Graham McNicoll

AI coding tools are genuinely impressive. And this is literally the worst they will ever be. They allow us to build products faster than ever. Engineers are now supervising agents instead of writing code. The volume of what ships to production has gone up significantly. Whether any of it actually moved a metric is still largely unknown. As great as these tools are, they haven't solved the fundamental problem of product development. Are we building the right things? Will this feature actually change how users behave? That is vibe coding without the safety net. You are moving faster. You just don't know where you're going. Wrapping your features in flags, even asking your agent to do it, and launching them as experiments closes that gap. You can measure the direct impact of every feature your agent ships. Turn it off if it's not working. Ship it fully once you know it is. AI can write the code. Knowing whether that code did anything worth shipping still requires you to measure it. That part is still on you. If you want to see how our MCP server fits into that workflow, 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.