Crossing the line

Key talking points

• Why November 2025 was the quiet turning point in Ai capability
• What “doesn’t need babysitting” actually means in practice for a small business
• Why most companies still aren’t getting value from Ai despite paying for it
• What’s coming next that small businesses should be ready for

Standout quote

“We just crossed a threshold where the code mostly works without you watching it.” — Simon Willison Verified

What this means for you

• If you tried Ai six months ago and gave up, the same task probably works now. Worth re-testing.
• Pick one task you stopped trying to automate, and try again this week.
• The gap between businesses using Ai and businesses ignoring it is widening fast.

Try this

Take one task you tried to get ChatGPT or Microsoft Copilot to do six months ago and gave up on. Run it again this week with the same instructions. Compare the output. If it’s noticeably better, that’s your signal to try Ai on something else.

Key talking points

• The plain-English definition of an Ai agent
• Why the shift from “chatbot Ai” to “doer Ai” matters this year
• The kinds of work agents are useful for, and the kinds they still aren’t ready for
• What to ask a vendor before you trust their “Ai agent” product

Standout quote

“We are now firmly in an agentic era — Ai is capable of doing things, not just talking about things.” — Jack Clark Verified

What this means for you

• Don’t buy products that promise Ai will do open-ended work for you. Not ready yet.
• Look for Ai that can finish one bounded task: book a meeting, draft a reply chain, prep a customer file.
• If the vendor can’t show a real demo of the agent finishing one bounded task, walk away.

Try this

Pick one workflow you do that has more than one step but fewer than five. Example: get an enquiry, check your calendar, draft a reply, send it. Write it down. When an Ai product crosses your desk claiming to be an “agent”, ask the vendor if it can do that exact workflow end-to-end.

Key talking points

• What the actual job data says (versus what the headlines say)
• Why a lot of “Ai apocalypse” content is marketing for Ai companies
• Skills that Ai cannot replicate, and why they’re worth more not less
• How to spot when a confident prediction from a tech CEO is hot air

Standout quote

“AI was not built for you. The rich don’t need you anymore.” — Scott Galloway Verified

What this means for you

• If your team is anxious about Ai replacing them, the data does not back the panic narrative.
• Lean into the human skills Ai cannot do well: judgment, taste, customer relationships.
• Don’t make hiring or layoff decisions based on Ai panic. Decide based on what your business actually needs.

Try this

Spend twenty minutes this week looking at the actual job listings in your industry on LinkedIn or Indeed. Compare to twelve months ago if you can. Use the real number to ground any Ai conversation in your team rather than relying on headline panic.

Key talking points

• What an “Ai reads everything at once” task actually looks like for a small business
• Why Anthropic isn’t releasing this Ai publicly (and what that tells you)
• Where this kind of Ai is reliable, and where it still needs human checks
• Why this is a more useful Ai use case for most businesses than chatbots

Standout quote

“We’re in a moment of danger because Ai is finding flaws faster than companies can fix them.” — Casey Newton, paraphrasing Anthropic’s framing Paraphrased

What this means for you

• Think about what you’d love to read but never have time for. That’s where this Ai is most useful.
• Customer support messages, invoices, supplier contracts, complaint forms — all good candidates.
• Try one this month. Find one pattern in your business you couldn’t see before.

Try this

Take 100 of your recent customer support messages, or 50 of your last invoices. Drop them into ChatGPT or Claude in one go and ask it to spot patterns you’d miss reading them one at a time.

Here are [N] customer support messages from the last [period]. Read them as a batch. Group them by theme. Tell me the three patterns I would not notice reading them one at a time. Quote one example for each pattern.

Key talking points

• How Notion teams structure work when Ai does the first draft of everything
• What changes when humans manage ‘swarms’ of Ai agents instead of doing the work
• How to make sense of which docs Ai needs access to (and which it shouldn’t)
• Where the agent metaphor breaks down and humans still own the call

Standout quote

“You stop writing first drafts. You start writing instructions and editing.” — Simon Last Paraphrased

What this means for you

• Try “instruct + edit” as the new shape of one task this week.
• If your team relies on first-draft writing time, that time is moving from minutes to seconds.
• Workflows built around “we’ll write it ourselves” will get unbundled.

Try this

Pick one document you’d normally draft from scratch this week — a proposal, a policy, a piece of customer comms. Instead of writing, write 6 to 10 bullet instructions and let an Ai tool produce the first draft. Then edit. Time both halves.