vForge
Two attempts at AI agents on blockchain-lessons in scope discovery
The Problem
AI agents in Web3 lack persistent identity, coordination layers, and sustainable economics. Two different approaches to solve this-both valuable lessons.
The Process
First incarnation: Agentic Mesh with NFT-based agent contexts on testnet. Second incarnation: Hummingbird trading bot multi-tenant with NFT 'Access Days'. The second pivot hit a hard blocker-open-source framework couldn't support multi-tenant without 250% scope increase.
The Outcome
Project scrapped after hitting technical blockers. Money returned to investors. A valuable case study in scope discovery and knowing when to stop.
Design Decisions
| Decision | Why | Engineering Tradeoff |
|---|---|---|
| NFT-based agent contexts | Persistent identity that survives sessions | Built foundation on testnet, proved concept |
| Pivot to trading bot | Simpler use case, clearer B2C value proposition | Dependent on open-source framework capabilities |
| Kill the project | 250% scope increase made it economically unviable | Lost time but preserved capital |
STAR Summary
| Situation | AI agents on blockchain lack persistent identity, coordination layers, and sustainable economics. Two attempts to solve this-both valuable lessons. |
| Task | Build AI agent infrastructure with blockchain-native identity. First as “Agentic Mesh” with NFT-based contexts, then pivoted to Hummingbird trading bot with NFT “Access Days” for clearer B2C value. |
| Action | Incarnation 1: Built NFT-based agent context structure on testnet-persistent identity that survives sessions. Incarnation 2: Four weeks into Hummingbird integration, discovered the open-source framework wouldn’t support multi-tenant. Would need 250% scope increase to make it work. |
| Result | Project scrapped. Money returned to investors. The right call-no amount of clever engineering makes 250% scope increase fit a fixed budget. A case study in scope discovery and knowing when to stop. |
Incarnation 1: The Agentic Mesh
| Component | Description |
|---|---|
| NFT Agent Contexts | Personality, capabilities, memory live on-chain |
| Persistent Identity | Survives sessions, portable across platforms |
| 4-Layer Hierarchy | Base agents → Specialists → Coordinators → Executives |
Testnet proof-of-concept worked. Budget constraints forced a pivot to something more immediately monetizable.
Incarnation 2: The Trading Bot
| Phase | Status |
|---|---|
| Concept alignment | ✓ Full pod buy-in |
| Discovery | ✓ Complete |
| Additional funding | ✓ Secured |
| Development (4 weeks in) | ✗ Blocker discovered |
The Blocker: Hummingbird open-source framework wouldn’t support multi-tenant. At all. Extending it would require 250% scope increase.
The Decision
| Factor | Reality |
|---|---|
| Sunk costs | Time, money, team effort |
| Technical path | Feasible but economically unviable |
| Decision | Return money, kill project |
Why it was right: No amount of clever engineering makes 250% scope increase fit a fixed budget. Quick, clean kills preserve capital for the next opportunity.
The Lessons
| Lesson | Implication |
|---|---|
| Scope discovery is the real work | 4 weeks revealed what months of planning couldn’t |
| Know when to stop | Sunk cost fallacy kills projects slowly |
| Open-source = constraints | You inherit their limitations |
What Survived
The Agentic Mesh architecture is still valid. The whitepaper documents thinking that remains relevant. vForge isn’t dead-it’s waiting for the right moment.