Product Roadmap
Intent is four products, one value stream. Each phase of the loop has its own maturity level, tooling, and investment needs. Intent v1.0 prescribes three independent layers: Compiled Knowledge Base → Transformation OS → Software Spec & Code. Each layer has its own maturity and roadmap. This page shows where each product stands and where energy goes next.
→
→
→
OBSERVE
Infrastructure Specced
Captures observations — patterns, friction, insights, hunches — from wherever practitioners work, and lands them as structured signals in .intent/signals/.
Signal schema & template
✓ Defined
MCP server (Claude Code, Cowork, Cursor)
✓ Built
GitHub Action (event emission)
✓ Built
Quickstart guide
✓ Written
GitHub native capture
Specced
AI tool plugins (ChatGPT, Copilot, Codex)
Specced
Signal clustering & pattern detection
Not started
Enhance
Signal deduplication — detect when the same observation is captured from multiple surfaces
Confidence scoring enrichment — automated scoring based on frequency, source diversity, recency
Build
Slack bot — reaction-based capture (:signal: emoji → PR with signal file)
Signal-to-intent promotion — when 3+ signals cluster, suggest creating an intent
Learn
Do practitioners capture signals in flow, or batch them later?
Which capture surfaces generate the highest-quality signals?
Transforms clustered signals into actionable specifications that AI agents can execute against. The shaping layer between "we noticed something" and "build this."
Work ontology (7 levels)
✓ Defined
Cross-functional shaping workflow
✓ Conceptual
CLI tools (intent-intent, intent-spec)
✓ Built
Spec validation tooling
Not started
Spec-to-agent handoff format
Not started
Enhance
Define "spec completeness" criteria — what makes a spec agent-ready vs. human-ready
Codify the shaping flow into a checklist, not just a diagram
Build
MCP tools: intent_create_spec, intent_validate_spec
Spec validation CLI — check completeness against defined criteria
Learn
What's the minimum spec an AI agent can execute against?
Does cross-functional shaping happen, or does the architect spec solo?
The phase where AI agents implement against specs. Intent is deliberately thin here — agents bring their own capabilities. Execute's job is to ensure specs flow to agents and execution events flow back to Observe.
Event schema for execution events
✓ Defined
Agent trace capture
Not started
Spec-to-agent handoff
Not started
Contract verification (pre/post)
Not started
Execution observability
Not started
Build
Agent trace adapter — lightweight hook to emit execution events to events.jsonl
Contract verification tool — check implementation against spec contracts
Entire.io integration — bridge agent reasoning captures into Intent's event stream
Learn
How much execution observability do teams want? Full traces or just start/stop?
Do agents need to read Intent specs natively, or is a human always translating?
OTel-native observability. Stack specced: Grafana Tempo + Mimir + Loki. Makes the Intent event stream visible — dashboards, digests, pattern detection, feedback loops. The learning layer that closes the loop back to Notice.
15 event types defined (event-catalog.md)
✓ Defined
OTel-compatible event schema (trace_id, span_id, parent_id)
✓ Defined
Observability stack architecture specced
✓ Specced
File Tail Adapter designed (events.jsonl → OTLP)
✓ Designed
Trace identity model (Intent = Trace, Spec = Span)
✓ Defined
Metrics model (counters, gauges, histograms)
✓ Designed
Grafana Observe dashboard specified
✓ Specced
GitHub Action event emission
✓ Built
OTel Collector deployed (Phase 1)
Not started
File Tail Adapter operational
Not started
Grafana dashboard live with real data
Not started
Signal clustering view
Not started
Weekly digest / report
Not started
See the observability architecture →
Enhance
Validate event schema against real data — are all 15 event types the right set?
Build
OTel Collector + File Tail Adapter — events.jsonl → OTLP pipeline (Observe, Now, Specced)
Grafana Dashboard — Tempo traces + Mimir metrics + Loki logs (Observe, Now, Specced)
Trace propagation — Intent = Trace, Spec = Span identity model (Observe, Now, Specced)
Agent instrumentation — bridge Execute events into Observe trace stream (Execute × Observe, Next, Designed)
Weekly digest generator — CLI or scheduled action that summarizes the event stream
Spec-to-outcome trace — show the full chain from signal through shipped capability
Learn
What do leaders actually want to see? Signal volume? Spec throughput? Cycle time?
Does the Observe product generate new signals, closing the loop?
The Knowledge Engine is a separate product with its own roadmap. It compiles raw sources into structured domain knowledge through the three-layer architecture: Compiled Knowledge Base → Transformation OS → Software Spec & Code. Each engagement validates federation, redaction, and compilation patterns before the next begins.
Engagement Rollout
1.Subaru — Most data, highest learning — reference implementation
2.F&G — Insurance domain — validates different industry
3.ASA — Healthcare — tests third domain
4.Cargill — Supply chain
5.Footlocker — Retail
Each engagement validates federation, redaction, and compilation patterns before the next begins. Subaru becomes the reference implementation — patterns discovered there become Core templates.
Investment Priority
Now
1 Validate Notice end-to-end — install MCP server on real repos, capture signals from real work
2 Enable GitHub Pages — flip the switch, make the site live
3 First real intents and specs — use the new CLI tools on actual work
Next
4 Intent dashboard v1 — HTML page that visualizes the event stream
5 Slack signal capture — reaction-based bot for team conversations
6 Spec validation — CLI tool that checks spec completeness
Later
7 AI tool plugins — ChatGPT, Copilot, Codex adapters
8 Agent trace integration — bridge Entire.io into the event stream
9 Signal intelligence — clustering, pattern detection, promotion suggestions
CLI Toolkit
Every Intent product has a CLI tool. Install them to your PATH or run from bin/ in the repo.
intent-signal
intent-signal "What you noticed"
Capture signals from the terminal. Writes to .intent/signals/ with frontmatter and emits signal.created event.
intent-intent
intent-intent "What needs to change"
Propose intents, link to signals, set priority and product area. Also: list, show, accept.
intent-spec
intent-spec "What to build" --intent INT-001
Create specs linked to parent intents. Also: list, show, approve.
intent-status
intent-status roadmap
Full system status: signal counts, intent pipeline, spec pipeline, event log, four-product maturity.
See the technology decisions behind the roadmap →