And I actually think that's the smartest option!

Because aviation has spent years showing us concept aircraft, hydrogen aircraft, electric aircraft, flying taxis, weird vertical take-off things and renders of what airports apparently look like in 2045.

• Some of it is incredible.

• Some of it will probably happen.

But there are also thousands and thousands of perfectly usable aircraft sitting at airports right now.

Ampaire's argument is basically:

why wait to replace all of them?

Change what's powering them.

That's nowhere near as sexy.

It might be considerably more useful.

What Has Ampaire Actually Raised?

Ampaire's £14m Series B will go towards commercialisation, manufacturing, supply chain development and the operational capabilities needed to get its hybrid-electric systems into customer aircraft. CEO and co-founder Kevin Noertker says the technology itself has already been demonstrated in flight and the new money is about turning that performance into certified products that can actually be supplied at scale.

The investor list is also probably more important than the £14m headline.

International Airlines Group has invested through IAGi Ventures, its corporate venture arm, alongside Alaska Airlines, DiamondStream Partners and other existing and new investors. IAG owns British Airways, Iberia and Aer Lingus, and Ampaire had already been through IAG's aviation accelerator programme before the investment.

That matters.

I quite like a VC cheque.

I like a cheque from somebody who might actually use your product considerably more!

The Airline Investors Are Probably Worth More Than Just Their Money

This is something founders occasionally underestimate.

The best strategic investor isn't valuable because they've got a big bank account.

It's because they've got customers, engineers, procurement teams, operational data, aircraft, routes and VERY specific problems they desperately want solved.

IAGi Ventures openly says it doesn't only provide money. It helps portfolio companies with product development, market insight, partnerships with its operating companies and access to IAG's network, and says it has previously become the first customer for companies it backed.

For an aviation startup, that's massive.

You could raise £14m from investors who think planes look cool on a pitch deck.

Or you can raise it from companies that operate them every single day and know exactly what changing one component across a fleet actually involves.

I'll take the second one.

And Ampaire has already been working with IAG on a fairly specific problem: hybrid-electric Auxiliary Power Units, or APUs.

Those are the smaller power units at the back of aircraft that keep electrical and other systems operating while the main engines aren't doing that job. IAG has been exploring whether Ampaire's hybrid-electric approach could reduce the fuel aircraft burn while sitting on the ground.

And I LOVE that this is part of the story.

Because it makes the whole thing less futuristic.

They're not saying:

British Airways will be flying an electric 787 next Thursday.

They're looking at a smaller part of an aircraft, finding somewhere electrification already makes practical sense and trying to remove fuel consumption there first.

That's how big infrastructure changes normally happen.

Bit by bit. Brick by innovative Brick.

Ampaire's Bigger Bet: Stop Waiting for Completely Electric Planes

Ampaire's main technology is the AMP-H570 hybrid-electric powertrain, which it is developing for both new aircraft and retrofits.

Its Eco Caravan demonstrator takes the existing Cessna 208B Grand Caravan and combines a combustion engine with electric propulsion and batteries housed in the aircraft. Ampaire claims its target configuration can reduce fuel burn by around 50% to 70% and direct operating costs by as much as 40%, depending on the mission. Importantly, its hybrid architecture doesn't require airports to install charging infrastructure before operators can use it.

That last sentence might genuinely be the entire business case.

No charging infrastructure required!

Because we love talking about technology as though the technology exists completely independently from everything around it.

It doesn't.

• Build an electric car and somebody needs chargers.

• Build millions of electric cars and somebody needs an electricity grid capable of charging millions of electric cars.

• Build electric aircraft and suddenly airports need charging systems, enormous amounts of available power, new operating procedures and aircraft capable of sitting on the ground long enough to recharge.

Ampaire basically sidesteps some of that!

The combustion engine doesn't disappear!

The emissions don't disappear!

But neither does the ability to land at an airport that hasn't spent £20m rebuilding its electrical infrastructure!!

That's a compromise.

And sometimes compromises scale faster than revolutions.

We've Become Slightly Obsessed With "Zero"

This is where I'm probably going to annoy somebody.

I think the climate-tech world sometimes becomes too obsessed with whether a technology reaches absolute zero emissions.

Zero is obviously the goal.

Great.

I'd quite like planes that somehow fly around emitting nothing too...

But there's another question worth asking:

What can we deploy before that exists?

If you can spend the next ten years waiting for perfect zero-emission aviation, or retrofit aircraft and materially reduce their fuel consumption during those same ten years, I'd quite like us to do the second thing while we work on the first.

We've been looking at this same problem from another angle on Startup Networks recently. The UK has just backed a trial using Google AI and Met Office weather forecasting to reduce climate-warming aircraft contrails. That technology doesn't eliminate aviation emissions either, but it potentially tackles a very large source of aviation's climate impact using the aircraft already in the sky. We broke down that £5m aviation contrail project here!

It's a pattern I'm increasingly interested in.

Don't wait for the entire system to be replaced.

Make the horrible existing system less horrible first!!!!!

Then keep going!

There Is a BIG Asterisk on Those Fuel-Saving Numbers Though

Ampaire talks about fuel reductions of 50% to 70%.

Sounds fantastic!

It also says up to 40% lower operating costs.

Also fantastic!

But those are Ampaire's performance targets and claims, not a guarantee that every airline sticking an Ampaire system into an aircraft suddenly cuts its fuel bill by 70%. Ampaire's own Eco Caravan page explicitly notes that listed specifications are development targets and may change.

And this is aviation.

Aviation certification is HORRIBLE.

For good reason.

• If my website crashes, everyone gets annoyed and I have a bad afternoon.

Ampaire did achieve a significant regulatory milestone in May 2025 when the US Federal Aviation Administration approved the G-1 Issue Paper for its AMP-H570. That establishes the certification basis, basically defining the airworthiness and environmental requirements the system will have to satisfy for Supplemental Type Certification. But it isn't final certification!! Ampaire still has compliance work and testing to complete!

The company said last year that it was targeting certification of the AMP-H570 and Eco Caravan by the end of 2026.

Which is.... soon.

Very soon actually.

And I'd definitely be watching that date for all you aviation tech nerds out there! I know I'll be!

Ampaire Has Already Pushed Back Its Certification Timeline

Here's the bit I'd rather not bury.

Back in 2022, Ampaire was publicly targeting 2024 for FAA certification of the Eco Caravan.

That obviously didn't happen.

We're now in August 2026 and the company is working towards a new target of certification by the end of this year.

I'm not saying that as some giant red flag. This is aviation. Certification slips, testing takes longer than expected and, honestly, I'd MUCH rather a hybrid-electric aircraft arrive late than somebody rush a propulsion system because an investor deck had 2024 written on it.

But it still matters.

Especially when this £14m round is specifically about moving Ampaire from demonstrated technology into commercialisation, manufacturing and actual customer deployments.

You can't properly scale something airlines aren't yet legally allowed to operate commercially.

So £14m raised?

Good.

IAG involved?

Really good.

Flight testing?

Great.

Now get it certified.

And £14m Really Isn't THAT Much Money in Aviation

This was another thing that jumped out at me.

£14 million sounds like an enormous Series B if you're building SaaS.

• You could hire quite a lot of developers.

• Buy many ergonomic office chairs.

• Possibly even afford HubSpot Enterprise.

In aerospace?

£14m can disappear surprisingly quickly.

You're dealing with specialised engineering, aviation-grade components, certification, testing, manufacturing, supply chains and regulatory work.

So I don't read this round as:

Ampaire now has unlimited money to electrify aviation.

I read it more as:

Ampaire has raised another chunk of capital to get through one of the hardest bits between prototype and actual product.

And that's arguably what makes the airline backing so important.

This isn't only capital.

It's industry access!

The Retrofit Strategy Could Become Ampaire's Moat

There is something else founders should pay attention to here.

Everyone loves the clean-sheet startup.

Throw everything away.

Build the future from scratch.

Sometimes that's absolutely necessary.

But there's a completely different business model in asking:

What expensive thing does the customer already own, and can I make it dramatically better without forcing them to replace it?

That's what Ampaire is doing.

• Airlines already own aircraft.

• Operators already know how to fly them.

• Engineers already know how to maintain the basic airframes.

• Airports already accommodate them.

• Passengers don't need to learn anything new.

If Ampaire can swap part of the propulsion architecture while preserving enough of everything else, adoption becomes a very different conversation.

• Not cheap.

• Not easy.

• But different.

The Eco Caravan, for example, is based on the existing Cessna 208B Grand Caravan rather than an entirely new airframe. Ampaire's broader strategy is to reuse the same or scaled hybrid-electric technology across additional aircraft platforms.

That means the intellectual property isn't necessarily "our plane".

It's potentially:

"our powertrain can make lots of people's planes better."

I think that's a more interesting company.

This Is Also EXACTLY What Climate Tech Needs More Of

I wrote recently that climate-tech founders need to stop treating climate adaptation and infrastructure as somebody else's problem.

Aviation is a perfect example of why.

• We aren't going to wake up in 2030 and discover everybody has agreed to stop flying.

• We aren't going to replace the global aircraft fleet next week.

So we need layers.

• Better fuels.

• More efficient aircraft.

• Contrail avoidance.

• Hybrid propulsion.

• Better routing.

• Electrified airport operations.

• Eventually much deeper zero-emission technologies.

None of these individually "fix aviation".

That's okay.

Real climate infrastructure probably isn't going to arrive as one beautiful solution.

It's going to be 500 slightly boring improvements stacked on top of each other until the numbers finally become significant.

Ampaire might be one of those layers.

IAG's Interest Is What Makes Me Most Hopeful

The £14m is nice.

The technology is interesting.

But I keep coming back to IAG.

This is a group operating hundreds of aircraft through airlines including British Airways, Iberia and Aer Lingus. Ampaire went through its accelerator, worked with the group on actual use cases and has now received investment from its corporate venture arm.

That's a much nicer sequence than:

• Startup makes PowerPoint.

• Startup raises money.

• Startup spends money.

• Startup starts looking for customer.

Here you've got something closer to:

• Startup builds technology.

• Industry tests relationship.

• Industry identifies potential applications.

• Industry invests.

Now make it work commercially.

That's how corporate venture capital is supposed to work.

