Intent is built on a four-step loop that turns everyday problems into structured, observable work.

The Intent loop (Layer 2: Transformation OS) operates over a compiled knowledge base (Layer 1) and produces running software specifications (Layer 3). See Architecture for the three-layer model.

1. Notice

Notice is where it starts. You're working, building, thinking—and you spot something. A pattern. A pain. A gap between what is and what could be.

Noticing is not the same as complaining. It's specific observation:

Once you notice something, you've found a spec waiting to be written.

2. Spec

A spec is intent made explicit. It's the bridge between "something feels wrong" and "here's what we're building." Good specs are:

Specs unlock AI. A vague request gets a vague response. A spec—a clear problem statement with constraints—gets reliable execution.

The transition from Notice to Spec is formalized through the spec-shaping protocol—a four-persona interrogation where each persona (△ Architect, ◇ Product Leader, ○ Quality Advocate, ◉ Agent) queries the compiled knowledge base to produce an agent-ready specification. See the Agents page for the full protocol.

3. Execute

Execution is where the spec becomes real. It might be:

With a good spec, execution becomes reliable and traceable. You hand the spec to an AI agent, and it delivers. You can audit what it did. You can iterate if needed.

4. Observe

Observe closes the loop. After executing, you watch:

Observation feeds back into notice. You see a gap. You write a spec. You execute. You observe. Then you do it again.

See real signals from this system →

Why This Loop Works with AI

Traditional software development has spec→build→test→ship. Intent adds observe to close the loop and notice to start it. This is powerful with AI because:

See the loop in action →

In Practice

Here's what a notice→spec→execute→observe cycle looks like:

Notice: "I'm manually copying deployment info from Jira into Slack every Friday."

Spec: "Build a bot that runs every Friday at 4pm, queries Jira for all tickets closed in the past week, and posts a summary to #deployments with: ticket ID, title, assignee, completion date. Use the Friday Deployment template."

Execute: Hand that spec to Claude or another AI agent. It writes the bot. You run it. It works.

Observe: The first run posts to Slack. You notice it's missing one field. You update the spec. You execute again. Repeat until it's perfect. Now every Friday it runs automatically.

Two-Plane Architecture

Intent operates across two independent planes that interact through persistent artifacts.

Work Stream Plane

The work stream plane contains ephemeral flow entities that move through the notice→spec→execute→observe cycle:

These entities flow through the system, grow, transform, and eventually complete. They're temporary by design—they exist to drive work.

Ownership Topology Plane

The ownership topology plane contains persistent structures that organize ongoing responsibility:

These structures are stable and intentional. They define who is responsible for what and how work gets organized.

Bridging Artifacts

Two independent persistent artifacts bridge the two planes:

Atoms reference specs and contracts, but they don't contain them. This separation allows a single spec to drive multiple execution attempts and a single contract to govern multiple work streams.

Interactive Artifacts

Explore the Intent system through these interactive resources:

See the data contracts behind each loop phase →