Ese Idukpaye

Learn. Build. Share.

[Ese Idukpaye @idnk2203]

9/
Here's the full map, and what belongs in that empty corner:

→ [https://t.co/1u4Lq9lgkk]

Six systems make your agent predictable before it ships anything. What checks on you?

[Ese Idukpaye @idnk2203]

Researchers at ETH Zurich tested whether AGENTS.md files actually help.

Across 138 tasks, they barely did. When AI wrote the file itself, results got slightly worse. And either way, they cost about 20% more to run.

[Ese Idukpaye @idnk2203]

8/
You can map any tool with two questions. Does it serve the agent's task, or your head across all of them? And does it enforce anything, or just hope you read it?

Three corners fill up fast. One stays empty.

[Ese Idukpaye @idnk2203]

Full write-up, the loop, the cost data, the sources: https://t.co/Hvu8NW10od

[Ese Idukpaye @idnk2203]

Every software engineer working with coding agents knows every morning you onboard a coworker who remembers nothing from the day before: my AI coding agent.

Then you onboard a second one: yourself, staring at 11 half-finished tabs, no clue which 3 are almost done.

[Ese Idukpaye @idnk2203]

So stop blaming your discipline for a context bug.

The fix isn't wanting it more. It's one honest, shared list you both run the loop on.

Wrote the whole thing up: the numbers, the sources, the loop. Link below 👇

[Ese Idukpaye @idnk2203]

I just completed the Graph Developer - Associate certification on Odyssey, the @ApolloGraphQL learning platform. https://t.co/qOSobeuj9B