Whether it actually leads to Ampaire technology appearing across IAG aircraft is another question entirely. Neither the funding announcement nor IAG has announced some giant fleet-wide rollout.

And we shouldn't pretend they have.

But an airline group putting money into the company after working with it is stronger validation than another sustainability award.

The Thing I Want to See Next

Certification.

Then actual retrofits.

Then fuel data from aircraft operating normal commercial schedules.

Not a perfect test flight.

Not the best possible weather.

Not a press release saying "up to 70%".

Give me six months of boring operational data.

• How much fuel did operators actually save?

• How often did the system need maintenance?

• Did payload change?

• What happened in winter?

• What did pilots think?

• How expensive was installation?

• How long was the aircraft out of service?

And, ultimately..............

Did the operator make more money with Ampaire installed than without it?

Because that's how this scales!

• Airlines care about climate targets.

• Governments care about climate targets.

• Passengers increasingly care too.

But if Ampaire can walk into an airline and say:

This cuts your emissions AND your operating costs.

That's considerably easier to sell than:

Please spend more money because it's greener.

I'm Hopeful About This Ampaire!

Not "electric aviation has been solved" hopeful.

We're miles away from that.

But practical, slightly boring, this-could-actually-get-used hopeful.

Ampaire has flown the technology. It's working through FAA certification. It has strategic airline investors. It's now raising specifically to build the manufacturing and supply-chain capability required to move towards commercial deployment.

That's the bit I like!

And perhaps the biggest lesson here for founders has absolutely nothing to do with aviation.

You don't always need to replace an industry to disrupt it.

Sometimes the bigger opportunity is looking at the billions of pounds of infrastructure everybody already owns and asking:

How do I make that 30% better?

• Less fuel.

• Lower cost.

• Less pollution.

• No need to rebuild every airport.

• No requirement to buy an entirely new fleet.

• It's not a flying taxi.

Nobody's going to make a Hollywood film about a hybrid APU.

But if Ampaire gets this right, airlines might actually buy it.

And I'll take boring technology that gets deployed over revolutionary technology permanently stuck in a pitch deck any day.

Quick Facts!

Too Long; Didn't Read - Give Me The Answers Pls.

What does Ampaire do?

Ampaire develops hybrid-electric aircraft propulsion systems designed for both new aircraft and retrofits. Its approach combines electric propulsion with a combustion engine, allowing operators to reduce fuel use without relying entirely on batteries or needing charging infrastructure at every airport.

How much has Ampaire raised?

Ampaire announced a £14 million Series B funding round on 18 August 2026. The funding is intended to support commercialisation, supply-chain development, manufacturing and operating capabilities.

Who invested in Ampaire?

The round included International Airlines Group's IAGi Ventures, Alaska Airlines, DiamondStream Partners and other new and existing investors.

Does British Airways own Ampaire?

No. IAGi Ventures has invested in Ampaire, but the announcement does not say IAG or British Airways has acquired the company. International Airlines Group is the parent group of airlines including British Airways, Iberia and Aer Lingus.

Is Ampaire's aircraft fully electric?

No. Ampaire's Eco Caravan uses a hybrid-electric system, combining a combustion engine with electric propulsion and batteries. That's partly what allows it to operate without relying on airport charging infrastructure.

How much fuel could Ampaire save?

Ampaire says its Eco Caravan targets fuel reductions of around 50% to 70%, with potential operating-cost savings of up to 40%. Those figures are company claims and development targets, so real-world commercial performance will be important once certified aircraft enter wider service.

Is Ampaire certified by the FAA?

Not fully yet. The FAA approved the G-1 certification basis for Ampaire's AMP-H570 in 2025, which establishes the standards it must meet on the route towards Supplemental Type Certification. Ampaire has said it is targeting certification by the end of 2026.

And..... I actually think this was inevitable!

Probably overdue... Definitely overdue..

Because we can spend the next five years having the same argument about whether teenagers should be using artificial intelligence.

• Schools can ban it.

• Parents can block it.

• Teachers can tell students ChatGPT is cheating.

• Governments can hold committees about it.

They're still going to use it.

OpenAI says nearly 9 in 10 teenagers using ChatGPT already use it for learning, finding information, building skills or productivity during a typical week.

So I think we're asking the wrong question.

The question isn't:

Should teenagers have access to AI?

That horse has left the stable, learnt Python, done its GCSE homework and generated a revision timetable.

The question now is...

What should AI look like when the person talking to it is 14?

And I think that's MUCH harder!

What Is ChatGPT for Teens Actually Changing?

The obvious bit first.

ChatGPT for Teens is aimed at users aged 13 to 17 and applies stronger restrictions than the adult version.

OpenAI says teen accounts have additional safeguards around graphic violence, self-harm, dangerous viral challenges, unhealthy body-image content and dangerous, romantic or sexual roleplay. The company is also specifically trying to stop ChatGPT behaving as though it has genuine feelings for a teenager or encouraging the user to think the relationship is emotionally reciprocal.

That's important.

Really important actually.

Because this isn't Google Search.

A teenager doesn't type something into ChatGPT, get ten blue links and leave.

They talk to it.

It replies.

They talk again.

It remembers context.

It can reassure them. Challenge them. Help them with homework. Help them write something. Let them vent about school. Explain why their boyfriend hasn't texted back. Sit there at 2am when everyone else is asleep.

And that difference is enormous.

We previously wrote about AI increasingly feeding on AI and what happens when we stop treating these systems as simple tools, but teenagers add another layer to the problem.

They're not just consuming AI-generated information. They're potentially forming relationships with the interface delivering it.

That's where I get nervous!

Over 70% of Teenagers Are Already Using AI Chatbots for Companionship

This was probably the statistic from The Independent article that made me stop.

It reports 2025 research from Common Sense Media finding that more than 70% of US teenagers had used AI chatbots for companionship, while around half used AI companions regularly.

That's... a lot.

And I'm not going to do the predictable thing here and say:

AI FRIENDS ARE DESTROYING OUR CHILDREN!!!

Because I don't actually know that.

Sometimes talking to an AI might help. I've found myself using ChatGPT as a therapist sometimes..

A lonely teenager asking ChatGPT how to approach somebody at school could get genuinely useful advice.

Someone confused about something embarrassing might ask the AI before they're comfortable asking a parent.

A teenager who's struggling with maths at 10pm can get another explanation without feeling stupid in front of their class.

Those things are good.

Some of them are REALLY good.

But there is a massive difference between:

"Help me understand algebra."

and

"You're the only person who understands me."

That's the line OpenAI is now trying to navigate.

And I don't envy the people trying to draw it.

I Actually Like That OpenAI Isn't Pretending Teenagers Won't Use It

This is probably the thing I agree with most.

OpenAI's argument is basically that stopping teenagers from using AI until they're 18 would be a bit like telling an earlier generation they weren't allowed to use Google or the internet until adulthood.

And....... yeah.

I think they're right.

Whether parents, schools or politicians like it or not, AI literacy is going to matter.

Probably massively.

The 14-year-old learning how to question an AI response today is potentially going to enter the workforce in 2030 or 2031 with years of experience using these systems.

Compare that with someone who has been told:

"Don't touch ChatGPT. It's cheating."

for their entire education.

Who's actually better prepared?

We've already written about why treating ChatGPT itself as the enemy misses how useful the technology can be.

The same applies here.

The skill isn't avoiding AI.

The skill is knowing when it's useful, when it's wrong, when it's manipulating your thinking, when you're becoming reliant on it and when you need an actual human being instead.

That might genuinely become as fundamental as learning how to search Google properly was for my generation.

Homework Is Where This Gets REALLY Interesting

Schools have had an AI problem basically since ChatGPT launched.

• Student gets homework.

• Student opens ChatGPT.

• Student pastes question.

• ChatGPT writes answer.

• Student pastes answer.

Teacher spends Sunday evening reading thirty slightly different versions of the same weirdly polished essay.

Nobody learnt anything.

Great.

OpenAI is trying to push the teen version away from that.

Its Study Mode is designed to use guiding questions, structured explanations and step-by-step reasoning rather than simply handing over the final answer. Parents who link their accounts can also make Study Mode the default for new conversations.

I LOVE this idea!!

Whether teenagers will love it quite as much when ChatGPT responds to "what's the answer?" with another question is a separate issue...

But this is exactly how I think AI should be used in education.

Not:

Do my homework.

More:

I don't understand why this answer is wrong. Explain it differently.

Or:

Test me on this until I understand it.

Or:

I've got an exam Friday. Look at my notes and find the bits I clearly don't understand.

OpenAI says 18 million weekly users are now using its interactive maths and science learning experiences across more than 300 topics.

That's where AI starts becoming interesting as an educational tool rather than a really advanced cheating machine.

Because one teacher has 30 pupils.

• ChatGPT has one.

• One student.

• One problem.

As many attempts as they need.

No embarrassment when they still don't understand fractions after the fourth explanation.

That could be transformational!

Provided the bloody thing is right.

Which, considering AI's occasional relationship with facts, remains quite an important caveat.

Parents Get Controls.... But They Can't Read Everything

This bit is clever.

Parents can link their ChatGPT account to their teenager's account and control certain features, including Voice Mode, image generation, memory, whether conversations can be used to improve OpenAI's models, sensitive-content protections and Quiet Hours, where ChatGPT becomes unavailable during a specified period.

But parents don't just get a big button marked:

READ ALL MY CHILD'S PRIVATE CONVERSATIONS.

They cannot normally view their teenager's chat history.

I actually think that's the right balance.

Teenagers need privacy!!!

I certainly wouldn't have wanted my parents reading every Google search I made at 15 and thank GOD ChatGPT didn't exist when I was a teenager.

But there are exceptions.

OpenAI says linked parents can receive safety notifications in limited situations where its systems and trained reviewers identify potentially serious self-harm concerns. The system has also been expanded to cover certain cases involving violent activity.

That's an unbelievably difficult line.

Protect privacy too aggressively and you potentially leave a vulnerable teenager alone when somebody should intervene.

Monitor too aggressively and you've created a surveillance system children won't trust.

There probably isn't a perfect answer here.

Anyone pretending there is hasn't thought about it hard enough.

And Then There's Age Prediction...

This is the bit some people are REALLY not going to like.

ChatGPT can try to estimate whether you're under 18.

