The Five Levels of AI-Assisted Development
Most teams think they're further along than they are.
Dan Shapiro's framework of five levels of AI-assisted development, borrowed from the autonomy levels of self-driving cars, is the most useful way to think about where your team actually is. The insight is that every level feels like you are done but you are not.
Level 0: Manual driving
The AI does autocomplete. You write every line. The bottleneck is typing speed. This was most of "AI coding" in 2022. Feels primitive now.
Level 1: Lane assist and cruise control
The AI handles discrete tasks. You direct what to do. The bottleneck is how fast you can assign tasks. Copilot in 2023. Useful, but you're still doing most of the driving.
Level 2: The trap
You pair with the AI like a colleague. It feels transformative. Productivity is up, morale is high, the work is satisfying.
This is where most teams are. And it's a trap. Shapiro says, every level feels like you are done, but you are not. You are still the rate-limiting factor. If you're not at the keyboard, nothing happens.
Level 3: Waymo with a safety driver
The agent works autonomously, but a human is still responsible and can take over. You are supervising, not driving. You spend the day reviewing code. So much code. Your work is diffs.
For many people this feels worse than Level 2. Shapiro is blunt that almost everyone stops here.
Level 4: Robotaxi
You are not a developer, and you are not a development manager. You are the PM. You write the spec, argue with it, plan the work, and review the plans. Then you leave for a few hours and check whether the tests pass.
Shapiro says he works here. The driver is gone. The person is not. Getting to Level 4 safely means mastering Level 3 first.
Level 5: Dark factory
The car is gone too. The process is a black box that turns specs into software. Humans are neither needed nor welcome. Shapiro knows a few teams under five people who do this. Interesting, but not where most teams are, and not where most teams should be aiming today.
Why the levels matter
The levels aren't a ladder you want to climb for its own sake. They're a diagnostic. Most teams think they're at Level 3 or 4 when they're actually at Level 2, because Level 2 feels like the destination.
Shapiro says almost everyone stops at Level 3, and his own description shows why: your work becomes diffs. The harness changes what the human reviews. You stop reading every diff. You approve the design before the code, and the evidence before the release. That's the Level 4 shape. That's not slowing the agent down. It's what makes going faster safe.
The workflow is free. Bring any agent. Keep your standards. github.com/oystro/oystro-oss
Bring Harness Engineering to Your Project
Oystro adds deterministic human approval gates and persistent memory to AI coding agents, so your codebase stays reviewable and maintainable for years to come.