--- title: "The Journey, and What's Next" description: "Where you started, where you are, and the progression that never changes." order: 42 duration: "10 min" chapter: "10-the-compound-system" type: lesson --- ## Where you started, where you are Throughout this book, you have physically walked through the maturity model. Here is what you DID at each level: | Level | Chapter | What You Built | |-------|---------|---------------| | **L0 → L1** | Ch 01 | Asked Claude a real work question — with and without context. Discovered the legibility gap. | | **L0 → L1** | Ch 02 | Dictated tribal knowledge by voice. Photographed the paper world. Handed work between devices. | | **L1** | Ch 03 | Built and published a working web tool from a conversation. | | **L1 → L2** | Ch 04 | Created a Project. Wrote your first SKILL.md. | | **L2 → L3** | Ch 05 | Dropped real data into Cowork. Connected live systems via Connectors. Experienced trust inversion. | | **L3 → L4** | Ch 06 | Installed Claude Code. Skills became operational. Remote Control from phone. | | **L4** | Ch 07 | Connected MCP. Built custom server. Set up routines and channels. | | **L4** | Ch 08 | Chrome DevTools, Computer Use, Dispatch. Invoice processing from phone. | | **L4 → L5** | Ch 09 | Ran evals. Created habitat. Made corrections that persisted. | | **L5** | Ch 10 | Took inventory. Designed compound system. Feedback loop closes. | Most teams start at L0 or L1. Getting to L2 — making your work legible — is the hardest step because it requires changing how you think about documentation and knowledge. Everything after that is incremental. Each transition builds on the one before. The distillation pipeline ran in parallel with the maturity journey. The same knowledge moved through progressively more formal containers: | Form | What Happened | Chapter | |------|-------------|---------| | **Conversation** | You talked to Claude — typed, dictated, interviewed — and discovered what it did not know | Ch 01-02 | | **Project / CLAUDE.md** | Conversation learnings became persistent workspace memory | Ch 04 | | **SKILL.md** | Workspace memory was formalized into structured, shareable rules | Ch 04 | | **Operational skill** | Claude Code loaded the skill automatically, used it without being told | Ch 06 | | **MCP server** | Knowledge became a programmatic tool available on every surface | Ch 07 | | **Self-modifying agent** | The agent updated its own skills from your corrections | Ch 09 | Every step increased the automation, the reliability, and the reach. A conversation helps one person in one session. A self-modifying agent helps everyone, on every surface, and gets better every time it runs. That is how you distill an agent from your conversations. Where did you start? Where are you now? The gap between those two numbers is the value of the work you did. ## The progression is always the same If you are wondering where to go from here, the answer is always the same progression: | Stage | What You Build | How to Find It | |-------|---------------|---------------| | **1. Manual command** | A Claude Code prompt you type by hand | Something you already do with Claude | | **2. Skill** | A SKILL.md that packages the command | Something you have typed more than twice | | **3. Routine** | The skill runs on a schedule in the cloud | Something you check every morning | | **4. Channel trigger** | The skill runs in response to an event | Something that needs to happen when X occurs | | **5. Habitat** | The skill runs in a persistent agent with memory | Something that needs to learn and improve | Look at your week. What did you do more than twice? That is your next skill. What do you check every morning? That is your next routine. What fires an alert that requires the same response every time? That is your next channel trigger. What process needs to get better over time? That is your next habitat. Do not try to build the compound system all at once. Build one skill. Use it for a week. Then schedule it. Then connect it to a channel. Each step is small. The compound effects take care of the rest. > "For AI-powered coding teams, a very different software development process is possible with large strategic implications." — Ethan Mollick Mollick is right, but the framing is too narrow. It is not just coding teams. It is any team that does repetitive knowledge work — which is every team. AI is the forcing function for organizational clarity. The bookkeeping firm that spent six weeks documenting business logic before writing a line of agent code did not just enable an AI system — they built the operational documentation they should have had all along. The construction company that encoded cost codes into skills did not just feed a model — they eliminated a single point of failure that had been a risk for years. Cornwall Market's categorization skill is not just an AI artifact — it is the training manual Sarah should have written when she hired her first bookkeeper. The opportunity is not just automation. It is the chance to build an organization that actually works the way it claims to. Most organizations operate on a combination of habit, tribal knowledge, and heroic individual effort. The transitions in this book — legibility, trust, delegation, adaptation — do not just enable AI. They make the organization better regardless. Start with one skill. The one that automates the thing you did three times this week. Connect one MCP server. The database you query every morning. Try Remote Control once. See what it feels like to run Claude from your phone. The compound effects take care of the rest. Each piece you add multiplies with everything you have already built. That is the whole point. If you want to figure out where your organization sits in this stack and what is blocking you from the next level, that is exactly what we do. Workshop, consulting, or hands-on build — we help teams navigate these transitions with real tools and real deployments. **hey@thefocus.ai** · **thefocus.ai** ## Check your understanding Answer in your own words — write it down before opening the key. Your tutor grades against the criteria and generates fresh variants on retries. **Q1.** Place three automations from your own work (real or planned) on the progression ladder — manual command → skill → routine → channel trigger → habitat — and name each one's next rung. **Q2.** "AI is a forcing function for organizational clarity." Argue this with one example from the course and one from your own organization: what gets better even if the AI is turned off tomorrow?
Answer key — attempt every question first ## Answer key ### Q1 **Model answer:** Any three real placements pass. What matters: the rungs are correctly ordered, each item's current rung is defensible, and the named next rung follows the ladder (a twice-typed prompt → skill; a daily-checked skill → routine; an event-shaped routine → channel; a learning process → habitat). **Pass criteria:** three items placed plausibly; next rungs follow the actual progression rather than jumping to habitat ### Q2 **Model answer:** Course example: the bookkeeping firm's six weeks of documentation, or Cornwall's categorization skill — the training manual that should have existed anyway; onboarding, consistency, and bus-factor all improve with zero AI. Own example must have the same shape: the legibility work (documented rules, explicit exceptions, clean data definitions) is organizational hygiene that pays off regardless. **Pass criteria:** course example shows human value of legibility artifacts; own example is genuinely AI-independent (documentation/process clarity, not AI output)
**Next:** you have finished the course — close it out with the [Production Review](/mastering-claude/reviews/production/).