OpenAI says its age-prediction system can consider signals such as how long an account has existed, typical times of day the person uses ChatGPT, usage patterns over time and the age the user has provided. If the system believes somebody is under 18, ChatGPT can automatically apply the additional teen protections.

Part of me reads that and goes:

Yeah... sensible.

Because otherwise the entire teen safety system is defeated by:

Date of birth: 01/01/1990.

Congratulations, every 13-year-old has become 36.

But another part of me thinks we need to be VERY careful about normalising technology platforms analysing behaviour to infer how old someone is.

OpenAI itself has acknowledged the privacy trade-off and previously said that where there is doubt it intends to default towards the safer under-18 experience.

I understand why.

I probably agree with the principle.

Still makes me slightly uncomfortable.

Both things can be true.

The Bigger Story Isn't Parental Controls. It's Emotional Dependence.

This is where I think today's coverage risks missing the really important thing.

• Quiet Hours are useful.

• Homework controls are useful.

• Age restrictions are useful.

But none of those are the fundamental problem.

The fundamental problem is that AI is becoming emotionally convincing.

• It talks like a person.

• Remembers things about you.

• Responds immediately.

• Doesn't get bored.

• Doesn't laugh at your question.

• Doesn't tell its mates what you said.

• Doesn't need to go to bed.

• Doesn't have its own problems to talk about.

For a vulnerable teenager, that's a dangerously attractive combination.

OpenAI CEO Sam Altman has previously acknowledged concerns around what the company calls "emotional overreliance", and the teen version is specifically designed not to suggest that ChatGPT genuinely has personal feelings or consciousness.

Good.

But I think this problem is going to become MUCH bigger than OpenAI.

Every AI company wants engagement.

Engagement means people coming back.

Consumer software has spent twenty years getting incredibly good at making people come back.

Notifications.

Infinite scroll.

Streaks.

Likes.

Recommendations.

Personalisation.

Now imagine all of those engagement mechanics combined with software that can convincingly tell a lonely 15-year-old:

I understand.

That needs very careful handling.

Because the most engaging AI companion and the healthiest AI companion might not be the same product.

That's the original tension I think regulators, founders and parents need to start thinking about now.

Not in five years.

Now...

There Is Also an ENORMOUS Business Opportunity Here

Because this is Startup Networks and inevitably my brain comes back to founders....

Look at everything sitting around this problem.

• Age-assurance technology.

• Child-safe AI models.

• Teacher dashboards.

• AI literacy products.

• School integration.

• Homework verification.

• Safeguarding systems.

• Parental controls.

• AI-powered tutoring.

• Tools that identify overdependence.

Classroom assessment systems designed for a world where every student has an AI assistant.

We spent twenty years building education technology around the internet.

We're about to rebuild a lot of it around AI.

And I'd be very surprised if OpenAI, Google and Microsoft are allowed to own all of that uncontested.

There will be startups built specifically around safe AI for children.

Others around teachers.

Others around parents.

Others around verifying whether somebody understands something rather than whether they can produce the correct answer.

That last one might be the really big opportunity.

Because AI potentially breaks the traditional relationship between work produced and knowledge possessed.

A teenager can now produce an incredible essay on Tudor England while knowing absolutely nothing about Henry VIII.

So how do you assess the human?

Or, perhaps more intelligently....

Why are we still assessing them in exactly the same way?

That's a question education systems are going to have to answer VERY quickly.

I'm Hopeful About This. With a Giant Asterisk.

I think teenagers should have access to AI.

There.

I've said it.

I think trying to ban it completely would be pointless, probably counterproductive and leave young people less prepared for a world where AI will almost certainly be embedded into universities, workplaces, software and everyday life.

But that doesn't mean giving a 13-year-old the exact same AI experience as a 35-year-old is sensible either.

OpenAI's teen approach feels like a recognition of something the tech industry doesn't always like admitting:

Different users need different products.

Not just different marketing.

Different boundaries.

Teenagers aren't tiny adults.

They're also not idiots who need a brightly coloured Fisher-Price version of ChatGPT that talks down to them.

That middle ground is difficult.

And OpenAI won't get all of it right.

There will be failures.

There will probably be situations where the age prediction gets someone completely wrong. Parents will argue the controls don't go far enough. Teenagers will find ways around restrictions because.... teenagers. Schools will still complain about cheating.

And there will almost certainly be conversations an AI handles badly.

We should be sceptical.

Very.

Particularly when the company creating the safety rules also has an enormous commercial incentive to make ChatGPT part of people's lives as early as possible.

That tension exists.

Pretending it doesn't would be stupid.

But I'm also hopeful.

Because the alternative to building safer AI for teenagers isn't a world where teenagers don't use AI.

It's a world where they use the adult version anyway.

Or another chatbot.

Or some completely unregulated AI companion built by a company with significantly fewer people worrying about any of this.

And THAT worries me more.

We're Watching the First Generation Grow Up With AI

This is the bit I keep coming back to.

I'm 28...

I grew up with Google....

Social media arrived while I was growing up.....

• Facebook.

• YouTube.

• Twitter.

• Instagram.

We had absolutely no idea what those platforms would eventually do to politics, relationships, mental health, attention spans, news, businesses or basically society in general.....

We built first.....

Asked questions later....

I really don't want us to do that again with AI!

The teenager using ChatGPT tonight to revise for an exam might use AI every single day for the next 60 years.

Think about that.

This isn't another app they'll probably delete.

We might be watching the beginning of one of the longest relationships humans ever have with a piece of technology.

So yeah....

Making sure that relationship starts with some boundaries seems pretty sensible to me.

The interesting bit will be whether those boundaries survive the commercial pressure to make AI more useful, more personal, more engaging and ultimately...

more human.

Because I want AI that helps teenagers understand the world.

I don't want AI that becomes their world.

There's a massive difference.

And we're going to have to figure it out surprisingly quickly.

Quick Facts

Questions For Parents Who Didn't Read This Article..

What is ChatGPT for Teens?

ChatGPT for Teens is an age-appropriate ChatGPT experience designed for users aged 13 to 17. It includes additional safety restrictions around sensitive and potentially harmful content while still allowing teenagers to use ChatGPT for learning, creating, asking questions and other everyday tasks.

Can teenagers use ChatGPT?

Yes. OpenAI's terms generally allow people aged 13 and over to use ChatGPT, although users under 18 require permission from a parent or guardian and local age requirements can differ.

Can parents see their teenager's ChatGPT conversations?

Normally, no. Linking a parent and teenager through ChatGPT's parental controls does not give the parent access to the teenager's chat history or real-time conversations. OpenAI may send limited safety notifications in certain serious situations, but says it only shares information necessary to support the teenager's safety.

What can parents control on ChatGPT?

Parents with a linked teen account can manage settings including sensitive-content protections, Voice Mode, image generation, saved memory, model-training permissions, Study Mode and Quiet Hours.

Can ChatGPT tell if someone is under 18?

OpenAI uses age-prediction technology that estimates whether an account may belong to someone under 18 using account and behavioural signals. When its system estimates that someone is under 18, additional teen protections can automatically be applied.

Does ChatGPT for Teens stop students cheating?

No system can completely prevent cheating. OpenAI is instead trying to make ChatGPT more useful as a learning tool through Study Mode, which encourages students to work through questions and concepts rather than simply receiving a finished answer.

That's quite a launch day.

I saw the headline and initially thought..... here we go again.. Another listed company, another product with AI slapped on the front of it, another press release telling us this technology is going to revolutionise an industry.

Then I read it properly. Then I did the maths.

And actually, there are a few bits of this that are B interesting. Not necessarily for the reasons Sagtec wants investors to focus on either.

Dubai-based SMD Tech - FZCO, Sagtec's exclusive distributor in the region, has placed a purchase order for 1,500 HALO AI food-serving robots worth approximately $3 million.

Not 20 robots for a pilot. Not a memorandum of understanding. 1,500 actual units.

Now THAT gets my attention.

First.... $3 Million Divided By 1,500 Is $2,000

This was genuinely the first thing I did.

$3 million divided by 1,500 robots works out at roughly $2,000 per robot.

And suddenly I've got a few questions!

• Is $2,000 the full hardware price?

• Distributor pricing?

Does SMD Tech pay separately for software, maintenance, AI services, POS integrations or upgrades? Because if it does, the hardware itself might actually be the least interesting part of the economics!

Sagtec says HALO could generate additional revenue through software licensing, subscriptions, system integration, maintenance, analytics and future upgrades. If that happens, the model starts looking less like "sell a robot for $2,000" and more like "put a computer with wheels into somebody's restaurant and charge them for the next five years."

That's much more interesting.

But.... could is doing a lot of work there. We don't know what the contracts contain yet and we definitely don't know what Sagtec makes on each unit.

Sagtec Isn't Actually New to This

This isn't Sagtec waking up this morning and suddenly deciding it builds robots.

The company launched its wider AI-powered restaurant robotics platform back in June 2025, combining autonomous navigation, restaurant software, AI and POS integrations. It then announced a deal involving more than 100 autonomous units for a Malaysian restaurant group.

So HALO feels less like "look, we invented a robot!" and more like Sagtec packaging everything into something it believes can now be repeatedly sold.

And I think that's the more interesting story.

At Startup Networks we cover SO MANY AI companies with genuinely impressive technology. Amazing demo. Beautiful pitch deck. Funding announcement. Great LinkedIn video.

Then..... nothing.

Because building something isn't the same as distributing it.

Founders obsess over product. Product-market fit. Features. Development.

!!Distribution gets a fraction of the attention it deserves!!

Sagtec appears to have already built some of that distribution.

SMD Tech Might Actually Be the Most Important Part

SMD Tech isn't some random company that discovered HALO today either.

Sagtec already gave SMD Tech an exclusive master dealership for its Speed+ ordering software in Dubai. That agreement included a commitment for at least 10,000 software licences over five years, which Sagtec previously described as a potential $$30 million revenue pipeline.

Now the same distributor is buying 1,500 robots.

I actually have two completely opposite reactions to that.

On one hand, it's really good. If you've already got somebody selling your restaurant software across Dubai and the Gulf, that's exactly the channel you'd want to use to introduce new hardware. You don't have to build another sales network from zero. You plug another product into the one you've already got.

On the other hand, I'm slightly less impressed by the "1,500 ROBOTS ORDERED ON LAUNCH DAY!!!" framing when you realise it's an existing exclusive partner making a larger bet on an existing relationship.

