Atlas — intelligence & communication layer: product research & go/no-go
Concept under test (Robert, 2026-07-18): “Atlas is the intelligence and communication layer for large-scale legacy refactoring. It explains what is changing, documents legacy systems and the new monorepo, maintains a shared glossary, and keeps teams updated on progress, decisions, and migration context. At the same time, it protects scarce human attention by capturing questions from agents, merging duplicates, rate-limiting noise, and routing each issue to the right owner. One human answer then unblocks every agent who asked, resumes their work, and is written back into the documentation and skills so the same question never needs to be answered twice.”
Goal: Decide whether Atlas — the expansion of factory app #3 (question triage, plan 001 verdict BUILD-ON-ZARUBA) into a full intelligence/communication layer — survives critical product scrutiny: build / build MVP / build on open source / integrate / partner / process-only / pivot / abandon. Do not assume the idea is good.
Scope note: Internal product — primary users are Robert’s fleet agents (API), Robert (operator/router), Tomas Zaruba (scarce answerer + digest consumer); “teams” is a future audience that does not exist yet. Standard product-research sections that only apply to external commercial products (pricing funnels, purchase, renewal/churn economics, marketing sites) are deliberately compressed or marked N/A rather than fabricated.
Read order
| # | Doc | pm workflow | What it does |
|---|---|---|---|
| 00 | SYNTHESIS | — | Executive recommendation — the decision (fulfills executive-recommendation.md) |
| 01 | Initial recommendation | — | Pre-research expert take: what Atlas should be, Ive/Norman/Rams lenses |
| 02 | Research | pm:brainstorm + pm:discovery | Problem discovery, evidence tiers, advisory board (Jobs/Ive/Cagan/Torres/Christensen), steelman both ways |
| 03 | Journey | — | Stakeholders + adapted internal journey maps |
| 04 | User stories | — | Stories by persona (agents are personas too), MVP classification |
| 05 | Competitive analysis | — | Comms/status products, AI doc tools, Q&A/triage landscape, “do nothing” |
| 06 | Open-source analysis | — | Foundation candidates with license/health/verdicts |
| 07 | UX / UI | — | API-as-UX spec, question/answer form design, anti-portal decision |
| 08 | Made to stick | — | SUCCESs pass for the humans who must adopt it |
| 09 | North star | — | Strategy, North Star Metric + alternatives, guardrails |
| 10 | Architecture | — | Data model, substrate integration, write-back pipeline, diagrams |
| 11 | Build vs buy | — | 8 sourcing options scored, one recommendation |
| 12 | PRD | pm:write-prd | Requirements, MVP scope, acceptance criteria, kill criteria |
| 13 | Pre-mortem | pm:pre-mortem | 18-month failure modes with triggers and owners |
| 14 | Validation plan | — | Riskiest assumptions → cheapest disproving tests |
| — | evidence/ |
— | Raw research dumps (web sweeps, internal ground-truth reads) |
File mapping vs the research brief: research.md → 02-research.md; executive-recommendation.md → 00-SYNTHESIS.md (anchor doc per rules/tools/plans.md); all other briefed files keep their names with NN- prefixes.
Out of scope
- Work-ledger, fleet monitor, dual-run harness (siblings — plans 002-*; plan 001 verdicts stand)
- Domain migration decisions (Tomas’s lane)
- Infra bring-up (billing gate,
monorepo-infra) - Legacy-audit knowledge-graph extractor (001 scoped it to the audit; referenced, not re-litigated)
Cross-references
| Path | What |
|---|---|
plans/001-factory-apps-validation/ |
Portfolio decision this plan expands on (verdict #3 triage + architect addendum: v1 zero-PR, INFORMATION_REQUEST + agent-questions group queue) |
plans/002-work-ledger-product/ |
Sibling product research — ledger rows Atlas would render into progress comms |
research/00-dossier.md |
Prior triage build-vs-buy verdict (HumanLayer dead, agent-inbox LangGraph-bound) |
analysis/legacy/SYNTHESIS.md |
Legacy estate the docs/glossary jobs would have to cover (1.05M active LOC) |
repos/monorepo-development/ |
Zaruba greenfield — tasks service, skills, docs conventions Atlas rides on (CONFIDENTIAL: reasons-and-decisions/ never cited here) |