AI Adoption · Engineering Playbook
Internal playbook · May 2026

AI Adoption for Software Engineering

A working reference for a consulting practice helping software engineering organizations adopt AI tools and practices. Not marketing — reading-first, organized for retrieval.

How this is organized. Catalog lists every area we audit and the solutions we propose. Synthesis is the top-level consulting position. Patterns, People, Harnesses, and Production are the supporting depth.

01 Start here

Catalog of areas & solutions

Every area inside a software org we can audit, with concrete agent loops, harness/skill recipes, and platform work.

7 sections

Synthesis

Five anchor ideas, expected-outcome table, house principles, risk frame, and the team's reading list.

Top-level position

Patterns & frameworks

Building Effective Agents, context engineering, context rot, harness engineering, plan mode, evals, MCP.

15 patterns · diagrams

Influential practitioners

Karpathy, Willison, Ronacher, Hashimoto, Cherny, Litt, Yegge, Husain, Yan, Martin, Breunig, Walden Yan.

13 figures

Harnesses & tooling

Filterable landscape of coding agents: Claude Code, Cursor, Codex, Amp, Devin, Aider, OpenHands and more.

Interactive filter

Production deployments

What real orgs ship — code review, issue→PR loops, on-call agents, MCP — and the security incident dossier.

+ incident dossier

02 The five anchor ideas

01 · MINDSET Vibe → Agentic Karpathy 02 · CRAFT Context engineering Martin · Breunig 03 · PLATFORM Harness > model Hashimoto 04 · DESIGN Workflow before agent Anthropic · Cognition 05 · RIGOR Evals first Husain Sequence of investment for any client engagement →
Five ideas, in the order they show up in client engagements. Mindset shift opens the door; context + harness + design carry the work; evals make it durable.
  1. From vibe coding to agentic engineering. Discipline replaces vibes; new craft is guiding agents. → Synthesis
  2. Context engineering is the new prompt engineering. Most agent failures are context failures. → Patterns
  3. The harness matters more than the model. Same model + better scaffolding moves SWE-bench Pro 5–15 points. → Patterns
  4. Don't build agents for everything. Don't build multi-agent unless work is genuinely parallel. → Patterns
  5. Evals first, autonomy second. Most failed LLM products share one root cause — no evals. → Patterns

03 Scope

We focus on software development organizations: engineering, platform, DX, SRE, and the developer-facing surfaces of support and internal knowledge. Other departments (sales, marketing, finance, HR, legal, data) are explicitly out of scope for this playbook — depth over breadth.

04 Source documents

Each HTML page mirrors a markdown source in the repo with the same citations. Verify pricing/feature claims before quoting — this field moves monthly.