Large enterprises are issuing mandates that sound decisive — 4x your output — without building the infrastructure to know what output even means. The result is not productivity improvement. It is performance theater.
I am watching this play out in real time. Engineering managers are scrambling to quantify something they have never had good tools to measure, and now the stakes are higher because leadership attached a number to it. The dashboards they do have track proxies — prompts sent, lines generated, tab completions accepted, cost per developer. None of that tells you whether the work mattered. And when the only CLI a developer uses does not even report the same metrics as the IDE tools, you get engineers showing zeros across the board while spending two thousand dollars a month on agent compute. The data is fragmented and the picture is incomplete.
The deeper problem is that every measurement system is gameable. Story points inflate. Lines of code reward verbosity. PR counts reward splitting work into trivial commits. This was true before AI, and AI makes it worse because now the agent can generate volume that looks like productivity but might be net negative if nobody reviews it properly. Top performers who adopted these tools a year ago are already running parallel agent sessions and using git worktrees. Telling them to 4x on top of that is like telling your fastest runner to also carry the team's gear.
What actually matters is not raw throughput metrics but whether the work flowing through your pipeline is the highest-value work, executed well, and reviewed before it ships. That requires instrumentation at the agent orchestration layer — knowing which tasks were agent-assisted, how much human interaction each PR required, and whether merged code held up or created rework. If your agents run through a platform that logs the full session, you have that data. If your developers are just running local CLI tools with no telemetry, you are flying blind and calling it a strategy.
The enterprises that win this transition will not be the ones who mandate 4x. They will be the ones who build the measurement layer that makes the mandate meaningful.
Related Essays
The Gaming Problem Never Goes Away
Any developer performance metric can be gamed. AI tools just give us new things to measure — and new ways to get it wrong.
The AI Coding Tool Wrinkle
AI assistants are generating a granular log of every prompt, accept, reject, and iteration. That's new data — and a new way to measure the wrong thing.
Top Performer Analysis: The Real Opportunity in AI Tool Telemetry
The interesting use of AI coding tool data isn't ranking. It's understanding how your best engineers actually work — and helping the rest of the team catch up.
Key takeaways
- Mandating 4x output without measurement infrastructure turns productivity goals into political games.
- Top performers who already adopted AI tools cannot be measured against the same baseline as newcomers.
- The real bottleneck is not developer speed but the absence of instrumentation around agentic work.
FAQ
Why can't enterprises measure the 4x productivity gains they're demanding?
Most organizations lack tooling that captures agentic work patterns like parallel sessions, agent-assisted PRs, and background task orchestration. Traditional metrics like lines of code or PR counts do not reflect how AI-augmented development actually works.
Are story points still a valid measure of developer productivity with AI tools?
Story points can approximate business value throughput if the backlog is genuinely prioritized and teams do not inflate estimates. But they are easily gamed, and they do not capture the new dimension of how much of the work was agent-driven versus human-driven.