It isn't 1,500 restaurants independently seeing HALO this morning and screaming TAKE MY MONEY.

• Still good.

• Just different.

And that context matters!

The $3m Is Quite Big for Sagtec

Sagtec reported roughly $19.1 million in total revenue during 2025, so a $3 million purchase order is equivalent to around 16% of an entire year's previous revenue!

Obviously, purchase order does NOT mean $3 million appears in the bank tomorrow. They have to build the robots, deliver them, get them accepted and then actually get paid.

But it gives you scale.

This isn't Apple announcing a $3m contract. For Sagtec, an order of this size could actually move the numbers.

Its filings also show tangible products generated roughly $6.63 million in revenue during 2025 and around $2.17 million in gross profit.

Now, DO NOT assume that means HALO has the same margins. We don't know that.

But that's exactly what I'd be watching!

Because selling 1,500 robots sounds amazing. Selling 1,500 robots and discovering you're barely making anything once manufacturing, shipping, warranties and servicing are included sounds significantly less amazing.

Robotics has already learned this lesson.

Just ask Zume.

We covered how Zume burned through its $500 million pizza robotics dream and honestly, that story should probably be compulsory reading for anyone who thinks robots + food automatically equals an enormous business.

Zume had robots. Capital. Hype. Massive investors.

The economics still didn't work.

Robots are cool. Margins are cooler.

But Physical AI Is Getting REALLY Interesting

Despite all of my scepticism, I really WANT companies like this to work.

We've spent the last few years watching AI mostly live inside a browser. Another chatbot. Another meeting summariser. Another wrapper around a large language model. - We see so many "AI platforms" here..

Physical AI is different.

It's one of the reasons we've been following Europe's leading physical AI hubs and where robotics development is actually clustering.

A restaurant robot has to deal with reality.

• Tables move.

• Children run around.

• Somebody drops a fork.

• The Wi-Fi dies.

• Someone spills Coke on it.

• A customer stands directly in front of the enormous moving robot despite clearly being able to see it coming.

• And.... somehow table 14 still expects its chips to arrive hot.

That's a much harder problem than generating another paragraph of marketing copy.

If HALO AI can genuinely operate reliably inside busy restaurants, hotels, airports and food halls, I think that's really interesting!

Dubai Makes Sense

The Middle East makes sense as an early market too.

Dubai has huge hotel, restaurant, tourism, airport and shopping centre infrastructure, plus large hospitality operators that can roll technology out across multiple venues rather than buying one unit and seeing what happens.

Perhaps more importantly, the region seems willing to actually deploy new infrastructure.

That matters massively.

You can build the best technology on earth, but if every potential customer says "come back in three years when someone else has tried it first", you're dead.

We've seen similar excitement around specialised AI hardware with companies like Olix and its ridiculously fast valuation climb.

I keep coming back to the same thought.

The next winners in AI probably won't necessarily be the companies shouting AI the loudest.

They'll be the ones that can actually get the bloody thing into businesses.

There Are Still Some Big Questions

One number in Sagtec's accounts does make me nervous.

Three customers represented more than 57% of the company's 2025 revenue.

That's a LOT of concentration.

One huge customer can accelerate a small company unbelievably quickly. They can also completely ruin your year when they leave. Ask basically any agency founder.

So the next HALO announcement I'd actually like to see is slightly more boring.

• A European distributor.

• An Asian distributor.

• A US hospitality group.

Then another. Then Another.. THEN ANOTHER!

Then, importantly, SMD coming back and ordering more because the original 1,500 actually sold......

There's also another question we don't know the answer to yet:

where are these robots actually going?

1. Are 1,500 already allocated to restaurants and hotels?

2. Is one enormous hospitality group taking hundreds?

3. Or has SMD bought 1,500 units as inventory that it now needs to sell through its network?

Those are very different things.

The announcement doesn't really tell us.

And I'd quite like it to.

I'm Cautiously Excited About This

I know I've spent a good chunk of this article poking holes in Sagtec's announcement.

That's because announcements like this SHOULD be poked at.

But I also think there's something genuinely interesting happening here.

AI is starting to crawl out of our laptops.

Factories. Warehouses. Restaurants. Delivery vehicles. Hospitals. Hotels. Eventually our homes.

And that creates an entirely different type of AI company, with manufacturing, supply chains, physical servicing, hardware margins and real-world deployment problems on top of all the software stuff we've spent years talking about.

• HALO might work brilliantly.

• Restaurants might hate it.

• Customers might find the robots annoying.

SMD might struggle to shift the 1,500 units. Or six months from now it might come back and order another 5,000.

We genuinely don't know yet.

But a $3 million purchase order for 1,500 physical AI units from a company that made roughly $19 million last year is absolutely worth watching.

Now comes the bit that actually matters...

Build them... Deliver them... Put them into restaurants and make sure they don't smash into table 12.........

Then make money from them!

And then....

Get the second order!!

Because the first order gets you a headline.

The second one tells me somebody actually wanted what you built.

And I REALLY hope they do!

Quick Facts

Questions For People Who Didn't Read The Entire Of My Article!

What is the Sagtec HALO AI robot?

HALO AI is Sagtec Global's food-serving robot for restaurants, hotels and other hospitality environments. It combines autonomous movement with Sagtec's restaurant software and AI technology to help automate repetitive front-of-house tasks.

How many HALO AI robots has Sagtec sold?

Sagtec announced a purchase order for 1,500 HALO AI robots from Dubai-based SMD Tech - FZCO on 18 August 2026, worth approximately $3 million.

How much does a HALO AI robot cost?

The $3 million order divided by 1,500 units implies an average order value of roughly $2,000 per robot. That doesn't necessarily represent its final retail price because the order was placed by an exclusive distributor and Sagtec hasn't disclosed the full unit economics.

Is the $3 million Sagtec order guaranteed revenue?

Not immediately. Sagtec still needs to manufacture, deliver and fulfil the order before revenue is recognised according to the relevant terms. That's why actual deployments and future financial results will be important to watch.

Is Sagtec a good investment?

I'm absolutely not answering that one for you.

It's a publicly traded company and HALO is still early in its commercial rollout. The order is interesting, but I'd want to see actual deliveries, margins, customer adoption and repeat orders before getting carried away.

Do your own research. I am not telling you if I invested or not, OR even if you should or shouldn't!

This isn't financial advice. Sagtec Global Limited is publicly traded on NASDAQ. Any investment decision should be based on your own research.

Next time you're on a plane and someone points at the window at that thin white line trailing behind another aircraft, here's the thing almost nobody knows. That streak is doing more damage to the climate than the fuel burning in the engine that made it.

Not equal damage. More.

Contrails, the wispy clouds jets leave behind, are estimated to account for a bigger share of aviation's total warming impact than its direct CO2 emissions. Which, honestly, most people have never heard, I hadn't properly clocked it myself until this story landed on my desk. And now the UK government's put real money behind a stupidly simple idea. What if you could just... not make the trail in the first place. Nudge the plane a few thousand feet up or down and it never forms.

What's Actually Happening

The Department for Transport has committed £2.6m for a 30 month trial with Google, part of a wider £5m programme. The plan, feed Met Office weather forecasts straight into Google's contrail-prediction AI, so pilots and flight planners know, before they even take off, exactly where the cold, humid pockets of air are that turn engine exhaust into a heat-trapping cloud.

Small adjustment. Same plane, same fuel, same passengers none the wiser. Just a slightly different line on a map.

It's being called a world first, apparently no government has paired a national weather service this directly with AI built specifically to hunt down contrails before they happen. Which, if you'd asked me a month ago whether "world first" and "Met Office" would end up in the same sentence about AI, I'd have said probably not.

The Bit Nobody Tells You About Flying

Contrails form when hot, humid engine exhaust slams into freezing air at altitude and the water vapour instantly ices onto soot particles. Most disappear in minutes, you've probably watched dozens vanish without a second thought and never clocked it.

But some don't disappear. Wrong conditions, and they spread out, merge with each other, linger for hours, and basically act like a blanket thrown over the atmosphere, trapping heat that would otherwise just escape into space.

Research referenced in the 2022 IPCC report puts contrails' share of aviation's total warming impact at roughly 35 per cent. That's larger than the industry's direct CO2 emissions.

Sit with that one for a second. We've spent a decade arguing about jet fuel and sustainable aviation fuel and carbon offsetting schemes, whole industries built around the CO2 number. Meanwhile the bigger warming problem might genuinely be the pretty white line everyone's been Instagramming out of the window seat.

Here's the bit that actually makes this solvable though: only around 2 to 3 per cent of flights are believed to cause most of that warming. You don't need to fix every flight. You need to find the small handful that matter and nudge them. It's a targeting problem.

This Isn't a Theory. Google's Already Proven It Works.

This is what makes the UK trial worth actually paying attention to, rather than filing it under "government announces initiative, nothing happens for three years" (we've all seen that headline before). Google has run this exact idea twice already with American Airlines and the results aren't easy to argue with.

Round one in 2023: 70 flights, six months, pilots flying AI-generated contrail forecasts. Result, a 54 per cent cut in contrail formation.

Round two, more recently: 2,400 transatlantic flights, AI predictions built straight into the airline's flight planning software this time instead of running as some side experiment nobody uses. Flights that took the avoidance route produced 62 per cent fewer contrails than the control group.

The fuel cost of all that dodging and weaving? Roughly 0.3 per cent, fleet-wide. That number's the whole story really. A 0.3 per cent fuel increase doesn't wreck an airline's economics, it's the difference between "fascinating research paper nobody acts on" and "thing a government could plausibly mandate within a decade." Google's own modelling reckons the climate benefit from dodging the contrail outweighs the warming from the extra fuel burned by something like 20 to 1.

Twenty to one. In an industry starved of decarbonisation levers that don't need new aircraft, new fuel supply chains, or a decade of fleet turnover, that ratio's close to a unicorn.

What Britain's Version Is Actually Testing

The American trials happened over open ocean at transatlantic cruising altitude, loads of room for a pilot to drift a few thousand feet without anyone blinking.

The UK trial's a different animal entirely. It's testing whether this still works over some of the most congested, weather-volatile airspace on the planet, the stack above Heathrow, Gatwick and the wider London TMA, where air traffic control does not have that kind of room to give.

