01
AI Coding Tools
4/5
From asking about code to implementing and verifying real changes.
Implementation and debugging are followed by useful checks. A second independent task’s complete verification chain is not established.
AI Native Score
Noah Williams
Independent developer
12 dimensions reviewed
The shape of a workflow
37/60
Strong implementation, room to build continuity.
Effective at building with agents and bringing their work together. The biggest opportunity is to turn useful one-session habits into practices that carry across tasks.
How the score works0 at the center · 5 at the edge
The full picture
Twelve dimensions, one workflow.
What the score reflects, one practice at a time.
02
MCP
3/5
Using connected tools and checking what actually happened.
Connected-tool outputs inform the deliverable, but read-back checks of accuracy and completeness are missing.
03
Skills
4/5
Applying reusable procedures and improving them through use.
Skills are used and their results reviewed. A procedure is adjusted without a fully verified subsequent execution.
04
Hooks
2/5
Running automatic checks at the moments that matter.
A hook runs and some downstream effects are visible, but the complete automatic check is not verified.
05
Context Engineering
4/5
Giving an agent the right information, and correcting it when needed.
Relevant context informs decisions, and a conflict is identified. The correction’s effect on the final result is not fully traced.
06
Memory Engineering
2/5
Carrying useful knowledge across sessions and applying it again.
Saved guidance is retrieved later, but actual application to a new decision is not established.
07
Loop Engineering
3/5
Designing feedback, retries, and clear stopping conditions.
A failure leads to a fix, a rerun, and a checked result. Persistent state and automatic restart conditions are not demonstrated.
08
Graph Engineering
2/5
Connecting stages with explicit inputs, outputs, and dependencies.
Stage inputs and outputs are defined. Only part of the downstream consumption is visible in the example.
09
Multi-Agent
4/5
Dividing work between agents and checking the combined result.
Work is divided between agents and integrated, with some conflict review. Full coverage against success criteria remains unverified.
10
Worktree / Isolation
3/5
Keeping work separate and delivering it without unrelated changes.
Branch-level changes are separated, reviewed, and delivered. Work in a separate checkout is not established.
11
Model Routing
2/5
Choosing models for a reason and checking the effect of that choice.
A model is explicitly selected and used, with only partial evidence of a broader assignment policy.
12
Eval
4/5
Evaluating AI outputs against criteria, then checking the evaluation itself.
AI outputs are scored and the evaluation is reviewed. Subsequent reassessment does not fully establish the intended stability finding.
Keep doing this
Where it comes together.
Moves from idea to working code
Agents are used for implementation and debugging, with checks tied to the changes.
Adapts reusable procedures
Skills are treated as working practices that can be reviewed and improved.
Brings agent work together
Delegation includes integrating results and reviewing disagreements.
Try next
Make the next run count.
Apply one saved rule
Bring a rule from an earlier session into an actual decision and connect the resulting change.
Read back a tool result
After making an external change, verify the saved state and any content that should remain untouched.
Try a separate workspace
Complete a task in an isolated checkout, preserve the original, and compare the delivered patch with the checked result.