LinkedIn·Monday, 17 August 2026·9d ago

A colleague who's wrapping up writes a decent handover note. The same colleague, told to leave right now, writes a bad one. Same person.…

Betacowork

976 followers

A colleague who's wrapping up writes a decent handover note. The same colleague, told to leave right now, writes a bad one. Same person. Same knowledge. Very different note. Language models behave the same way, and that difference is worth a chunk of your usage limit. Here's the fix, and it's one instruction you give before you run out of room. You tell the model to write everything it knows about the project into a markdown file, framed explicitly as a handoff to its own future self. A common convention is to name it "handoff.md", but the name matters far less than the framing. When a platform force-compacts your conversation mid-task, the model is interrupted. It summarises under pressure, at a moment it didn't choose. Ask for the handoff deliberately, before the limit hits, and you get something noticeably better. Andrew's phrasing in the room was that the model is aware it's compacting, which is exactly what the forced version takes away. This is Part 2 of The AI Cost Nobody Explains. Part 1- why your limit drops while you're away - is linked in the comments. From our 14 July Local AI event at Betacowork in Brussels, with Toon Vanagt and Andrew Yourtchenko. 🔔 We run AI events in Brussels; the next one is in September. Follow us to hear when registration opens. 🔁 Repost if this saves someone a wasted afternoon. #AI #LocalAI #Brussels #Freelancing #Entrepreneurship #coworkingspacebrussels