An AI-powered Go-To-Market strategy engine built on Claude Code's multi-agent orchestration. It takes a product brief and scenario context, then produces a complete, source-cited GTM strategy through a pipeline of specialized AI agents with human review gates between each stage.
This repo is both the system (agents, skills, schemas) and a complete worked example: a full GTM strategy for Docebo AgentHub, an enterprise AI learning platform.
Disclaimer: The Docebo AgentHub run is an independent, illustrative analysis produced by this engine from publicly available information, to demonstrate the system on a real-world scenario. It is not affiliated with, endorsed by, or commissioned by Docebo, and it does not use any confidential or insider information. Market figures and recommendations are AI-generated and should be independently verified before any real decision-making.
Most "AI does your strategy" demos are a single prompt that returns a wall of unverifiable text. This engine is built like a real analyst team:
- Separation of concerns — each agent owns one job (market sizing, ICP, channels, messaging, execution) instead of one model doing everything.
- Grounded, not hallucinated — agents do live web research and must cite sources for every factual claim, with explicit per-question confidence ratings and flagged gaps.
- Structured output — every agent emits JSON validated against a schema in
schemas/, so outputs are consistent and machine-consumable. - Human-in-the-loop — a review gate sits between every agent; downstream agents only run on approved upstream output.
- Guardrails — explicit constraints prevent the model from inventing product capabilities or recommending channels that earlier stages didn't justify.
Product brief + scenario context
│
▼
┌─────────────────────────────────────────────────────────────┐
│ Agent 1 — Market Intelligence (10 questions, Q1–Q10) │
│ built from 4 parallel research sub-agents: │
│ • market-sizing • geography │
│ • competitive-landscape • market-structure │
└─────────────────────────────────────────────────────────────┘
│ ▼ human review gate
▼
Agent 2 — ICP Definition (6 ICP dimensions)
│ ▼ human review gate
▼
Agent 3 — Channel Strategy (GTM motion + channels + partnerships)
│ ▼ human review gate
▼
Agent 4 — Messaging Framework (positioning + per-persona messaging)
│ ▼ human review gate
▼
Agent 5 — 90-Day Plan (phased execution, milestones, metrics)
│
▼
Final GTM strategy (one JSON output per agent) + executive summary
An orchestrator agent collects inputs, runs each agent in sequence, manages the review gates, and passes approved output downstream. Each later agent only consumes the approved outputs of the agents before it — so the strategy compounds instead of contradicting itself.
| Stage | Agent | Job |
|---|---|---|
| Orchestrator | gtm-orchestrator |
Runs the pipeline, manages review gates and data flow |
| 1 | market-sizing, geography, competitive-landscape, market-structure |
Market intelligence research (10 questions) |
| 2 | icp-definition |
Ideal Customer Profile across 6 dimensions |
| 3 | channel-strategy |
GTM motion, channel prioritization, partnerships |
| 4 | messaging-framework |
Positioning angles and per-persona messaging |
| 5 | ninety-day-plan |
Concrete phased 90-day execution plan |
Passive reference documents that enforce consistency across agents:
web-search-protocol, quality-check, output-format, human-review.
.claude/
agents/ # Orchestrator + 8 subagent definitions (the system prompts)
skills/ # Shared protocols: web search, quality check, output format, review
schemas/ # JSON output schema per agent (validation contract)
runs/
docebo-agenthub/
inputs/ # The product brief + scenario that drove this run
staging/ # Agent 1's intermediate sub-agent outputs
outputs/ # Final output per agent (JSON) + executive summary
review/ # Human review notes per agent
gtm-engine-full-output.md # All agent outputs consolidated in one file
CLAUDE.md # Project instructions / constraints the agents operate under
The runs/docebo-agenthub/ directory is a complete run
for a real product scenario — launching Docebo AgentHub, an agentic-AI platform
that turns enterprise knowledge (Confluence, SharePoint, Google Drive, Slack,
Teams, etc.) into trackable learning content for enterprise L&D teams.
Start here:
- Executive summary — the one-page takeaway.
- Full agent outputs — every agent's complete output in one file.
- Individual JSON outputs in
outputs/.
Claude Code — agents are defined as markdown files with frontmatter; skills are reusable reference documents; the orchestrator coordinates subagents and human review gates.