← Back to work
v0.2

DiscoveryOS

A methodology-as-system that prevents premature solution-building. Four AI agent roles, five core artifacts, and structured interview guides enforce discovery before design, all encoded in markdown with zero code dependencies.

The Problem

Consultants and service designers have a universal failure mode: they design solutions before understanding the real problem. They skip discovery, make assumptions about buyer motivation, build service packages based on what they can deliver instead of what clients need, and only discover the mismatch when revenue doesn't show up.

The root cause isn't lack of skill. It's lack of discipline. There's no system enforcing evidence collection before design decisions. DiscoveryOS makes it structurally impossible to skip discovery by encoding the discipline into the framework itself.

The Framework

Four phases with non-negotiable gates between them. You cannot advance to the next phase until the current one's evidence requirements are met. Each phase has a dedicated AI agent role with explicit boundaries on what it can and cannot do.

1

Discover

The Evidence Collector catalogs raw inputs: interviews, market signals, observations. No interpretation, no patterns, just structured evidence with unique IDs (E-XXX). Gate requirement: 5+ evidence entries across 2+ types, gaps explicitly flagged.

2

Define

The Pattern Extractor surfaces themes across 2+ independent evidence sources. The Assumption Challenger questions every claim. Outputs: Problem Map, Buyer Logic Map (Jobs To Be Done), and an assumption register where every belief is named and tracked.

3

Develop

The Architect designs three core artifacts (Service Architecture, Business Model, and Proof System), always presenting 2+ options per decision. Every design choice traces back to evidence. No artifact exists without a source chain.

4

Deliver

Final Assumption Challenger review. Validation activities test remaining assumptions against reality. Artifacts are finalized only after evidence gaps are closed or explicitly declared as risks.

Key Decisions

Evidence traceability via ID system

Every piece of evidence (E-XXX), pattern (P-XXX), assumption (A-XXX), and challenge (C-XXX) gets a permanent, unique ID. The chain is auditable: Artifact → Pattern → Evidence → Raw Source. This prevents the most dangerous consulting failure: recommendations that sound good but aren't grounded in reality.

Non-leading inquiry

Interview guides enforce past-behavior questions: "Tell me about the last time..." instead of "Would you..." Hypothetical questions tell you what people think they'd do. Behavioral questions tell you what they actually did. The guides include verbatim scripts, redirect phrases, and post-interview processing checklists.

Assumption discipline

Every assumption is named, tracked in a register, and challenged. The Assumption Challenger agent exists specifically to surface unstated beliefs. Untested assumptions are declared, not hidden. This is the core of the framework: you can't design a good service on beliefs you haven't examined.

Agent role boundaries

Each AI agent has explicit can/cannot rules. The Evidence Collector cannot interpret; it can only catalog. The Pattern Extractor needs 2+ independent sources before declaring a theme. The Architect must present multiple options. Tight boundaries prevent role drift and keep each phase honest.

What I Built

4 agent definitions

Collector, Extractor, Challenger, Architect

5 core artifacts

Problem Map, Buyer Logic, Service, Model, Proof

3 interview guides

Practitioner, Client/JTBD, Failed Prospect

100KB+ research

JTBD methodology, credence goods, inquiry methods

Phase gate system

Non-negotiable criteria between phases

Zero code deps

100% markdown, Git versioned, AI-native

What I Learned

The biggest insight from building DiscoveryOS is that methodology is architecture. The same principles that make good software (separation of concerns, explicit interfaces, validation gates) apply directly to service design processes. A framework isn't a checklist; it's a system with constraints that prevent common failure modes.

I also learned that the hardest part of evidence-driven work isn't collecting evidence. It's resisting the urge to interpret too early. The Evidence Collector agent's "no interpretation" rule was the most difficult constraint to enforce, and the most valuable. Premature pattern-matching is how consultants end up solving the wrong problem with confidence.