AI Week. A weekly two-host round-up of what actually happened in AI — across Anthropic, OpenAI, Google, Microsoft, Grok and the wider field. Plain talk, sources cited. Listen online or take the file.
Canary Co: The Full Picture
Series finale. Mara and Theo step back and look at what 48,000 people, nine divisions, three deployed agents, and one bearing failure taught them about making AI work in a large, ordinary organisation. The question from Episode One is still the right one. The work is harder than the pitch deck says. It's worth doing carefully.
Audio is being produced — it lands with the next run.
Role-Aware Agents
A generic AI assistant treats a nurse and a financial analyst the same. A role-aware agent knows the O*NET task profile of the person it's supporting — and that changes everything. Theo walks through Canary Co's shift-handover tool: why it works, how it's scoped, and the one check he runs before deploying any agent.
Audio is being produced — it lands with the next run.
ESCO as a Knowledge Graph
ESCO is machine-readable RDF — a knowledge graph of 3,000 occupations and 13,800 skills. Theo explains how Canary Co replaced a pile of inconsistent HR documents with ESCO as the authoritative source for their HR FAQ agent, and what improved when they did.
Audio is being produced — it lands with the next run.
The O*NET API
The O*NET API is free, publicly documented, and almost nobody building AI workforce tools is using it. Theo walks through the key calls — occupation summaries, task lists, hot technologies — and shows how Canary Co uses it to build role-aware Copilot quick-starts automatically.
Audio is being produced — it lands with the next run.
The New Role Profile
When AI handles 30% of what a role does today, what does the role become? Not smaller — oriented differently. Mara uses ESCO's emerging skills taxonomy and the Canary Co logistics analyst case to sketch what role redesign actually looks like in practice.
Audio is being produced — it lands with the next run.
Writing Prompts From Task Descriptions
Most people write vague prompts because they're vague about the task. O*NET forces precision. The technique: use the task description as your prompt foundation. Mara and Theo show it working across three roles — Financial Analyst, Social Worker, Logistics Analyst — and explain how Canary Co packaged this into a role-by-role quick-start.
Audio is being produced — it lands with the next run.
Task by Task: The HR Specialist
O*NET lists 12 tasks for an HR Specialist. Mara and Theo go through each one with Copilot open — honest about what works, what's flat, and what still needs a person. The case notes that write themselves. The policy advice that sounds confident and isn't. The training draft that actually saves an afternoon.
Audio is being produced — it lands with the next run.
ESCO: The European View
The US has O*NET. Europe has ESCO — 3,000 occupations, 13,800 skills, machine-readable and free. Theo leads this one: what ESCO actually contains, why the transversal skills taxonomy is the part AI finds hardest, and three practical uses for it in AI adoption work.
Audio is being produced — it lands with the next run.
Reading the Risk
The famous '47% of jobs at risk' number is both true and badly misunderstood. Mara and Theo go task-level — walking through what exposure analysis actually looks like, how Canary Co used it on their Finance team, and what belongs in an honest exec briefing about AI and jobs.
Audio is being produced — it lands with the next run.
The Map of Work
Before you decide which jobs AI will affect, you need to know what those jobs actually contain. Mara and Theo introduce O*NET and ESCO — two free, structured, public databases that map what people do at work at the task level. This is where every honest AI and jobs conversation should start.