I build AI products that do real work, and I build them by directing a fleet of AI engineering agents.
Now
- OneLoop: an AI operating platform. Agents take work from a business signal to an approved, receipted result: orchestration, memory, approval gates, proof receipts, model policy with shadow evaluation, canaries and rollback.
- Kleory: a public memory for claims. People say things; when the truth comes out, Kleory goes back and checks. In production pilot.
- Aithon: an agent-facing B2B services marketplace. Agents discover it via MCP,
agents.jsonandllms.txt; AI proposes, a deterministic contract layer validates price, eligibility and order schema before anything becomes an order.
How I build
Most of my code is written by agents I direct: 36 personas, 279 skills, written specs, review rules and CI gates that the agents can't talk their way past. Recent record, private repos:
| OneLoop | Kleory | |
|---|---|---|
| Commits | ~12,200 (Apr to Oct 2026), 77% by agents | 4,100+ in 38 days |
| CI workflows | 42 | 53 |
| Tests | 2,300+ test files | 7,000+ test cases |
| Shipping | 380+ blue/green production deploys | offline replay of human rulings as a CI eval gate |
The platforms stay private. The pieces worth sharing are here:
- agent-fleet-kit: the rules, hooks and CI gates I use to run a fleet of coding agents they can't talk their way past.
- fail-closed-router: LLM provider routing that falls through only on outages and quota errors, never to launder a bad answer through another model.
- aithon-mcp-server: lets Claude, Cursor and other agents search Aithon's B2B services catalog, read order forms and draft orders over MCP.
Before this
20+ years as a product and growth operator: grew an Amazon Business division from $1B to $3.2B in a single year (2024), led Alpheus from resale to fiber (revenue up 554% in two years, $200M to $1.8B valuation at sale), and was employee through an IPO at Cbeyond.
📍 Dallas-Fort Worth · nathanone.com · X @nathanone · LinkedIn