AI Pulse · Under the Radar

Two indie developers built the same tool this week without talking to each other. Mario Zechner and Dillon Mulroy both shipped extensions that let you annotate an AI coding agent’s last message and feed those notes straight back into the agent while it’s still working. Not after it finishes. Not in a pull request review. Right in the loop.

The pattern matters because it inverts the usual workflow. Most teams treat agent output like junior dev work: let it run, then review the pull request, then send it back for fixes. These tools skip that. You see the agent’s reasoning, you add a comment, the agent reads it and adjusts before it commits. It’s faster, and it keeps context warm. The fact that two people built it the same week, independently, suggests the need is real. It also lines up with a Hacker News thread from the same week arguing that software factories fail when humans only review finished output instead of staying in the loop.

This is the kind of thing we’re building into the Omni Command Centre, where review gates and feedback loops sit inside the workflow, not bolted on after. If you’re running agents that write code, marketing copy, or data pipelines, the question is whether you’re catching mistakes early or cleaning up messes late. The tooling is starting to assume you want the former.

The shift is quiet but practical. Agents work better when they get corrections before they compound errors. If you’re testing coding agents now, watch for tools that let you steer mid-task, not just approve or reject at the end. The original tweet shows one example. Expect more.