"Adopt AI or die."
Every other flippin’ headline this year says some version of it. I wrote in August that, as advice to an MD with an actual business to run on a blustery Monday morning, it's worth nothing. If anything I was being generous.
When me and Richard sit down with a founder or an MD, these are the questions that come up:
Where do we begin?
How do I trust it?
Will it help, or is it just another thing to manage?
What happens when it gets something wrong?
Only one of those is really about the tech.
So ignore the "or die" bit.
Pick one job that eats hours or margin every week, put a named person in charge of it and start there.
Want some guidance? Drop me me a DM.
Every system I build starts with the same question.
What happens when it's wrong?
Not if. When. A model will misread an invoice. A forecast will get a bank holiday wrong. The question is what the system does next.
Ours stop. If the evidence is stale, the agent doesn't guess, it fails closed and puts the decision in a queue for a person.
If you can't say what your system does when it's wrong, it isn't ready to switch on. Neither are we.
20% of respondents say the cost of running AI, including token costs, has already constrained how much they use it. (McKinsey 2026).
Yet 60% are planning to increase their AI investment over the next year.
Which is quite funny when you think about it.
And we’re starting to see this exact problem in client conversations.
You build something. It works. People like it. You get the feedback, the testimonials, everyone gets a bit excited about what you could build next.
Then you work out what it costs to run the whole thing properly every month 😂
And suddenly “let’s build more” isn’t necessarily the obvious answer.
Who still needs to sign something off?
What happens when it gets something wrong?
Where do you absolutely NOT want AI making the decision?
And does any of this still make sense when it’s doing something 1,000 times a month instead of 10?
Sometimes it does.
But sometimes the best answer is actually pretty boring.
One agent. Doing one annoying job really well. Saving someone hours every week. With a human still in control.
I’m becoming increasingly convinced that more AI doesn’t automatically mean progress.
If it saves time, works reliably and you can afford to keep the bloody thing running, you’re probably doing alright.
If you’re at that point where you’ve proved AI can do something but you’re not sure what you should actually build next, give me a shout.
That’s becoming one of my favourite conversations to have.
3
30
At Woodlark, we don't do demos.
Day eight of a Blueprint, we take your last month and run it through the system we've designed. Real invoices, real orders, real forecast history.
You see every decision it made and why. You see the ones it got wrong, flagged, not hidden.
If it doesn't hold up, you've found out for the price of a Blueprint, not a Build. That's the point of the fortnight.
A demo shows you what the software can do. Your own data shows you what it would actually do on a Monday.
Nobody tells you that when your business fails, you grieve it like a death.
WAF closed in February and for weeks I didn't know what to do with myself. Not in a productive "what's next" way. In a staring at the wall, can't sleep, replaying every decision kind of way.
You build something for four years. You recruit people. You ask them to trust you. You raise money from people who backed your vision. And then one day you're standing in a warehouse working out how to return other peoples stock while your staff are being made redundant around you.
That bit doesn't make the LinkedIn posts. The "lessons learned" content comes later, once you've had enough distance to sound composed about it.
The honest version is messier. I was angry for a while. Angry at the market, angry at decisions we made too late, angry at myself for not seeing things sooner. Richard and I had conversations during that period that were harder than anything we faced while the business was actually running.
But the bit I keep coming back to is this. The grief wasn't really about the business. It was about the identity. For four years I was "the WAF guy" and then suddenly I wasn't anything.
Starting Woodlark with Richard wasn't some grand pivot strategy. It was two people who trust each other deciding they weren't done yet.
That's it. No masterplan. Just not done.
1
25
We launched Gardn in 20 countries and 13 languages in under a year.
Our team? Two founders and 30 AI agents.
No localisation agency. No international expansion playbook. No dev team sitting in a room whiteboarding sprint cycles for six months.
Richard and I built the entire thing from a coworking space in Nottingham with a monthly model bill that would make most VCs spit out their coffee.
The thing nobody tells you about launching in 20 countries is that the hard part isn't the tech. It's the thousand small decisions that stack up. Which markets first. Which languages actually matter versus which ones look good on a pitch deck. How you handle customer support in Portuguese when neither of you speaks a word of it.
