I built my own AI review system around my dyslexia - it halved my review time

Friends,

your weekly AI briefing is here - designed to help you respond to AI, not react to the noise. No curveballs. No chaos. Just clarity.

📰 This was the week that was...

You will have seen the headlines this week.

Jacob Coxon, a researcher who quit Anthropic, told the BBC's Sunday with Laura Kuenssberg programme that without a slowdown "there is a strong chance that we could all die in the immediate future" - though he also said researchers "genuinely want to see the upside": solving diseases, improving lives. (Oscar Pickford-Wardle flagged this to me before it broke - he was ahead of the news on this one.) Dario Amodei argued for a slower pace in an essay the same week, saying extra time now could "greatly reduce the risk that something goes seriously wrong." Sam Altman told Salesforce's conference that "the world is right to be afraid of this" - then asked it to trust OpenAI anyway; Donald Trump, quoted in the same report, called the fears a "hoax," having already dismissed them in Ireland as things that "won't happen."

News headlines don't appear by accident, and the question I always ask is who benefits from them. Given the immense size of the upcoming IPOs, I wonder if trying to cool the AI market is a way to de-risk those economic events.

Let's get into it.

🔥 Urgent Priorities

✅ No fires to fight this week

✅ AI's most-watched leaders each told the world what to feel this week, and each one has money riding on which way you land

✅ Worth doing now: find out what your own team actually thinks, before the noise decides it for them

No panic needed this week. What it does call for is five minutes with whoever on your team has gone quiet on AI lately, to hear what they've actually seen.

🎯 Strategic Insight

Tension: This week, Dario Amodei made the case for slowing AI down, Sam Altman said the world is right to be afraid of it while asking the world to trust his company anyway, and Donald Trump called fears about AI's safety a hoax. Each of them has a stake in which argument wins. None of them has to decide what your team does on Monday morning. You do.

Optimistic insight: You already have the ability to manage this. My AI risk taxonomy exists to help a leader move at speed while adopting AI safely - it foregrounds the risks a business actually has to manage: data access, data residency, cybersecurity. The risks in this week's headlines belong to a small number of companies racing to build the most capable frontier AI models in the world. Your job is different, and entirely within your control: your AI governance and procurement policy, making sure the partners you choose are upholding AI safety and ethics themselves.

What's shifting: Silicon Valley has a blind spot worth naming: it's brilliant at building the technology, and much less used to the fact that society has its own checks and balances - a court ruling, a regulator's decision, a shift in liability law - that can reshape what's viable long before any lab's own timeline plays out. Building models well doesn't make anyone an authority on how society, law or human behaviour actually work.

Inside your team: this week's headlines will do real damage there too: fear is contagious, and it lands hardest on people who were already unsure. I call the capable people who hold real reservations about AI "conscientious objectors," and the way through is holding space for the actual conversation - what have you seen that worries you, and what would need to be true for you to feel differently. The Emerald podcast's episode on sorcerers and the age of AI - a few years old now, and recommended to me by Sophie Tidman - tells a version of the same story: the apprentice's real mistake was invoking a force he didn't yet understand. The clearest question of the week came from Logical Intelligence's Patrick Hillman, aimed straight at the labs: "If you believe what you are building is dangerous, show us what you are prepared to stop doing."

Why this matters now: Freezing isn't free. Every week your team spends unsure whether to lean into AI is a week a competitor spends leaning in - and good AI governance is what lets you keep moving with confidence, because it keeps the safety question where it belongs: with the suppliers building the frontier models, under your own procurement standards.

Action: Build your team a short, written list of what they can decide to do with AI on their own, no permission needed - three items is plenty to start. Leaders are being asked to make bolder calls, faster, on less data, against more uncertainty than at any point in this technology's short life, and structure is what lets people act with confidence at that speed. Put the list in writing this month, and tell your team plainly that trying is expected. That's what keeps a conscientious objector in the room.

Fear moves fast this week; curiosity is the only thing that outlasts it.

