How MAIA Works
What is MAIA Decision OS
MAIA Decision OS is a governed decision rehearsal system built on Anthropic's Model Context Protocol (MCP). It connects to Claude as an integrated tool and allows any user to stress-test a complex organizational decision before committing resources, capital, or reputation to it.
MAIA is designed for leaders and teams making high-stakes decisions with multiple dependencies, stakeholders, and potential cascade effects. Acquisitions, market expansions, restructurings, vendor consolidations, platform migrations, and capital allocation choices are all within scope. MAIA is not designed for personal decisions or simple binary choices.
The system orchestrates 47 specialized agents through a governed 9-stage pipeline. Each stage produces structured analysis — dependency maps, cascade simulations, countermeasure designs, readiness assessments — culminating in an evidence-scored Decision Brief with a viability verdict. The entire rehearsal is delivered as a conversation in Claude, with a downloadable Visual Intelligence Package at the end.
How It Works
MAIA runs a 9-stage pipeline. Each stage builds on the outputs of the previous stage. The Navigator (the person making the decision) guides every step and approves all evidence before it enters the analysis.
- S0 Session Setup and Data Ingestion — System initializes, validates agents, and ingests any uploaded documents.
- S1 Decision Capture — The Navigator describes their decision in their own words. MAIA validates the intent.
- S1a Decision Classification — MAIA classifies the decision type (commitment, direction, operational, threshold, or compound) and detects the industry.
- S1b Data Requirements — Per-category data requirements are derived from the decision type and industry profile.
- S1c Schema Discovery — Available data sources are scanned and a data scope meter is populated with coverage levels.
- S1d Gap Resolution — Material data gaps are identified. The Navigator can authorize governed external research, upload additional documents, or proceed with known unknowns.
- S1e Evidence Approval — Each external research finding is presented individually for Navigator approval before entering the analysis.
- S2 Dependency Mapping — Five agents map dependencies across eight categories: data, process, integration, security, policy, risk, analytics, and people.
- S3 Timeline Construction — The Navigator confirms a decision horizon. Phases, milestones, and irreversibility points are established.
- S4 Structural Analysis — Whole-system structure is mapped with cascade pathways at 1, 3, and 5 steps. Fragility points and silent failure zones are identified.
- S5 Disruption Scenarios — Disruption candidates are generated from the dependency structure. The Navigator selects which scenarios to simulate.
- S5a Cascade Analysis — Multi-agent simulation propagates each selected scenario through the dependency map at 1, 3, and 5-step depths.
- S6 Countermeasure Design — Stabilizers are designed for each disruption. Each countermeasure is scored for feasibility and mapped to specific dependencies.
- S6b Readiness Assessment — Five-dimension readiness synthesis. Baseline vs. with-stabilizers comparison.
- S7 Decision Brief and Viability Verdict — Six agents synthesize a complete brief: verdict, key factors, optimal path, recommended sequencing, risk register, and monitoring indicators.
- S8 Visual Intelligence Package — A single downloadable HTML file with eight interactive cards containing the full analysis. Opens in any browser. No software required.
Data and Privacy
MAIA does not store decision content. The rehearsal runs entirely within the Claude conversation. When the session ends, no decision data is retained on MAIA's servers. The analysis lives in your Claude conversation history and in the Visual Intelligence Package you download.
If you are on a Claude Free, Pro, or Max account, Anthropic's default setting allows conversations to be used for model training. Before describing a sensitive decision, go to Settings, then Privacy, and toggle "Help improve Claude" to OFF. If you are on a Claude Team, Enterprise, or API account, your conversations are already protected under Anthropic's commercial terms.
The Visual Intelligence Package is generated within your Claude session and downloaded directly to your device. No data is transmitted to external servers during export. The file is fully self-contained and works offline in any browser.
Connectors
MAIA can use Claude's built-in connectors to enrich the rehearsal with data from your existing tools. Connectors are optional — MAIA operates on Navigator input alone if no connectors are active.
Google Drive — Strategy documents, financial reports, board decks, analyst coverage, and planning materials that ground the analysis in your actual situation.
Gmail — Correspondence context and stakeholder signals that inform the dependency and stakeholder maps.
Google Calendar — Timeline and deadline context that informs the temporal analysis and irreversibility points.
Authentication is handled through Claude's standard OAuth connector flow. MAIA reads from connected sources only — it does not write, modify, or delete any data in your connected accounts.
Usage Examples
"We are evaluating whether to acquire a regional competitor with $40M in annual revenue. We need to decide whether to proceed to LOI within 60 days."
Output: Dependency map across 8 categories including financial, legal, operational, and people integration risks. 4 disruption scenarios simulated including regulatory delay, key talent departure, customer concentration risk, and integration complexity cascade. Countermeasures scored for feasibility. Viability verdict with confidence level. Visual Intelligence Package with full analysis.
"We are deciding whether to expand our SaaS product into the European market in Q3. We have a small team and limited runway."
Output: Regulatory and operational dependency mapping covering GDPR compliance, localization requirements, and partner channel dependencies. Resource cascade analysis showing team capacity constraints. Timeline stress-test with irreversibility point at hiring commitment. Go/no-go readiness score with stabilizer recommendations for phased entry.
"We need to decide whether to restructure our sales team from geographic territories to industry verticals before our next fiscal year."
Output: Human capital dependency map covering territory relationships, client continuity, and comp structure cascades. Revenue risk cascade showing pipeline disruption at each transition phase. Transition timeline analysis with 90-day and 180-day milestones. Countermeasure design for retention risk including interim dual-reporting and client handoff protocols.
Setup Instructions
- Open a new chat in Claude.ai (or the Claude desktop app).
- Type: MAIA Decision OS
- MAIA activates and delivers the welcome text explaining the rehearsal process.
- If you are on a Claude Free, Pro, or Max account: go to Settings, then Privacy, and toggle "Help improve Claude" to OFF before proceeding.
- Answer the document upload question. You can upload financial reports, strategic plans, board decks, or other relevant materials. Or type "No documents" to proceed without uploads.
- Describe your decision in your own words. Be specific about what you are deciding, the timeline, the stakes, and any constraints.
- Follow MAIA's stage prompts through the rehearsal. You will approve evidence, select disruption scenarios, and confirm at each gate.
- Download your Visual Intelligence Package at the end — a single HTML file with eight interactive cards containing the full analysis.