Comparison · development methodology · last balanced Apr 20, 2026
From autocomplete to autonomous — what the ledger says.
Five levels from autocomplete to autonomous agents, scored on criteria that matter. I name the level I operate at, the one most readers are actually at, and what moving a rung up costs.
Progression
Five levels from manual to autonomous.
Level 01
Manual
I write every line; no tool help.
AI autonomy · 0%
Level 02
AI‑Assisted
I drive; AI autocompletes at the line level.
AI autonomy · 20%
Level 03
AI‑Augmented
AI takes the first pass on a file; I review and integrate.
AI autonomy · 45%
Level 04
Agentic
AI plans across files and runs its own tests; I architect.
AI autonomy · 75%
Level 05
Autonomous
AI runs the loop end-to-end; I set the brief and spot-check.
AI autonomy · 95%
Level breakdown
The five rungs, in order.
- 01 · ManualAutonomy · 0%
What AI does
NoneWhat I do
Everything — architecture, implementation, testing. - 02 · AI‑AssistedAutonomy · 20%
What AI does
Autocomplete · small refactor suggestions.What I do
Drives every decision · reviews each line. - 03 · AI‑AugmentedAutonomy · 45%
What AI does
First-pass drafting · function-sized generations.What I do
Reviews · integrates · writes the tests. - 04 · AgenticAutonomy · 75%
What AI does
Plans · implements · runs tests · opens PRs.What I do
Architects · reviews · gates merges. - 05 · AutonomousAutonomy · 95%
What AI does
Owns the loop · handles routine ops.What I do
Sets brief · spot-checks outcomes.
Where are you?
Answer each honestly. The highest “yes” is your current level.
- L1I write all code by hand without AI tools in my editor.
- L2I accept Copilot-style inline suggestions line by line.
- L3I let AI draft a whole function, then I read, integrate, and test.
- L4AI agents plan across files, run tests, and open PRs I review.
- L5AI runs a full loop for a class of repeat work with minimal oversight.
Further reading
Keep reconciling.
Related to comparison: ai-assisted-vs-agentic-development
Verdict · audited 2026-04-20 · 5 criteria
Level 4 is where real client work actually ships.
Most of the client work I ship runs at Level 4 — agentic planning and testing under my architectural review. Level 5 is earned for repeat-shape work I run the same playbook on (the seven-site sprint, the daily build log). Level 3 is the honest resting place for early-week learning on a new stack. Levels 1–2 are where I go when the point is to think clearly about the code rather than emit it.
- Audited
- 2026-04-20
- Criteria
- 5
- Status
- on the books
- Voice
- first-person
- Shape
- spectrum
Next step