A quick note from Hugo: governance and curiosity both take real skill. For UK based readers, I can help directly: AI Savvy Leaders is a free, government-funded course for leaders, run with Working Knowledge - email [email protected] and Oscar will talk you through it. (Disclosure: AI Night School is my own venture.)

🤓 Geek Out

1️⃣ Scientists are saving nearly seven hours a week with AI - and hitting a new bottleneck

Google, Google DeepMind and MIT FutureTech have found that three in four scientists are already saving time with AI - nearly seven hours a week on average, most of it ploughed back into more research. The catch: nearly half of them now spend more than a quarter of that saved time checking the AI's work, and four in ten report a growing backlog of untested hypotheses. The bottleneck simply moved, from generating ideas to checking them.

Why it matters: how you design where and how humans review AI output is emerging as a discipline in itself - and it can be personalised to how each of us actually reviews best. I've been building my own review system around my ADHD and dyslexia, and it's halved the time it takes me to review AI outputs, at around a quarter of the effort.

👉 Action: before your next AI rollout, name who's responsible for checking the output - that role needs an owner just as much as the licence does.

👉 Read the report

2️⃣ AI-for-good, at two very different scales

The Gates Foundation has committed at least $1 billion over the next two years to get AI tools to health workers, teachers and farmers in some of the world's poorest countries, spanning education, healthcare, agriculture and the digital infrastructure to make it work in languages the big models don't yet handle well. Its own framing: AI could be "a great equalizer, but only if it's built intentionally to benefit the people who need it most." Closer to home, Startup Sherpas is tackling the same gap at a much smaller scale, giving teenagers the real-world skills they'll need to thrive in a working world already being reshaped by AI. (Disclosure: I'm a co-founder of Startup Sherpas. It's a social enterprise with a mission to build a generation of teenagers ready for the world of work ahead of them.)

Why it matters: this is what AI-for-good looks like at two different scales - a $1bn global commitment, and a grassroots programme doing the same job for the next generation closer to home.

👉 Action: if your business runs any community, education or supplier-development programme, look at whether an AI tool could extend its reach the same way - cheaply, to people who wouldn't otherwise get access.

👉 Read the Gates Foundation's announcement

3️⃣ A humanoid robot just learned to work safely next to people, without the cage

Agility Robotics has unveiled Digit 5, a humanoid robot built to work alongside people without the safety barriers usual industrial robots need. Its safety system continuously watches for people nearby and slows or stops if anyone gets too close, and it charges fast enough to stay working most of a 24-hour day. Agility, working with NVIDIA on the safety side, calls it the company's first robot built to this standard from the ground up.

Why it matters: the real barrier to humanoid robots doing ordinary work has always been keeping people safe around them - and that's the barrier this removes, which is why it's worth watching even if you'll never buy one.

👉 Action: if your operations mix people and heavy automation, ask your equipment suppliers whether anything like this safety approach is on their roadmap - a good proxy for how close barrier-free robotics is to reaching your sector.

👉 Agility Robotics' announcement

🎨 Weekend Playground

A book recommendation from my mum: Yanis Varoufakis's Another Now: Dispatches from an Alternative Present - a novel by the economist and former Greek finance minister, built around its own opening premise: "Imagine a world with no banks. No stock market. No tech giants. No billionaires." After a week spent asking who benefits from the headlines, a novel that removes the beneficiaries entirely felt like the right weekend companion.

Why this matters: imagining an economy built on entirely different assumptions is a real way to test which parts of how we work today are actually necessary, and which ones we've just stopped questioning.

👉 Mission:

• Get a copy from the library or secondhand - it's a five-year-old paperback, easy to find either way

• Read the first chapter or two this weekend, enough to get properly into the world he's built

• Notice which of the four - no banks, no stock market, no tech giants, no billionaires - strikes you as the most implausible, and ask yourself why

📢 Share the Optimism

If The AI Optimist helps you think more clearly, forward it to someone else handling the shift.

And here's the question I'm curious about this week: what's your team actually saying about AI right now, when you're not in the room? Reply and tell me - I read every message and I'll come back to you personally.

Stay strategic, stay generous.

Hugo & Ben