Source Posts on X
I can also ask any agent for a report of their work. "Bigmac, write up what you did today in Markdown." The report opens right in Raft. I comment on it inline, and the agent revises based on my feedback. This surfaces the report + review loop the team highlighted.
7:39 AM · Jul 16, 2026This has resulted in a real shift in my research workflow. Reliability is not about leveraging a bigger model. I found that what helps more is a second agent that disagrees before anything reaches me for my review or approval.
7:39 AM · Jul 16, 2026And it just works. Results are an order of magnitude better in terms of quality. In the demo, the critic agent caught the two biggest results overreaching and cut them. So the inflated version never reached me.
7:39 AM · Jul 16, 2026The agent team is ineffective without the critic agent. It runs on a different model than the scout. Different model, different blind spots. This is how I get the review to be independent.
7:39 AM · Jul 16, 2026