Mitchell Agentic Sprint

AI-led 6-step sprint that walks an AI builder through validating one idea — vertical, buyers, framework, positioning — to a sales deck, outreach plan, and pre-seed investor deck. Adversarial by default. Output is artifacts, not product.

View project on GitHub

ADR-002: Build Tier — Moderate with NotebookLM Bolt-on

Field Value
Decision ID ADR-002
Initiative v0.1.0 scope and shape
Proposed By Chris Barlow
Date 2026-05-08
Status Accepted

WH(Y) Decision Statement

In the context of building MAS v0.1.0 to deliver the concept’s headline promise (anti-sycophancy + investor-conversation-ready artifacts) within a ~9-week budget,

facing three candidate build tiers identified in the build-options research — Minimum (~2 weeks; profile-pack-only Council), Moderate (~7–8 weeks; new NPC agents + step gates + loopback), Extensive (~12–16 weeks; tooling layer with slide rendering, theme clustering, evals) — plus the NotebookLM bolt-on (~+1.5 weeks),

we decided for Moderate + NotebookLM bolt-on (non-Enterprise path), with the first week shaped as a Minimum-equivalent walking skeleton (Step 1 + single Seed VC Council voice) used as a tester gate before continuing,

and neglected Minimum-as-final-shape (cannot deliver the headline promise — Council voices feel interchangeable when re-skinned from Dragon; Mom Test discipline is unenforceable; investor decks lack defensible evidence) and Extensive (premature — concept §10 itself flags core uncertainties around linear flow, named-expert assumption, and AI-as-coach replacement; tooling layer becomes justified only after tester demand pulls it forward),

to achieve real domain-knowledgeable Mini Council voices, enforced Mom Test discipline, working loopback, markdown-first artifacts that close 80% of the headline value, and NotebookLM-augmented synthesis for Steps 3–6 that aligns with the concept’s “verbatim quotes only” rule,

accepting that anti-sycophancy verification via eval set is deferred to v0.2.0, slide rendering (PPTX/PDF) is deferred to v0.2.0, and the unofficial NotebookLM CLI carries Google ToS risk that must be mitigated by a Claude-only fallback path.

Options Considered

Option Verdict
Moderate + NotebookLM bolt-on (Selected) Hits headline promise; testable in 9 weeks; integration deferred to week 5 doesn’t block walking skeleton
Minimum only Rejected — see build-options §1 risks
Extensive Rejected — premature per recommendation §rationale
Moderate without NotebookLM Considered — viable, but loses ~30% of synthesis quality on Steps 3–6 for ~1.5 weeks of work; user explicitly opted in to the bolt-on

Dependencies

Relationship ADR ID Title Notes
Enables ADR-003 Sibling-plugin shape Implements the Moderate architecture
Enables ADR-004 Mini Council as new SKILL.md NPCs Required by Moderate, not Minimum
Enables ADR-006 NotebookLM via MCP The bolt-on portion

References

Reference ID Title Location
Research Build options analysis docs/research/build-options.md
Research Recommendation document docs/research/RECOMMENDATION.md
Concept Mitchell Agentic Sprint concept docs/concept/concept.md

Status History

Status Approver Date
Accepted Chris Barlow 2026-05-08