We got it wrong plenty of times. Early versions of the German copy were apparently hilarious for all the wrong reasons. A user in Japan flagged that our plant recommendations would have killed everything in their climate zone.
But here's the thing. Every single one of those problems got fixed in hours, not quarters. Because when your agents are governed properly and your feedback loops are tight, you move at a speed that doesn't make sense to people used to traditional teams.
Most founders I speak to think going international means raising another round and hiring 15 people. We did it on a credit card and a lot of stubbornness.
The app is now live, growing, and we still haven't hired a single employee.
40
Buzz is great on so many levels.
@jack - constantly one step ahead, I doff my cap sir.
I think it hits the right note in terms of being useful and fun.
The power of it is probably yet to be known - but it will certainly have shaken @SlackHQ up and it's day 1.
If there will be more agents than humans on earth within the next 12 months - who should we be building for?
Being able to achieve terminal level power, LLM agnostic, locally and be encouraged to share compute/ go open source - is just next level.
I've used it all day today and just had fun, but also started to test it and it certainly is powerful whilst not being overbearing.
Hopefully me and the team will be spending our days in here - and hopefully tomorrow with Opus 5 @AnthropicAI.
Much kudos.
2
175
Your lawn hasn't died, it's gone dormant. In heat, grass sends its energy to the roots and waits for rain. The kindest move this week is doing less: mower blades up, no feed, no watering. #GrowTogether
Download Gardn: gardn.world?utm_source=x&utm…
20
You only get one first impression. This week we're auditing every flow, screen and edge case in Gardn before we go live. Two founders, one app, one launch. 'Done' never really arrives — you just decide to go. #GrowTogether
1
27
Anthropic: “Fable is an agentic coding superweapon, capable of developing cyber- and bio-weapons at unprecedented speed and scale. We cannot in good faith release it without guardrails.”
China: “lmao here’s Fable but open-source. Good fkn luck”
Introducing Kimi K3: Open Frontier Intelligence
🔹 2.8 Trillion Parameters, 1 Million Context, Native Multimodal
🔹 Kimi Delta Attention enables up to 6.3x faster decoding in million-token contexts
🔹 Attention Residuals deliver ~25% higher training efficiency at <2% additional cost
🔹 Built for long-horizon agentic coding and self-evolving workflows
Kimi K3 is now live on on Kimi.com, Kimi Work, Kimi Code, and the Kimi API.
Open Weights by July 27, 2026.
🔗 API: platform.kimi.ai
🔗 Tech blog: kimi.com/blog/kimi-k3
AI is modern witchcraft: it works before society agrees what kind of power it is.
Pandora's box is open. It will not lock again.
The question is not who gets magic.
It's who may cast in public, what they can touch, and what happens when they get it wrong.
2
23
Every AI agent product wants your trust on day one. Full inbox access, connected accounts, off it goes.
We're building the opposite. An AI executive team that starts with zero authority and earns it — one approved decision at a time.
Trust as a ledger, not a toggle.
1
40
Trent Peek retweeted
Replying to @England_Pod
@England_Pod If you need to fill the gap before the Ghana game, @TJPeek and I have built a version of 38-0 for England at the World Cup. You can choose your all time England XI and see how they would have fared at any World Cup from 1950 to present day. 60yearshurt.com
If you're looking to fill the gap before England play Ghana, I've made this 38-0 style game for England at the World Cup. How would a front three of Stanley Matthews, David Beckham and Harry Kane fare in any World Cup from 1950-2026? Can you bring it home? 60yearshurt.com/
3
65
60 Years of Hurt 🏴
2002 World Cup
🟩🟨🟩🟩🟩🟩🟩
CHAMPIONS 🏆
60yearshurt.com/
3
31
For the past 18 months, I’ve been building something new.
It’s called Gardn. 🌱
An app that remembers your garden.
We’re now preparing to launch across 20 countries and multiple languages.
Still feels a bit surreal.
More soon. 🌍#GrowTogether
1
4
33
Trent Peek retweeted
It will never feel normal to see Forest signing players like Kalimuendo, McAtee and (potentially) Douglas Luiz and Hutchinson. 10 years ago, we signed Matt Mills on a free. #nffc
Trent Peek retweeted
Christmas Pilgrimage 2024 at @sneintonprimary: a thread

















