Jonathan Haas
I build the operating layer for accountable AI work: agent orchestration, evals, code review, security guardrails, and the feedback loops that make delegation trustworthy. Previously co-founded ThreatKey and worked at Vanta, Carta, and Snap.
Usually in San Francisco. Always down for a walk. Find me by email, GitHub, or X.
Now: turning AI agents from impressive demos into coworkers you can measure, correct, and safely leave alone.
Operating thesis
- Accountability over autonomy
Agents become useful when their work is inspectable, reversible, and tied to real evidence.
- Evals as product taste
The eval is not just a test suite. It is how a team decides what the system has earned permission to do.
- Context before cleverness
The best systems know the codebase, the customer, the policy, and the operator behind the request.
- Small tools, real stakes
I like personal tools that prove an operating model before they try to become platforms.
Currently thinking about
- Agents and evals14 posts
How to make autonomous work measurable, correctable, and worth trusting.
- Product judgment13 posts
Why the tiny product decisions are usually the control system.
- Founder lessons13 posts
A map of founder traps that mostly look reasonable while you are inside them.
- Security and systems12 posts
Risk is usually a systems-design problem wearing a policy costume.
- Personal systems11 posts
Software gets more interesting when it is allowed to fit one operator exactly.
Best entry points
- How I operate coding agents1 min
The Real Work of Orchestrating AI Coding Agents
- Why product taste is operational1 min
Somebody Gave a Shit: The Quiet Power of Product Detail
- Why small personal tools matter1 min
The Rise of Single-Serving Software
- How founder advice goes sideways1 min
The Three Types of Startup Advice (And Why They're All Wrong)
Recent writing
- The Real Work of Orchestrating AI Coding Agents
Three concurrent coding agents taught me the actual bottleneck: not prompting, but assignment, evidence, review, and release control.
- Building Kestrel: A Context-Aware AI Desktop Assistant in One Session
How I built a full LittleBird clone with screen context reading, meeting recording, arena mode, and MCP tool support — from scratch to packaged .app in a single coding session.
- DiffScope: What Happens When You Give a Code Review Agent Real Context
Most AI review tools see a diff. DiffScope sees the diff, the callers, the type hierarchy, the team history, and knows when to shut up. Here is how.
- The 10-Minute AI POC That Becomes a 10-Month Nightmare
Five lines of Python and an API key produce a working demo. The gap between that demo and a production system contains failure modes the prototype...
- Why Your AI Strategy is Actually a Spreadsheet Strategy
Most enterprise AI transformations are solving problems that spreadsheets handle at 1/50th the cost. The misalignment is driven by career incentives,...
- The AI Agent Gold Rush: Why Everyone's Building Picks and Shovels
Most AI agent infrastructure is premature. The agents themselves barely work. The industry is selling Formula 1 equipment to people still learning to...