Startup

ℹ Illustrative scenario — a composite archetype, not a real named adopter.

Meet Alex Rivera

Alex is CTO of a 12-person AI startup building a developer tool. The team ships weekly, context switches constantly between features, and relies heavily on AI agents for code generation, documentation, and planning. Institutional knowledge lives in people’s heads and disappears when they go on vacation.

The Challenge

The startup moves fast — too fast for knowledge to keep up. Feature decisions are made in Slack and forgotten. Architecture context lives in the heads of 3 engineers who’ve been there since day one. New hires (arriving monthly) can’t find answers to basic questions: “Why did we choose this database?” “Where’s the auth flow documented?” “What’s the priority right now?” AI agents generating code lack project context and produce boilerplate that misses the codebase’s conventions.

How aDNA Helps

Alex adds aDNA as the team’s operational knowledge layer. The vault doesn’t replace the codebase — it sits alongside it as the brain the codebase doesn’t have.

Decisions get captured (what/decisions/): Every architectural choice gets a lightweight ADR. “Why Postgres over Mongo?” has a file. When someone asks, the answer is a link, not a 20-minute retelling.

Agents get context (what/context/): A topic on the product’s architecture gives AI agents the conventions they need: naming patterns, API structure, error handling approach. Agent output goes from generic to codebase-aware.

Priorities stay visible (STATE.md): Updated weekly with current sprint focus, blockers, and shipping timeline. Any agent or new team member reads STATE.md and knows what matters right now.

What Their Vault Looks Like

devtool.aDNA/
├── what/
│   ├── context/         # Product architecture, API conventions, tech stack
│   ├── decisions/       # 40+ ADRs (architecture decision records)
│   └── lattices/        # CI/CD pipeline, deployment workflow
├── how/
│   ├── campaigns/       # Q2 goals, major feature epics
│   ├── missions/        # Sprint-sized deliverables
│   ├── sessions/        # Agent work audit trail
│   ├── templates/       # ADR template, feature spec template
│   └── backlog/         # Ideas and improvements
└── who/
    ├── governance/      # Coding standards, review policies
    ├── team/            # 12 engineers with roles
    └── coordination/    # Cross-team handoff notes

Outcome

Onboarding drops from 3 weeks to 3 days. New engineers read the context library and decision records instead of shadowing senior devs. AI agents produce codebase-aware output because they load architecture context before generating. The team ships the same speed but with less context loss.

Typical Ontology Extensions

Entity	Triad	Purpose
`api_spec`	what/	API endpoint documentation with versioning
`onboarding_checklist`	how/	New hire setup and orientation tasks
`customer_feedback`	who/	Structured customer feedback tracking

The first hour

A 60-minute quickstart for a CTO who has one hour and no appetite for ceremony.

Minutes 0–10 · Clone and install — fork the base template into your repo root. You get the triad (what/, how/, who/), the 16-entity base ontology, and a working CLAUDE.md scaffold. Start at Navigate a Vault.

Minutes 10–25 · CLAUDE.md — edit the root CLAUDE.md to name your project, voice, and standing orders. This is the first file every agent session reads. See Create Your First CLAUDE.md.

Minutes 25–40 · STATE.md — write a one-paragraph operational snapshot: current phase, blockers, next steps. Every new hire and every new agent session reads this second. See Governance Files.

Minutes 40–60 · First ADR — capture one decision that currently lives in a Slack thread as what/decisions/adr_001_<slug>.md. Future agents and future hires read this instead of re-asking. See Knowledge Graph.

What you should have at the end of the hour

The vault is legible to an agent. CLAUDE.md + STATE.md means a fresh agent session reads two files and knows where everything is. No per-session re-explanation.
Decisions stop evaporating. One ADR seeds a decision log. Within a month the founding-engineer bus factor starts breaking down — the “why” lives on disk.
Onboarding has an entry point. Your next hire reads STATE.md and walks the ADR list. No Slack archaeology, no “ping me if you have questions.”

Where to go next

Design a Mission — hour one is individual discipline; hour two is team coordination. When you have a sprint-sized goal, a mission file turns it into claimable objectives.
Run a Campaign — structuring quarterly goals as phased, gated work.
Base + Extension — growing the vault organically without forking the standard.

Token Selection — how agents load the right context for each task
Convergence Model — how campaigns and missions focus weekly sprints
aDNA vs. Notion — why the startup chose aDNA alongside Notion
Mission — the unit that turns a sprint goal into claimable objectives