AI Development Methodology

P9 — Self-improvement velocity

Pillar goal: the methodology improves itself via the self-development cycle, faster than ad-hoc maintenance could — and the loop’s outputs compound across phases.

Last updated: 2026-08-19

Related:

• Brief: Capability layer 9
• Strategy phase: all four phases (primary throughout — this is the cross-cutting mechanism)
• Depends on: P1 Doc completeness, P3 Doc currency, P7 Community feedback loop, P8 Maintenance sustainability
• Feeds into: every other pillar (the loop’s outputs become methodology improvements that elevate every layer)
• Delivering epics: (none yet)

1. Overview

This is the compounding mechanism for the entire project. If the methodology gets better on its own through the self-development cycle, every other pillar benefits. If the cycle stalls, the methodology is static.

The cycle structure (per the self-development bootstrap plan and self-development/strategy/00_master_plan.md):

1. Apply the methodology to its own development.
2. Notice gaps, friction, missing patterns during loop runs.
3. Memory entries surface; clusters become methodology improvements.
4. Improvements ship as methodology releases.
5. Next loop run benefits from improved methodology.

This is the only pillar that’s primary in every phase of the strategy. Phase 1 bootstraps the cycle (Steps 0–4 of the self-development bootstrap); Phase 2+ exercises the cycle and harvests improvements.

The pillar’s existence is what makes the project’s vision sustainable: maintainer hours don’t scale linearly with adoption because the cycle absorbs recurring work into the methodology itself.

2. What this pillar covers

Cycle component	What “operational” means here
Self-development folder	self-development/ exists, is populated through Step 4 of the bootstrap, mirrors the methodology’s own PROJECT_STRUCTURE.md.
Autonomous loop	self-development/AUTONOMOUS_LOOP.md adapted from the template; constraints, stopping conditions, targets defined. Loop can run unsupervised between user check-ins.
Memory hygiene	The loop records insights as memory entries (or notes that become memory entries). Maintainer reviews memory periodically per the stdlib growth loop.
Methodology promotion	Clusters of memory entries become methodology improvements via the existing promotion path. Improvements ship as releases.
Cycle attribution	CHANGELOG entries explicitly note “this came from the self-development cycle” for shifts the loop produced.
Cycle health	Cycle operates within constraints (doesn’t damage abstract methodology docs autonomously, surfaces blockers to HUMAN_NEEDED.md, etc.).

3. Exit criteria

The pillar is delivered (evergreen) when:

• Steps 0–4 of the self-development bootstrap are shipped (v1.7.0 through v1.x with autonomous loop config).
• At least one autonomous loop run has completed end-to-end without maintainer intervention mid-run.
• The cycle has shipped methodology improvements (not just self-development cleanup); CHANGELOG entries attribute specific changes to cycle output. See brief/05_success_metrics.md “Early signals”.
• The cycle has surfaced patterns the solo maintainer wouldn’t have discovered alone — measured by the maintainer’s “this came from the cycle” notes in CHANGELOG. See brief/05_success_metrics.md “Sustained signals”.
• The cycle has not damaged the abstract methodology docs via the autonomous loop (no incidents where the loop touched methodology/*.md without explicit user direction).

Re-tested: every methodology release that came from the cycle (per the attribution criterion); quarterly during Phase 2 to verify cycle health.

4. Dependencies

Depends on:

• P1 — Doc completeness. The cycle can only improve what exists. A methodology with gaps too large can’t be improved by the cycle alone.
• P3 — Doc currency. The cycle works on the current methodology; stale docs send the cycle in the wrong direction.
• P7 — Community feedback loop. Cycle improvements are even better when informed by community feedback (Phase 2+).
• P8 — Maintenance sustainability. The cycle is what makes sustainability achievable; the two are mutually reinforcing.

Feeds into: every other pillar. Cycle outputs are methodology improvements that elevate every other capability layer.

5. Anti-patterns

• Treating the cycle as autonomous in everything. The cycle has constraints (no autonomous changes to abstract methodology docs; no third-party PRs; no production deploys). Letting the loop expand its scope produces incidents.
• Setting milestone targets the loop can’t reach. If the loop’s target is “achieve 500 stars,” the loop has no leverage — that’s adopter behavior. Targets must be actionable by the loop itself.
• Skipping the user check-in. The user is the gate at the milestone level. Cycle runs that bypass user review and ship directly to main bypass the methodology’s own DoD.
• Letting cycle outputs ship without cross-AI review. The methodology’s cross-AI validation is a gate; the cycle doesn’t skip it.
• Mistaking activity for improvement. A cycle that runs and runs but produces no real methodology improvements is overhead, not value. Measure outputs (releases shipped, patterns surfaced), not runtime.
• Ignoring the cycle’s cost in compute / tokens. The cycle runs on AI tools that cost money. Sustainable cost is part of P8 sustainability; not a separate concern but worth noting here.

6. Current state (last reviewed 2026-08-19; entries below date from v1.8.0 unless noted)

Bootstrap progress:

• ✓ Step 0 (brief) shipped in v1.7.0.
• ✓ Step 1 (strategy + pillars) shipping in v1.8.0 (this release).
• ✓ Step 2 (first epics) — shipped; nine epics chartered (E01–E09), seven closed, one parked, one active.
• ✓ Step 3 (first items) — shipped; item IDs now run to BL-0038.
• ✓ Step 4 (autonomous loop config) — shipped as self-development/AUTONOMOUS_LOOP.md, with the tiered-autonomy matrix.
• ✓ Step 5+ (loop runs) — at least one recorded run (loop-notes/2026-05-25.md). This list corrected 2026-08-19; it had described Steps 2–5 as pending since v1.8.0.

Cycle health (pre-operational):

• Constraints are spelled out in the bootstrap plan (loop must NOT modify abstract methodology docs autonomously, etc.).
• The stdlib growth loop and memory-as-leading-indicator patterns are in place in the abstract methodology to receive cycle outputs.
• Semi-annual self-evaluation cadence is the meta-check on whether the cycle is producing real improvements.

Honest: this pillar is the most ambitious and most speculative. The infrastructure is being built in v1.7.0 → v1.x; whether the cycle actually produces compounding improvements is testable only after Step 4 ships and the loop has run for some weeks. Until then, the pillar’s exit criteria are aspirational.

7. Delivering epics

(None yet. The bootstrap steps 2–4 are essentially this pillar’s first epic-like work — they belong as proper epics in Step 2 of the bootstrap.)

This site is open source. Improve this page.