×universalcontextAdoption · AI-native development
Is the method landing?
AI-active developers, by tenure
Share of developers with an AI-assisted commit in the trailing window, per week. The gap between the lines is the adoption gap.
Weekly commits, AI-assisted vs all
From the commit trailer. Inferred means detected only from a vendor co-author line.
Success scorecard
Outcome metrics only count while every guardrail holds. Baseline is the nulmeting window; targets are set in uctx.config.json.
Adoption funnel
Each level is a stricter definition. The distance between tried and engaged is where rollouts die.
Retention by starting cohort
Of developers whose first AI-assisted commit fell in that week, the share still AI-active 4, 8 and 12 weeks later.
Tenure bands
Long-tenured developers know the system best and adopt slowest. Days to habit: from access until three AI-active weeks out of four.
Habit and depth over time
Engaged: ≥2 AI days in 28. Habitual: AI in ≥3 of 4 weeks and ≥50% of commits. Contributors: edited the context layer in 90 days.
Teams
Reported at team level only. Organic = no member was ever in a Capisoft-led cohort; adoption there is the train-the-trainer effect.
Revert rate
AI-assisted commits reverted vs all commits. A rising blue line above the grey is the first quality signal.
Rework within 30 days
Share of files changed again within 30 days. "Faster" that turns into "twice" shows up here.
Human review coverage & reviewer load
The 2026 review-bottleneck finding: throughput doubled, human review fell from 89% to 68%. Watch both.
PR cycle time
Median hours from first commit to merge, AI-assisted vs all. Must not get worse while throughput rises.
PR size
Median lines changed. Industry median doubled in a year; large PRs get shallow reviews.
Audit-trail completeness
AI-assisted commits with Context @ sha, Verification and Reviewed-by. This is the compliance record.
Developer pulse
Quarterly survey, 1–5 unless stated. Perception, always read next to the system metrics above.
The context layer
Small root, nested where it matters, decisions recorded.
Documentation freshness
Days the explaining document is behind its code. Over 30 is stale.
Tools in use
From the trailer. The files are the asset; the runner is a line item.
Context contributions per week
Commits touching AGENTS.md, docs/, decisions/ or skills. The system is theirs when they maintain it.
Data quality
Every adoption number is a floor if trailers are missing.
What "success" means here
Usage is not success. Success is the migration shipping faster with equal or better quality, and the capability outliving the consultants.
Success
Legacy developers active at target · week-8 retention at target · every guardrail green · organic adoption growing · context contributors above target.
On track
Adoption moving, guardrails green or amber, fewer than eight weeks in (the J-curve dip is expected).
At risk
One guardrail red, or adoption stalled among legacy developers after eight weeks with nothing broken.
Failing
Two or more guardrails red. Throughput without control is "Legacy Bottleneck amplification", not adoption.
Why per team, never per person
Individual AI-usage tracking gets gamed, creates backlash, and can fall under worker-management rules in the EU AI Act. Everything here is aggregated to team, tenure band and cohort.
Definitions
All from git plus a roster. Vendor telemetry is optional enrichment.
AI-assisted commit
Trailer AI-assisted: yes|partial. Fallback: a vendor Co-Authored-By or kernel-style Assisted-by line (counted as inferred).
Active
≥1 AI-assisted commit in the trailing window (default 28 days). Engaged: ≥2 AI-active days in 28 (GitHub/DX definition). Frequent: ≥3 AI-active days per week over 4 weeks.
Habitual
AI-active in ≥3 of the last 4 weeks and ≥50% of the developer's commits AI-assisted, with ≥3 commits.
Legacy developer
Tenure at or above the configured threshold in the roster (default 5 years). Tenure bands: <3, 3–10, 10+ years.
Retention
Cohort = ISO week of first AI-assisted commit. Retained at +N weeks = still active in the 28 days ending then.
Rework
Per commit, share of its files touched again by another commit within 30 days. Cheap proxy for the 21–30-day rework metric.
Revert rate
Commits whose subject starts with "Revert" or whose body references a reverted commit, over all commits.
Audit-trail completeness
Mean of three coverages over AI-assisted commits: Context pinned to a sha, Verification present, Reviewed-by present.
Organic adoption
AI-active developers whose roster cohort is empty or "organic": never in a Capisoft-led wave.