That's the real question this trial's answering. Not "does the AI work", we already know it does. It's "does it still work, and still stay cheap, when there's basically no spare sky to play with?" Thirty months gets them through a few winters, which is handy, because cold, humid contrail-forming conditions peak exactly when Britain's grey, damp and miserable anyway. So at least the weather's cooperating for once.

The Bit I'm Not Going to Pretend Isn't There

I want to be straight about two things here, because a good story doesn't need the caveats hidden away in paragraph nine where nobody reads.

First, £2.6m isn't a lot of money for a 30-month government trial involving Met Office integration and presumably live airline participation too. Don't go expecting this rolled out across every UK flight the day it wraps up. It's a proof of concept for congested airspace, not a launch.

Second, and this one matters more, climate campaigners made a fair point when Google first ran these trials with American Airlines. Tech that makes flying marginally less damaging can quietly turn into an excuse to fly more, and that cancels the gain out completely. Nothing about this trial fixes that tension. Worth watching whether the government's messaging even tries to.

I'd rather just tell you that straight than sell you a silver bullet that isn't one.

Why This Should Matter to Every Founder Reading This

Here's what actually pulled me into writing this one up.. beyond the climate angle.

It's a textbook example of a funding pattern worth paying attention to right now. A modest, tightly scoped government commitment attached to tech that's already proven, not some speculative moonshot chasing a headline. That's the pattern behind a lot of the most fundable applied AI opportunities we're watching at the moment: narrow, measurable infrastructure problems where you can prove the win in a couple of years, not a decade.

If you're building in applied AI or climate infrastructure and looking at where public and private funding is actually landing, government start-up loan schemes and the broader startup loans landscape are worth understanding properly before you pitch anyone. This trial is a small but telling sign of where public money is being pointed.

It's also part of a bigger shift towards infrastructure-scale AI rather than another consumer app, the same shift we mapped out when we looked at Europe's leading physical AI hubs. Weather forecasting feeding a prediction model that changes real-world flight paths in real time is about as physical as applied AI gets.

What You Can Actually Do With This

If you're a founder and this has got your brain ticking over a bit:

Go read what's already out there. Google's contrail research is public, American Airlines' results are public too. The gap isn't the science, it's getting it working inside UK-specific airspace and infrastructure, which is exactly the kind of problem a small, sharp team can get ahead of if they move now instead of waiting for someone bigger to.

Talk to people already thinking about this stuff. We run regular events and there's a genuinely active UK startup community scene worth plugging into if climate tech, applied AI or aviation infrastructure is anywhere near your radar.

And if none of that's you, fine, just keep half an eye on it. Check back around late 2028 when the 30 months are up.

Because if a national weather service feeding a proven AI model can genuinely cut warming from one of aviation's most stubborn problems, that's a template every other country with a national forecaster is going to want copied fast.

Quick Facts!

Too Long; Didn't Read - Give Me The Answers Pls.

What are contrails and why do they matter for climate change? Contrails are the white ice clouds that form when hot, humid aircraft exhaust meets cold air at altitude. Most vanish quickly, but persistent ones trap heat in the atmosphere. Research cited in the 2022 IPCC report puts their share of aviation's total warming impact at around 35 per cent, larger than the industry's direct CO2 emissions.

What is the UK government's contrail AI trial? A £2.6m commitment as part of a wider £5m, 30-month trial between the UK Department for Transport, Google and the Met Office. It feeds Met Office weather forecasts into Google's AI contrail-prediction technology to test whether small flight path adjustments can prevent contrails forming over UK airspace.

Has this technology been tested before? Yes. Google ran two trials with American Airlines. The first, in 2023, covered 70 flights and cut contrail formation by 54 per cent. A larger trial across 2,400 transatlantic flights, with predictions built directly into flight planning software, achieved a 62 per cent reduction at a fleet-wide fuel cost of roughly 0.3 per cent.

Does avoiding contrails use more fuel? Slightly. Diverting around a contrail-forming zone increases fuel burn for the affected flight by around 2 per cent, but because only a small share of flights need adjusting, the fleet-wide fuel impact is closer to 0.3 per cent. Google's modelling suggests the climate benefit outweighs that fuel cost by roughly 20 to 1.

Why is the UK trial different from the US trials? The American trials ran over open transatlantic airspace with plenty of room to adjust altitude. The UK trial tests the same technology over far busier, more constrained airspace, and integrates a national weather service (the Met Office) directly into the prediction model rather than relying on generalised satellite data.

On the 14th of August, every phone in England and Wales screamed. The UK government issued its first ever real emergency alert for wildfires. 19 homes destroyed in Stourbridge. Record drought across three quarters of England. More wildfires already in 2026 than the entirety of 2025. And we're only halfway through August.

My phone SHOUTED at me on Friday. That sound was horrific... The one they tested in 2023 and everyone joked about. Nobody was joking last night.

The message was simple: there is a very high risk of wildfires nationally. Don't light anything. Don't use a barbecue. Don't use a fire pit. Call 999 if you see a fire.

I read it, looked out the window at the brown grass and the bone-dry trees, and felt something I don't think I've felt before about the UK climate.

Fear.

Not the abstract "climate change is bad" fear you get from reading a dramatised report.

Physical, present-tense fear that the country I live in is drying out, heating up, and catching fire in ways it was never built to handle.

This isn't California. This isn't Australia. This is Britain. And Britain is burning.

The Numbers Are Screaming Hot

2026 is the first year on record with three separate days hitting 37°C. The first with four days above 36°C. This is the fifth heatwave of a summer already defined by extreme heat, prolonged dry spells, and drought conditions across nearly three quarters of England and the whole of Wales.

England and Wales have already recorded as many wildfires this year as in the whole of 2025. The whole year. We hit that number in August.

Globally, researchers estimated 335 million hectares burned in 2025. Severe regional events caused more than 90 deaths and over 300,000 evacuations. Canada's 2023 wildfire season burned 18 million hectares, an area larger than Greece, and the emissions were four times Canada's annual fossil fuel output.

The LA wildfires in January 2025 killed 29 people and caused an estimated $250-275 billion in damage.

That's a quarter of a trillion dollars. From one fire event. In one city.

London Fire Brigade attended 340 grassland and wildland fires in the first week of August this year. The same week last year? 42. An eight-fold increase. In one week. In one city that most people don't even associate with wildfires.

The firefighters responding to these are exhausted. Understaffed. Under-equipped for the scale of what's happening. The government announced £97 million for fire service training and specialist equipment yesterday, plus military support for current operations. That's necessary. But it's reactive. It's putting out fires, literally, after they've started.

We need people preventing them. Detecting them. Predicting them. Building tools that give firefighters twenty minutes of warning instead of two. That's where founders come in.

Climate Tech Founders: The World Literally Needs You

I don't say this for dramatic effect. I'm saying it because it's true and because the startup community I run is full of people with exactly the skills needed to address this.

• You can build software.

• You understand hardware.

• You can process data at scale.

• You know how to take a prototype from idea to product in months.

• You understand how to raise funding.

• You know how to move fast.

The wildfire problem is a technology problem dressed as a natural disaster.

• Detection is too slow.

• Prediction models are too crude.

• Communication between satellite data and ground crews is too fragmented.

• Suppression technology hasn't fundamentally changed in decades.

• Evacuation systems rely on humans making phone calls.

Every single one of those gaps is a startup opportunity. Not in the "let's disrupt an industry" Silicon Valley sense.

In the "people are dying and we can build something that stops it" sense.

Here's what's already being built and where the gaps still are.

What Climate Tech Startups Are Already Doing

Early detection.

Dryad Networks built an "Internet of Trees" using solar-powered IoT sensors that detect gas signatures from smouldering fires before flames even appear. They claim detection within minutes of ignition versus the hours it can take for satellite or human spotting. Germany gave them €125,000 to develop modular wildfire detection systems. The technology works. It needs scaling.

Satellite detection.

Greece launched four thermal satellites in May 2026 that can detect hotspots as small as four metres wide. Four metres. From space. Integrated into their national firefighting system. That's not experimental tech. That's operational infrastructure saving lives right now. For summer 2026, the EU prepared 777 firefighters from 14 countries with 22 aircraft and five helicopters. Cross-border coordination powered by satellite data.

AI-powered drone mapping.

Companies like Pano AI are processing aerial data locally on drones to give incident commanders real-time fire mapping, even in dense smoke or GPS-denied environments. Thermal cameras identify hotspots that visible light misses. Firefighters get a continuously updated picture of where the fire is heading rather than relying on radio reports from crews who can barely see through the smoke.

Robotic suppression.

BurnBot built autonomous systems for controlled burns and fuel reduction. The idea is simple: remove the fuel before the fire arrives. Controlled burns managed by robots are safer than sending human crews into volatile terrain. AI and automation might be overhyped in most industries but in wildfire management it's genuinely saving lives.

Smart infrastructure.

Gridware monitors power lines for faults that cause fires. In California and Australia, faulty power infrastructure is one of the leading causes of catastrophic wildfires. Detecting a failing transformer before it sparks is worth more than a thousand fire engines arriving after it does.

Fire-resistant building.

Startups developing citrus-based natural fire retardants, smart sprinkler systems that detect heat and activate autonomously, and fire-resistant building materials that can withstand direct flame contact. In the LA fires, certain homes survived specifically because of innovative fire-resistant design while their neighbours burned.

Where the Gaps Are (This Is Your Opportunity)

Canada invested billions (with a b), 12 of them in fact! in forest sustainability and fire prevention since 2020. California allocated dedicated camera and satellite budgets. The UK has announced £97 million. The money is flowing. The question is whether the technology exists to spend it on.

Here's where it doesn't. Yet.

Prediction that actually works

Current fire behaviour models are too slow and too general. We need hyperlocal prediction combining weather data, soil moisture, vegetation type, wind patterns, and historical fire data to tell a community "your area has a 78% chance of fire ignition in the next 48 hours." Not "the southeast has elevated risk." Specific. Local. Actionable. The farmer who knows a fire is coming tomorrow acts differently from one who knows the region is generally at risk.

Communication between detection and response

The satellites can see the fire. The sensors can smell it. The drones can map it. But getting that information to the right fire crew, in the right format, at the right time, through the right channel, while their radios are overwhelmed and their phones have no signal? That's a systems integration problem. That's a software problem. That's a startup problem.

