Directory
Techniques
148 of 163 stories map to at least one of these techniques. A story can use more than one, so it may appear on more than one page. Counts show how common a technique is in this collection, not whether it works.
Context
93 stories use at least one technique in this stage.
-
Rules, skills and context files
The agent is steered with written instructions and context: AGENTS.md/CLAUDE.md, skill files, rule docs, migration guides, golden examples, indexed code or telemetry.
-
Spec or plan agreed first
A written spec, plan, design doc or brief is produced (often by the agent from old code) and settled or approved by people before implementation starts.
Doing the work
94 stories use at least one technique in this stage.
-
Small, bounded batches
Work is cut into small PRs, components, phases or batches with capped size or volume, instead of one big-bang change.
-
Parallel agents on isolated copies
Several agents or sessions work at once, each in its own worktree, container, clone or branch, coordinated through git or tickets.
-
Orchestrator and worker agents
A lead, planner or orchestrator agent splits the work and hands pieces to worker or specialist agents, sometimes with a judge.
-
Deterministic tools plus LLM
Classic tooling (codemods, code search, static analysis, build graphs, scripted pipelines) does the mechanical or discovery part and the LLM handles bounded steps.
-
Background agents on in-house infrastructure
Agents run in cloud sandboxes or on internal platforms, triggered from tickets, Slack, CI or schedules rather than a developer's laptop.
Verification
83 stories use at least one technique in this stage.
-
Automated verification
The agent's work is checked automatically: tests, compilers, linters, CI gates or fix loops, including parity harnesses, coverage or eval sets built before the change.
Review and merge
69 stories use at least one technique in this stage.
-
Human review
A person reviews or approves the change (or the plan) before merge, or the owning team makes the final call.
-
Agent review
A separate agent or a different model reviews, audits or tries to break each change.
-
Risk-tiered routing
Rules that route each change by risk: low-risk changes auto-merge or get agent approval, risky areas go to a human.
Release
7 stories use at least one technique in this stage.
-
Staged rollout of shipped code
The shipped code goes out gradually: feature flag, canary with rollback, pre-release, closed testing or a rehearsed data migration.
Organisation
17 stories use at least one technique in this stage.
-
Enablement, spend controls and tool pilots
Company-level moves: preinstalled or self-hosted setups, peer and champion adoption, budgets and token policies, and pilots or proofs of concept of the tool before wider rollout.