Leadership · Dino

How I build with coding agents

Dino is built with coding agents inside rules I set: machine-enforced constraints, failing tests first, honest failure as a non-negotiable, and checks only a human can weaken.

Founder · Engineering lead

The failure that matters

Coding agents rarely write code that fails to compile. Their characteristic failure is code that looks right and reports success when it should not. My rules exist to make that failure impossible to ship.

The rules

Hard constraints, enforced. About 38 numbered constraints, most enforced in CI, built around evidence, a plan → change → prove → log → stop discipline, and specifications that stay true to the code.
Red first. For every bug fix and every feature, the failing test is written and its failure captured before the code that fixes it.
Honest failure is the moat. Any module that returns a report, score or metric must prove that a degraded run is never reported as success.
Invariants. One report per run, no cross-tenant access, no silently dropped tools, no raw secrets in reports, strict dependency direction, no test theatre.
Checks only a human can weaken. The enforcement files are sealed. Loosening a check takes a written spec, review, and a reseal only a person can perform.

The result

Agents do the volume. I own the architecture, the rules, the review and the release. It is the same operating model I ran at Circo, taken further.

System

Leading an AI-native engineering team

Building and leading a 16-person engineering team across backend, web, mobile, ML and data, with one shared operating model for coding agents.

LeadershipAgentsOperating model

System

Governing coding agents in a polyglot codebase

One instruction set for agents across TypeScript, Java, Go, Python, Swift and Kotlin: shared rules, local boundaries, human review.

AgentsGovernance

System

Deterministic verification and verifiable evidence

Dino validates API behaviour against defined requirements and produces reproducible results backed by execution evidence, sealed into records anyone can verify.

DeterminismEvidenceAttestation

How I build with coding agents

The failure that matters

The rules

The result

Related systems

Leading an AI-native engineering team

Governing coding agents in a polyglot codebase

Deterministic verification and verifiable evidence

The failure that matters

The rules

The result