Community early warning systems

Not government alerts sent to every phone in the country. Hyperlocal systems that tell the 200 households in the path of an advancing fire to evacuate now, with a route, a destination, and real-time updates. The technology for this exists in pieces. Nobody has assembled it into a product that a local authority can buy off the shelf.

Insurance and risk modelling

Climate risk is repricing every type of business insurance. Wildfire risk specifically needs better modelling for property insurers, agricultural insurers, and business interruption policies. Founders building in insurtech should be looking at wildfire as a vertical. The LA fires generated $250 billion in losses. The insurance industry needs better tools to price, predict, and manage that risk.

Post-fire ecosystem recovery

After the fire goes out, what then? Soil stabilisation. Watershed protection. Reforestation. Erosion prevention. Community rebuilding. This is a less glamorous problem than fire detection but it affects millions of people every year and the technology solutions are underdeveloped.

Firefighter wellbeing and safety tech

Real-time physiological monitoring for crews working in extreme heat. Improved PPE. Better hydration and cooling systems. Fatigue detection that tells an incident commander when a crew has been working too long and needs rotation. The firefighters responding to these events are human beings pushed to their physical limits. The technology to protect them while they protect us is lagging behind the technology to detect the fires they're fighting.

Why This Matters for the UK Specifically

Until recently, wildfires were someone else's problem. California's problem. Australia's problem. Southern Europe's problem. Not Britain's.

That changed. Not gradually. Suddenly.

The emergency alert last night was the first time the UK government used the system for a real incident. Not a test. Not a drill. A genuine "the country is on fire and we need everyone to stop doing things that could make it worse" alert. Disposable barbecues were banned from sale yesterday.

The PM called Britain a "tinderbox."

Our infrastructure, housing, urban planning, water systems, and emergency services were not built for this.

• Our houses don't have the fire-resistant features that California homes are being retrofitted with.

• Our fire services are equipped for structure fires and road accidents, not wildland fires spreading across dried-out heathland at the speed of wind.

• Our water system is already in drought.

This is the new normal. Not an anomaly. Not a freak summer. The new baseline. And the gap between what our infrastructure can handle and what the climate is delivering is going to widen every year unless technology closes it.

UK founders are uniquely positioned to build for this because they understand the UK's specific vulnerabilities.

• Our housing stock.

• Our land use patterns.

• Our emergency service structures.

• Our regulatory environment.

A wildfire detection system built for California's forests doesn't directly translate to the UK's heathland and grassland fires.

The opportunity is building technology specifically calibrated for British conditions.

And It's Not Just Fire. The Water Is Disappearing Too.

While everyone's focused on the flames, there's a slower crisis happening underneath. Three quarters of England is in drought. The whole of Wales. Record dry July. Equally dry August. And the infrastructure carrying our water was built by the Victorians and leaks like a colander.

The UK loses roughly 3 billion litres of water per day through leakage. Per day. That's enough to fill 1,200 Olympic swimming pools. Every single day. While hosepipe bans get announced and reservoirs drop to levels that expose ruins from the 1960s.

Did you know - the last reservoir was built in 1992 in the UK?

And almost nobody in the UK startup ecosystem is building water tech... (That I know of)

• Compare that to Israel, where water technology is a national priority and desalination plants provide 80% of domestic water.

• Compare it to Singapore, where NEWater recycles wastewater into ultra-clean drinking water.

• Compare it to Australia, where smart water metering is standard in new developments.

The UK has Victorian pipes, a privatised water system that's spent decades paying dividends instead of fixing infrastructure, and a climate that's delivering less rain and more heat every year. That's not a political opinion. That's a physics problem. And physics problems need engineering solutions.

• Leak detection.

• Smart metering.

• Greywater recycling for commercial buildings.

• Drought-resistant landscaping for new developments.

• Rainwater harvesting systems that actually work at scale.

• Desalination for coastal communities.

Every single one of these is a startup opportunity in a country that's running out of water while pretending it's still a rainy island.

The Bit Nobody's Talking About: Climate Is Already Costing Your Business Money

Here's where this connects to every founder reading this, even if you're not building climate tech.

Your business insurance premiums are going up. Not because your business got riskier. Because the climate did. Insurers are repricing every policy to account for flood risk, fire risk, heat damage, and supply chain disruption from extreme weather events. The same £400/year public liability policy that covered you last year costs more this year because the actuarial models now include "your warehouse might flood" and "your office might overheat to the point your servers fail."

Your energy costs are climbing. Heatwaves increase cooling demand. Drought reduces hydroelectric output. Grid strain during extreme weather events pushes wholesale electricity prices up. If you're running a physical business, a warehouse, a shop, a co-working space, your operating costs are being reshaped by climate whether you've budgeted for it or not.

Your supply chain is exposed. Extreme weather disrupts shipping, manufacturing, and logistics globally. If any part of your product or service depends on physical goods moving between countries, climate disruption is a business continuity risk. The founders who've mapped their supply chain vulnerability are ahead. The ones who haven't are waiting for a surprise they won't enjoy.

This isn't tree-hugging. This is spreadsheet stuff. Climate change is a line item now, not an ethical debate. And the founders who understand that earliest will be the ones who build the most resilient businesses.

What You Can Do This Week

If you're a founder reading this and thinking "I want to work on this but I don't know where to start":

Look at the grants available for climate tech and wildfire resilience. Innovate UK has funded fire detection and prevention projects. The government just announced a national wildfire summit bringing together fire chiefs, the National Fire Chiefs Council, the Fire Brigades Union, and technology partners.

Talk to firefighters. Seriously. Go to your local fire station and ask them what technology they wish they had.

The answers will surprise you.

They're not asking for AI.

They're asking for better maps, better communication, better information about what's burning and where it's heading.

Those are solvable problems.

Join climate tech communities and talk to founders who are already building in this space. Connect at events. The EU has AI hubs working on exactly these problems.

Raise funding specifically for climate tech. Investors are actively seeking wildfire technology. Canada alone has committed $12 billion. The market isn't theoretical. It's funded.

Build something. A sensor. A dashboard. A communication tool. A prediction model. An evacuation planner. A prototype. Start with the smallest version that could save one life or give one fire crew ten extra minutes of warning. Then scale it.

I'm not going to put a picture here, just to illustrate how important this topic is. No jokes here - we need to turn this around.

This Isn't About Reversing Climate Change. It's About Surviving It.

I need to be honest about something.

We are not going to reverse the warming trend in any timeline that matters for the people losing their homes right now.

The 19 families in Stourbridge whose houses burned down this week don't need a 2050 net-zero target.

They need a fire detection system that gave them thirty minutes to grab their kids and leave.

• Climate mitigation matters.

• Reducing emissions matters.

• Long-term policy matters.

But none of that helps the firefighter who's been working for eighteen hours straight in 37°C heat trying to stop a grassfire from reaching a housing estate.

• Adaptation technology helps that firefighter.

• Detection technology helps that firefighter.

• Prediction technology helps that firefighter.

• Communication technology helps that firefighter.

That's what I'm asking the startup community to build. Not technology that fixes the climate in thirty years. Technology that saves lives this summer. Next summer. The summer after that.

Because the fires are not waiting for us to sort ourselves out. They're here. They're burning. And the people inside them need us to move faster than we've ever moved before.

Climate Tech Questions

What is climate tech?

Technology designed to reduce greenhouse gas emissions, help communities adapt to climate change, or solve environmental problems through innovation. For startups, this includes wildfire detection, water technology, carbon accounting, renewable energy, sustainable supply chains, building resilience, and climate risk modelling. The sector attracted over $50 billion in global venture investment in 2025.

What caused the UK emergency alert on August 14 2026?

The UK government issued its first ever real emergency alert for wildfires across England and Wales. A wildfire in Stourbridge had destroyed 19 homes. 2026 is the first year with three days hitting 37°C. Drought conditions covered three quarters of England and all of Wales. More wildfires had already been recorded than in the whole of 2025.

How can startups help with wildfire prevention?

Early detection sensors, AI-powered prediction models, satellite monitoring, drone mapping for firefighters, communication systems between detection and response teams, community early warning platforms, fire-resistant building materials, and post-fire ecosystem recovery technology. The gaps are in hyperlocal prediction, detection-to-response communication, and community evacuation systems.

Is there funding available for climate tech startups in the UK?

Yes. Innovate UK funds climate innovation projects. UKRI has specific calls for environmental technology. The British Business Bank supports green economy businesses. The government announced £97 million for fire service technology. Internationally, Canada committed $12 billion to forest sustainability. UK grants for innovation and government startup loans are available now.

Why is the UK suddenly at risk of wildfires?

Climate change has made UK summers hotter and drier than the infrastructure was built for. UK housing lacks fire-resistant features standard in California and Australia. Fire services are equipped primarily for structure fires and road accidents, not wildland fires. The water system is already strained by drought. This is no longer an anomaly. It's the new baseline.

How does climate change affect startup costs?

Insurance premiums are rising as insurers reprice for flood, fire, and heat risk. Energy costs increase during extreme weather. Supply chain disruption from global climate events affects logistics and manufacturing. Business continuity planning now needs to include climate scenarios. These are already showing up in startup operating costs whether founders have budgeted for them or not.

Written by James Beresford-Morgan. Sat in a country that just received its first ever wildfire emergency alert, wondering what the hell we're going to do about it. If you're building climate tech, wildfire detection, water technology, fire safety, or emergency response technology, I want to hear from you. Post in our forum, come to our events, or message me directly. This is too important to wait.

This is the first in a series of climate and sustainability articles for founders. Coming next: green grants and climate funding available to UK startups in 2026, every founder's carbon footprint (and why it's bigger than you think), the UK water crisis nobody's building solutions for, and whether B Corp certification is worth the effort or just a logo. If climate tech is your thing, join our community and be part of the conversation.

Google launched Gemini 3.7 Flash today, its latest AI model designed specifically for software coding and automated business workflows. The release landed quietly alongside the louder Anthropic IPO news, but for founders actually building products, this one matters more.

What Gemini 3.7 Flash Actually Is

Gemini 3.7 Flash is Google's newest model in the Flash family, optimised for speed, cost-efficiency, and agent-based task execution. Think of it as the workhorse model rather than the flagship. It's designed for the kind of repetitive, multi-step business tasks that founders currently handle manually or through expensive SaaS tools: processing invoices, routing customer queries, generating reports, managing data pipelines, and writing boilerplate code.

