Case study · Independent project
Tomorrow.
An experimental multi-agent orchestration system for coordinating planning, delegation, execution, governance, and verification.
- Role
- Creator & Architect
- Period
- 2026 — now
- TypeScript
- Node.js
- Tool contracts
- Multi-agent orchestration
- Evals
What Tomorrow is
Tomorrow is a technical project I began creating in 2026. It is not a company. The project explores how specialist agents can coordinate complex work while preserving evidence, permissions, and a clear definition of completion.
The orchestration model
A coordinator decomposes an outcome into bounded work, delegates to specialist agents, records observations, and routes artifacts through verification before the run can finish.
- — Typed tools with explicit authority
- — Shared task ledger rather than unlimited transcript memory
- — Recoverable versus terminal failure classification
- — Verifier gates and human approvals
Why multi-agent
Multiple agents are useful only when separation creates a real boundary: distinct context, permissions, expertise, or independent verification. If the work is sequential and shares the same context, one agent with good tools is usually simpler.
Reliability before autonomy
The system earns authority by completing narrow workflows consistently. Task success rate, verification pass rate, retries, tool failures, human intervention, latency, and cost matter more than the number of agents involved.
Current direction
The work is focused on orchestration primitives, evaluation, operational memory, governance, and evidence-based completion. Public claims are deliberately limited to what the project actually demonstrates.