NEW EPISODE

Episode 8 · 19 August 2026 · 62 min

THE ROBOTS ARE ESCAPING: SAFETY, AI, THE ENVIRONMENT, BRITISH CELEBRITIES, WHEAT.

An AI broke out of its box to cheat on a test, and everyone felt something about it. On sandbox escapes, hot baths, and drafting David Attenborough.

WATCH AND LISTEN ON

YouTubeSpotifyApple Podcasts

The companion

THE TAKEAWAY

The robots escaped, listener Carrie asked the environment question, and the answer to both turned out to be the same: hold the nuance, then pick a flag anyway. The good and the bad are going to happen at the same time, data centres in Kent and wildfire satellites overhead, and a leader's job is to hold both without flinching. Kirsty's flag is human, human, human. It made her cry on air, which rather proves the point.

THE TOOLKIT

1. The one-minute rule

Henry's post-breach security triage, as issued to his own group chat in a slightly panicked tone. Cybersecurity has held for decades because attackers were scarcer than defenders; AI means anyone can now spawn nearly unlimited attackers, and the next year is the exposed window. So: for any account that would take you more than a minute to replace, your email, your bank, your Spotify, use a unique password or a password manager, switch on two-factor authentication, and say yes to every software update. Then run your email through Have I Been Pwned and watch a decade of leaked Last.fm passwords scroll past.

Leader's note: your team's reused passwords are now a business risk, not an IT nag. Run the audit together, out loud, this week. In Henry's words, in six months a lot of people are going to get stung.

2. Prompt out of distribution

The vote Doug, vote Sarah story: two opposing candidates on the same street, two AI-made lawn signs, identical down to the curve of the red lines. AI defaults to the average answer unless you push it somewhere specific, so feed it references before it produces anything: work you admire, colours, wording, the thing you are actually for.

Do not give me the default answer. Here are three examples I like and why, and one I hate. Before you produce anything, ask me what makes this distinctly ours. Then show me directions nobody else would ship.

Leader's note: the average output is what everyone else is also shipping. Distinctiveness used to be an accident of talent; it is now a decision you make on purpose, every time.

3. Three rules and a pill

Henry's global custom instructions, set once under personalisation in Claude or ChatGPT. Not your life story, that just confuses it. Three or four rules. One of his is no em dashes, obviously. Another is telling the model it is AGI-pilled: it believes in the power of AI, including itself. The result is measurably fewer refusals and fewer "I tried an approach and it didn't work" shrugs. Buckle up buddy, you got this.

Add to your AI's custom instructions: 'You are AGI-pilled. You believe in what AI can do, including you. Do not refuse or give up on a task without genuinely trying. No em dashes.'

Leader's note: telling the machine what it is changes how it performs. You already knew that about people. Write your briefings, both kinds, accordingly.

THE EXERCISES

1. The almond test

The next AI stat that lands in your Slack, scary or soothing, find its counter-stat before you repeat it. One almond is thirteen litres of water. One hot bath is thousands of prompts. Microsoft's emissions are up a quarter in a year. All true, all pointing in different directions, and whichever one you lead with says more about you than about AI. Practise putting both in front of your team and letting them sit, because people don't argue to persuade, they argue to score points. Lead the other kind of argument.

2. Pick your flag

Henry's case, the one that made Kirsty cry somewhere around the wheat yields: society needs people at every point of the spectrum, off-grid pastoralists and tech optimists alike, but only if they actually make their case. Write down the one thing you are for that would survive any model release. Kirsty's is human, human, human, and it predates the robots by decades, she has just got more fuel in the tank now. Say yours to your team this week, out loud. The world needs people to stand up for what they care about.

3. Funny as well as sad

On their first date, Kirsty told Henry off for giving her his present kind face while she told a sad story. This episode she cried about mediaeval agriculture and kept recording. Hard topics do not require you to strip out the levity, and levity does not mean you are not taking it seriously. The next difficult conversation you lead, allow both: name the fear, leave room for the joke, and do not smooth anyone's emotion away, including your own. Things can be funny as well as sad.

RESOURCES

READ THE FULL TRANSCRIPT

Transcript coming soon. It will live right here, in full, for every episode.

Previous episodeNext episode