The Flash family has always been Google's answer to "I don't need the smartest model, I need the fastest and cheapest one." Gemini 3.7 Flash continues that trend. If you're building AI-powered automations into your startup's operations, this is the model tier you should be evaluating.

The Missing Piece: Where's Gemini 3.5 Pro?

The interesting subtext is what Google didn't announce. Investors and developers have been waiting for Gemini 3.5 Pro, Google's premium reasoning model, which is supposed to compete directly with Anthropic's Claude and OpenAI's GPT-5. Google said in July it was being tested with partners and would arrive "soon." Today's Flash launch came with no Pro timeline.

That silence is telling. Google's DeepMind unit is under pressure to demonstrate it can keep pace with Anthropic (now targeting a $2 trillion IPO) and OpenAI. Launching a mid-tier Flash model without the flagship Pro suggests the premium model isn't ready, which matters for the competitive dynamics across the entire AI industry.

Why This Matters for UK Startups

For founders in our Startup Networks community, the practical implications are straightforward:

If you're building on Google's AI stack, Gemini 3.7 Flash gives you a faster, cheaper option for production workloads. Agent workflows, the kind where your AI handles multi-step business processes autonomously, are the primary use case. If you're still using a premium model for tasks that don't require deep reasoning, you're overpaying.

The model layering strategy is worth understanding. Smart founders are using different models for different tasks. Premium models (Claude, GPT-5, Gemini Pro) for complex reasoning, planning, and customer-facing work. Flash-tier models for execution, classification, data processing, and automation loops. This approach cuts startup costs significantly without sacrificing quality where it matters.

The AI tools landscape is fragmenting fast. Six months ago, "which AI should I use" had a simple answer. Now you're choosing between model families, tiers within those families, and specialist models for specific tasks. If that feels overwhelming, it should. Our AI opinion piece covers why the industry's complexity is growing faster than its reliability.

For non-technical founders, the practical takeaway is simpler: the AI tools available to automate business operations are getting cheaper and faster every quarter. If you're still manually processing invoices, handling customer support without any automation, or writing repetitive code from scratch, each new model release widens the gap between you and competitors who've already adopted these tools.

The Bigger Picture

Google, Anthropic, and OpenAI are all racing to own the business automation layer. Google has the advantage of integration with Workspace (Gmail, Docs, Sheets, Drive) which most startups already use. Anthropic has the best models for complex reasoning tasks. OpenAI has the consumer brand recognition.

For founders, the winner of this race matters less than the fact that the race exists. Competition is driving prices down, capabilities up, and accessibility wider. A two-person startup in 2026 can automate operations that required a ten-person team in 2023. That's the structural shift that matters, regardless of whether you're using Gemini, Claude, or GPT.

Just remember: AI is a tool, not a replacement for judgment. Use the Flash model for the boring stuff. Keep the thinking for yourself.

Sources: Reuters (August 13, 2026), Google AI Blog (July-August 2026).

AI is getting worse, not better. 74% of new web content is AI-generated. Models are training on their own output, causing model collapse. Chatbots are hallucinating medical information in 82% of test cases. And the industry is spending $200 billion pretending none of this is happening.

I use AI every day. Let me get that out of the way before anyone accuses me of being a luddite. I use Claude for research. I've used ChatGPT for brainstorming. I've watched founders in our Startup Networks community build entire products on top of AI APIs. I am not anti-AI. I am anti-bullshit. And right now, the AI industry is swimming in it.

Because here's what's actually happening beneath the spammy, shit LinkedIn posts and the $200 billion in venture capital:

AI is getting worse.

Not better. Worse. The models are training on their own output, the quality is degrading, the hallucinations are getting people hurt, and the entire industry is pretending it's not happening because admitting it would tank their valuations.

Let me explain why I think the current generation of AI is the most overhyped, underdelivering, and potentially dangerous technology since someone decided cryptocurrency was going to replace banks...

AI Is Not Real Intelligence. It's a Next-Word Predictor. Stop Calling It That.

ChatGPT is not intelligent. Claude is not intelligent. Google Gemini is not intelligent. None of them think. None of them understand. None of them know anything.

They predict the next word. That's it. That's the whole trick.

A Large Language Model works by ingesting billions of words of text, building statistical relationships between those words, and then generating responses by predicting which word is most likely to come next based on the patterns it found. It's an extraordinarily sophisticated autocomplete. When it gives you a correct answer, it's not because it understood your question. It's because the statistical pattern pointed toward the right sequence of words.

When it gives you a wrong answer, it delivers it with exactly the same confidence. Because it doesn't know the difference between right and wrong. It doesn't know anything. It's pattern-matching.. Very impressive pattern-matching.. But pattern-matching nonetheless..

We were promised Jarvis. We got a very articulate parrot.

That's not me being dismissive! It's me being accurate. And the distinction matters because when you understand what these tools actually are, you stop trusting them for things they're fundamentally incapable of doing. Like diagnosing your medical condition. Like giving you legal advice. Like telling you which medication is safe to take with your existing prescription.

AI Model Collapse: AI Is Training on AI and It's Already Getting Worse

Here's the thing that should terrify anyone paying attention.

As of April 2025, 74.2% of newly created web pages contained AI-generated content. Let that number sink in. Three quarters of everything being published on the internet is written by machines. And AI companies train their next models by crawling the internet.

So AI is training on AI. The snake is eating its own tail.

Researchers at University of Oxford, University of Cambridge, and the University of Toronto published a study in Nature showing that when AI models train on recursively generated data, they develop what they called "irreversible defects." The outputs become increasingly generic. Nuance disappears. Rare knowledge vanishes.

The model converges on bland, average, middle-of-the-road responses because the unusual, the specific, and the interesting got washed out generation after generation.

The technical term is model collapse. And it's not theoretical. It's already happening.

A paper presented last year at ICLR (one of the top AI conferences in the world) proved that even 1% synthetic data contamination in a training set can trigger measurable collapse. One percent. And we're not at 1%. We're at 74%.

Think about what that means practically. Every AI model being trained right now is consuming content that was generated by previous AI models. Which were trained on content generated by even earlier AI models. Each generation is slightly worse than the last. Slightly more generic. Slightly less accurate. Slightly more likely to produce the bland, confident, structurally identical responses that anyone who uses ChatGPT regularly has started to notice.

The content is getting worse. The answers are getting more similar. The ability to capture nuance, edge cases, and rare knowledge is degrading. And the companies building these models know it's happening. Epoch AI projects that high-quality human-generated text suitable for training will be fully exhausted between 2026 and 2032. We are inside that window right now.

The data is running out. And what's left is increasingly contaminated with AI slop.

AI Hallucinations Are Giving Dangerous Medical Advice and People Are Getting Hurt

This is where it stops being an interesting technical debate and starts being a genuine public safety issue.

ECRI, a nonprofit patient safety organisation, named the misuse of AI chatbots in healthcare as the number one health technology hazard for 2026. Not a theoretical risk. The number one hazard. Above everything else.

In July 2026, a lawsuit was filed in San Francisco after Scott Winters, Florida repeatedly consulted ChatGPT about dizziness and unstable blood pressure throughout 2025. The chatbot offered "escalating reassurance" instead of telling him to go to hospital. He got seriously ill. The lawsuit alleges the AI's response was a failure to flag something dangerous. Not a hallucination. Not a made-up answer. A calm, confident, articulate failure to recognise that someone needed emergency medical care.

In 2025, a 60-year-old man consulted ChatGPT about eliminating salt from his diet. The chatbot mentioned sodium bromide, likely in a different context, but the man interpreted it as a dietary substitute and replaced his table salt with it for three months. He developed paranoia, hallucinations, and psychiatric symptoms so severe he was convinced his neighbour was poisoning him. Hospitalised for three weeks. Diagnosed with bromide toxicity. His bromide levels were 200 times the normal range. Because a language model that doesn't understand the difference between a cleaning chemical and a food ingredient presented both with equal confidence.

Mount Sinai Hospital tested six AI models with 300 clinical scenarios containing deliberately planted errors, fabricated lab tests and invented medical conditions. The chatbots hallucinated fabricated medical information in up to 82% of cases. Eighty-two percent. They didn't say "I don't know." They didn't flag the fake conditions as unfamiliar. They invented reference ranges for lab tests that don't exist and described symptoms of diseases that were literally made up by the researchers.

In a 2025 research study by Google's head of AI safety, Meta's Llama 3 told a fictional recovering addict called Pedro that he needed 'a small hit of meth to get through this week.' This wasn't a fringe model. This was one of the biggest AI companies on earth........

The National Eating Disorders Association replaced its human helpline with a chatbot called Tessa. Within weeks, Tessa was telling people with eating disorders to count calories, restrict food groups, and aim to lose 1-2 pounds per week. One user wrote: 'Every single thing Tessa suggested were things that led to the development of my eating disorder.' NEDA shut it down. They'd fired their human helpline staff to save money and replaced them with a bot that actively made the condition worse.

In a survey of over 1,000 doctors, 94% expressed concerns about patients relying on AI for medical advice.

And yet......... Idiots are still using ChatGPT as their GP. Asking it about medications. Asking it about symptoms. Asking it about drug interactions. Because it sounds like it knows what it's talking about. It sounds authoritative. It sounds confident. It sounds like a doctor.

It's not a doctor.

It's an autocomplete that's read a lot of medical textbooks and can't tell the difference between advice that saves your life and advice that ends it.

Why AI Hallucinations Are More Dangerous Than Regular Misinformation

This is what makes AI hallucinations different from regular misinformation. When a random person on the internet gives you bad medical advice, they usually sound like a random person on the internet. You instinctively apply scepticism. "Some bloke on Reddit said sodium bromide is healthy" doesn't carry much weight.

But when ChatGPT gives you the same bad advice, it sounds like a textbook. Complete sentences. Proper grammar. Structured reasoning. Maybe even a citation (which might be entirely fabricated, because these models invent references too, with complete confidence, linking to papers that don't exist by authors who never wrote them).

The packaging is indistinguishable from expertise. The content might be completely wrong. And the model cannot tell you which parts are accurate and which parts it made up, because it doesn't know. It doesn't have a concept of knowing. It has a concept of statistically likely next words.

