Graham McNicoll

Most Engineers I know are supervising agents now, not writing the code. Features that took weeks take hours. The volume of what ships to production has gone up significantly. The fundamental question of product development remains unsolved: are we building the right things, and did any of it move a metric that matters? Those questions have not gotten easier. In some ways, they have gotten harder because the volume of code reaching production has increased without a corresponding increase in the ability to measure what any of it does. That is vibe coding without the safety net. Moving fast with no visibility into whether the direction is right. It is like trusting your codebase to a brand new employee you have never met and hoping they do the right thing. Wrap every feature your agent ships in a flag and run it as a controlled experiment. Measure what it actually did before committing to a full rollout. Speed and measurement work together. That combination is what separates thrashing from a real learning rate. AI writes the code. Knowing whether the code did anything worth shipping still requires you to measure it. That part has not changed. If you want to see how we wire this up with our MCP server, send me a DM.

Only one snapshot so far — the engagement-over-time curve appears once the daily scrape has captured this post at least twice.