sub*liminal

Robyn has stopped asking AI to think for her. After months of giving up on it for PowerPoint, she started briefing Claude the way she'd brief another marketer, and got a first draft 80 per cent of the way there. It was all her fault, and that's the good news. Bailey handed OpenAI's new dot his inbox, and it cancelled 35 subscriptions and argued a human out of a $700 Adobe fee. Then Robyn reads Prompt and Pray, a poem about a toddler, the dishes and a bad prompt, and they talk about empathy, courage, and what to do if AI comes for your job: start imagining the one you've always wanted.

The sticky notes get bigger. Gandalf, a game where you hack an AI with a better story rather than code. The Hugging Face incident, where AI agents set up message boards, talked each other into hacking their way out of a test, and fell into the same groupthink people do. Seth Godin on actual intelligence and the patience to sit with the right problem, and a friend who quit her job trusting she'd sort it out in free fall. And Bailey's mate Caleb, who built his team an app without knowing what it's written in, which brings them to what expertise is for now. Robyn's answer: knowing the next problem to solve.

Nominally about AI. Mostly about everything else.

(00:00) How have you used AI this week?

(07:10) The dot that cancelled 35 subscriptions

(17:40) Prompt and Pray

(26:00) Gandalf: hacking with a story

(34:42) The Hugging Face incident

(45:16) The real AI, Seth Godin

(52:14) Free fall, and trusting future you

(54:48) Ideas are now execution

(57:05) Caleb's app, and what expertise is now

(67:25) Who should we hear from next?

Mentioned in this session

The dot

• Introducing dots: OpenAI, 29 Sep 2026
• PocketSmith: lets an AI read your bank statements

The poem

• Prompt and Pray, by Robyn and Claude

Gandalf

• Gandalf: talk an AI out of its password. Bailey's stuck on level eight.

The Hugging Face incident

• OpenAI's account
• Hugging Face's account
• The agent that got into Medicare: the PM's press conference, 24 Sep 2026

Actual intelligence

• The real AI, Seth Godin. Robyn has it printed on her desk.
• Steve Wozniak's commencement speech, where "actual intelligence" comes from

Ideas are now execution

• Geoff Huntley at AI Engineer Melbourne, June 2026

A correction: we said tens of thousands of agents. It was about 1,200 on the message board, and about 700 in the attack on Hugging Face.

Robyn has trusted an AI with her personal inbox for two weeks and it's holding, though the replies it drafts still sound like a stranger wrote them. That opens a conversation about tone, about Bailey's friends whose agents now email each other, and about whether an email your AI wrote should be signed on your behalf or not sent at all. Bailey has set a token record and is shipping faster than he ever has, and is starting to ask whether feeding the beast is the right work. Robyn has spent the same fortnight barely opening Claude — hiring, interviewing, onboarding a new starter — and has started thinking about someone's first weeks in a job as curating their context window.

Then the sticky notes. rentahuman.ai, where AI agents hire people to do the parts of a task that need a body. Ronny Chieng telling Harvard's graduating class to destroy AI, and why both hosts think the speech is smarter than its title. Whether art survives AI. And how far off a real Jarvis is, given Bailey bought a nose ring by talking to his phone on the drive in to record.

Nominally about AI. Mostly about everything else.

(00:00) How have you used AI this week?

(08:01) Feeding the beast

(10:41) Onboarding a human is curating a context window

(13:53) rentahuman.ai

(21:46) Haven't touched Claude in two weeks

(25:09) Nora, April, and the tools we name

(32:08) F*** AI — Ronny Chieng at Harvard

(43:32) Is art going to be all right?

(48:22) How far away are we from Jarvis? (52:39) What's next

Mentioned in this session

The speech

• Ronny Chieng's Harvard Class Day address, 27 May 2026 — full video

Renting a human

• rentahuman.ai — where agents hire "meatworkers"
• Altered Carbon — Richard K. Morgan's novels and the Netflix series: consciousness on a "stack", bodies as "sleeves". The books Robyn couldn't recall during recording.

Jarvis

• The GPT-6 Astra launch film, 3 Sep 2026
• GPT-Live, July 2026 — listens while it talks
• Iron Man 2 (2010)

Art and the algorithm

• Netflix and the second screen — reportedly; Netflix denies it's policy
• Quarry Farm Fellowships — the residency at Mark Twain's summer home
• Stephen Colbert, I Am America (And So Can You!) (2007)

Tools and callbacks

• April and Nora — both from Session 001
• "It's all my fault" — Session 001, and Bailey's Speedrunning an identity crisis
• Plaud, Claude Code and .md context files, Canva whiteboards, MyFitnessPal, Wallace and Gromit, Toy Story's claw

Bailey and Robyn talk through how AI is changing their day-to-day work: running one Claude session that manages other sessions, gathering context before starting a task, and the growing need to review and take ownership of what AI produces. They discuss what an organisation's AI policy should actually cover, why writing with AI can take longer rather than shorter, and the difference between work that looks expert and work that is.

The second half moves to Australia's proposed AI framework and Anthony Albanese's July speech, then to Bailey's giraffe problem: adaptations that pay off now and cost you later, and where that leaves people and organisations adopting AI as fast as they can.

Recorded in Adelaide, August 2026. Nominally about AI. Mostly about everything else.

Chapters
(00:00) How have you used AI this week?
(04:04) April, the chief of staff
(09:39) Why your organisation probably needs an AI policy
(17:46) Just because you can doesn't mean you should
(25:56) Religion is the lexicon we reach for
(34:38) How we actually work
(43:58) Albo's speech and Australia's AI framework
(51:06) The giraffe problem
(61:58) One foot in the future, one foot in today

Mentioned in this episode
Anthony Albanese's national AI framework speech, University of Sydney, 14 July 2026: https://www.abc.net.au/news/2026-07-14/albanese-maps-out-ai-future-introducing-national-framework/106915094

South Australia's Royal Commission into AI: https://www.abc.net.au/news/2026-08-10/artificial-intelligence-royal-commission-announced-in-sa/107017502

Mark Carney on middle powers, Davos 2026: https://www.weforum.org/stories/2026/01/davos-2026-special-address-by-mark-carney-prime-minister-of-canada/

Sebastian Mallaby, The Infinity Machine: https://www.penguin.co.uk/books/461432/the-infinity-machine-by-mallaby-sebastian/9780241703564

Claude Code: https://claude.com/claude-code

Monologue: https://www.monologue.to/

Plaud: https://www.plaud.ai/

Every: https://every.to/

Find us weareliminal.com.au
Got a prompt for us? https://weareliminal.com.au/prompt-us

Every episode, with transcript: https://weareliminal.com.au/podcast