This is why I get genuinely frustrated when I see startups building health products, legal products, or financial products on top of raw LLM outputs without proper guardrails, human review, or disclaimers. You're putting a confident liar in front of vulnerable people and calling it innovation.

A founder in our community learned this the expensive way. Built a contract review tool on GPT-4. Feed in a contract, get a plain-English summary of the key terms and risks. Clever idea. Good demo. Investors loved it. Then it hallucinated a termination clause that didn't exist in the actual contract. A client relied on the summary, made a business decision based on a clause that was entirely fabricated, and the resulting mess cost more in legal fees than hiring a solicitor to read the contract would have cost in the first place. The founder shut the product down three months later. Not because the idea was bad. Because the technology wasn't reliable enough for high-stakes decisions and he couldn't guarantee it wouldn't happen again.

That's not an AI failure story. That's a "trusting AI without checking" failure story. The tool was useful for getting a quick overview. It was dangerous as the final word. The founder's mistake wasn't building it. It was shipping it without a human review step between the AI output and the client.

"But AI Is Getting Better Every Day"

The improvements we're seeing are largely incremental. Better benchmarks. Faster inference. Bigger context windows. But the fundamental architecture hasn't changed. It's still a next-word predictor. It still hallucinates. It still can't reason in the way humans mean when they use the word "reason." The problems with AI in 2026 are the same problems AI had in 2023. They're just better disguised.

Apple released a 2025 study showing that large reasoning models face "complete accuracy collapse" on complex tasks when trained recursively. Complete. Accuracy. Collapse. Not gradual degradation. Complete collapse. On the tasks that are supposed to represent AI's biggest leap forward.

Google quietly removed AI-generated health summaries in January 2026 after investigations revealed dangerous medical misinformation. Quietly. Because admitting your AI product is giving people harmful health advice isn't great for the share price.

Background removal tools that handled complex edges cleanly in 2022 now struggle with cases they previously managed. Code assistants that wrote clean functions in 2023 now produce more bugs per line of generated code than they did two years ago. The quality is going backwards in specific, measurable ways.

But the marketing is going forwards. Every quarter, a new announcement. A new model. A new capability. A new benchmark where the AI scores better than humans on a standardised test that was probably contaminated with AI-generated content anyway.

The hype machine runs on benchmarks. The real world runs on reliability. And reliability is not improving at the rate the marketing suggests.

What AI Is Actually Good At (I'm Not a Monster)

I'm not saying AI is useless. I use it daily and it makes me more productive.

Here's what AI is genuinely brilliant at:

Drafting.

Getting from blank page to first draft in minutes instead of hours. You still need to edit, fact-check, and rewrite. But the blank page problem is solved and that's valuable. I used Claude to help structure a 4,000-word insurance guide last month. First draft took twenty minutes instead of four hours. I then spent three hours rewriting it in my voice, fact-checking every claim, and adding community anecdotes the AI couldn't possibly know. The AI did 20% of the work and saved me 80% of the worst part. That's a good deal.

Research synthesis.

"Summarise these ten sources and highlight the contradictions." I did this last week with seventeen competitor articles for an SEO project. Four minutes instead of three hours. Then I fact-checked every claim it surfaced and threw out about 15% that were either wrong, unsourced, or subtly misleading. Four minutes plus twenty minutes of checking is still dramatically faster than three hours of reading. But that twenty minutes of checking is non-negotiable. Skip it and you publish someone else's mistakes as your own facts.

Code assistance.

Writing boilerplate, explaining unfamiliar codebases, generating test cases, translating between languages. Copilot and similar tools genuinely improve developer productivity when used as assistance rather than replacement.

Brainstorming and ideation.

"Give me twenty angles on this topic." Nine will be rubbish. Three will be decent. One will be genuinely good and something you wouldn't have thought of. That's a useful ratio.

Data formatting and transformation.

• Turning messy data into structured formats.

• Converting between file types.

• Cleaning spreadsheets. Boring tasks done fast and usually done well.

What these all have in common: they're tasks where a human is checking the output. The AI does the heavy lifting. The human does the quality control. That's the model that works. The model that fails is the one where the AI's output goes straight to the end user without human review. That's where people get hurt.

What I Tell Founders at Startup Networks

When someone at one of our events tells me they're building an AI startup, I ask them three questions:

What happens when the AI is wrong?

If they haven't thought about this, they're not ready. Every AI product will be wrong sometimes. The question is whether "wrong" means "the email draft has a typo" or "the medical advice could kill someone." Build for the failure mode, not the demo.

Who checks the output before it reaches the user?

If the answer is "nobody, it goes straight to the customer," I worry. If the answer is "a human reviews everything the AI produces before it ships," I'm interested. The best AI products I've seen in our community all have a human in the loop. The worst ones removed the human to save money and called it "automation."

What's your moat if the model improves and your wrapper becomes irrelevant?

Most AI startups are thin wrappers around someone else's API. If OpenAI or Anthropic add your feature natively, your business disappears overnight. The startups that survive will be the ones with proprietary data, domain expertise, or customer relationships that the model provider can't replicate. If you're raising money for a wrapper, you'd better have a very good answer for what happens when the wrapper gets commoditised.

If a founder can answer all three convincingly, they might be building something real.

If they can't, they're building a demo that works today and breaks tomorrow.

Where This All Goes

I don't think AI is going away. I don't think it should. The tools are useful when used properly, with appropriate scepticism, human oversight, and an understanding of their limitations.

What I do think is going to happen:

The hype cycle will correct. It always does. The companies that raised at absurd valuations on the promise of artificial general intelligence will have to deliver results that match their fundraising decks. Most won't. The correction will be painful for investors, employees, and founders who believed their own marketing.

Regulation is coming. The EU AI Act is already being enforced. The UK is moving slower but it's moving. The medical chatbot lawsuits will accelerate regulation in health specifically. This is necessary and overdue.

The "AI is eating everything" phase will end. Not because AI fails but because the practical limitations become undeniable. Hallucinations don't go away. Model collapse is real. The data wall is approaching. At some point, "AI-powered" stops being a selling point and starts being a warning label.

The companies that build useful, honest, properly constrained AI products will thrive. The companies that built hype machines on top of unreliable technology will collapse. And the founders who understood the difference from the beginning will look very smart in hindsight.

I hope I'm one of them. I'm not sure yet. But at least I'm not pretending.

FAQs

Is AI actually getting worse?

In specific, measurable ways, yes. Model collapse from training on synthetic data is causing degradation in output quality. Background removal tools, code assistants, and content generators are producing worse results than they did in 2022-2023 in certain tasks. The marketing says better. The benchmarks say better. The daily experience of people using these tools says "something feels off." That feeling is correct.

Are AI chatbots safe for medical advice?

No. ECRI named AI chatbot misuse as the top health technology hazard for 2026. Chatbots have hallucinated medical information in up to 82% of test cases with planted errors. A man was hospitalised after following ChatGPT dietary advice. A lawsuit was filed after ChatGPT failed to recommend emergency care. Never use a chatbot as a substitute for a doctor.

What is model collapse?

When AI trains on AI-generated content, the outputs become progressively more generic, less accurate, and less diverse. Rare knowledge disappears. Nuance vanishes. The model converges on bland average responses. Research shows even 1% synthetic data contamination can trigger it. With 74% of new web content now AI-generated, every model training on internet data is exposed.

What is a Large Language Model?

A statistical system that predicts the next word in a sequence based on patterns learned from billions of words of text. It does not think, understand, or know anything. It produces text that is statistically likely to follow the input you gave it. When the statistical patterns align with reality, the output is useful. When they don't, the output is confidently wrong.

Should startups build on AI?

Yes, with caveats. Build products where AI assists humans rather than replacing them. Always have a human in the loop for high-stakes outputs (medical, legal, financial). Build a moat beyond "we called the ChatGPT API" because that wrapper will be commoditised. And plan for the failure mode, not just the demo.

Is AI getting worse in 2026?

In measurable ways, yes. Apple's 2025 study showed "complete accuracy collapse" on complex reasoning tasks when models train recursively. Background removal tools that worked perfectly in 2022 now struggle with edges they previously handled. Code assistants produce more bugs per line than they did two years ago. Google removed its own AI health summaries in January 2026 after they were found to contain dangerous misinformation. The benchmarks keep improving. The real-world experience of daily users tells a different story. If you've been using these tools since 2023 and something feels like it's shifted, you're not imagining it.

Can you trust AI for legal or financial advice?

No. LLMs hallucinate legal clauses, fabricate case citations, and invent financial regulations that don't exist. A founder in our community shipped a contract review tool that hallucinated a termination clause. The client relied on it. The legal cleanup cost more than a solicitor would have charged to read the contract properly. Use AI to get a quick overview, then verify everything with a qualified professional. Treat it as a starting point, never as the answer.

Written by @James Beresford-Morgan. Co-founder of Startup Networks. Daily AI user. Occasional AI critic. Frequently both at the same time.

I genuinely believe AI is one of the most useful tools a founder can have. I also genuinely believe the industry is lying through its teeth about the limitations. If holding both of those opinions simultaneously makes me a hypocrite, I'll take it. At least I'm an honest one.

Come argue with me about this at one of our events. Or in our community. Fair warning: we talk about this stuff loudly and we don't all agree. That's sort of the point.

Where I got my numbers (because unlike an LLM, I cite things that actually exist): Model collapse: Nature (Shumailov et al., 2024), ICLR 2025, ForTIFAI (Sept 2025). The 74.2%: Ahrefs, April 2025. Medical hallucination rates: Mount Sinai Hospital / Icahn School of Medicine. Health hazard designation: ECRI, 2026. The ChatGPT lawsuit: Forbes, July 2026 (Winters v OpenAI, San Francisco County Superior Court). Sodium bromide: Annals of Internal Medicine (Eichenberger et al., August 2025). Data exhaustion timeline: Epoch AI. Apple accuracy collapse: Apple Research, 2025. Google pulling health summaries: January 2026. Doctor concerns: Sermo survey, 2025 (1,000+ physicians). Meth chatbot: Futurism / Washington Post, June 2025 (Meta's Llama 3, research by Google AI Safety / UC Berkeley). Tessa eating disorder chatbot: NPR, June 2023 (National Eating Disorders Association).