Writing
Software, AI, founders, systems, markets, and the recurring gap between a good idea and what survives implementation.
A look at the software, tools, experiments, games, and creative projects I've built in 2026 by treating AI, automation, and good systems as leverage rather than substitutes for judgment.
Desktop AI agents are no longer just chat interfaces. They can read files, inspect repositories, execute commands, use browsers, and act inside the same environments where companies keep source code, credentials, and client intellectual property. That changes the security model considerably.
Scylla started as an attempt to restrict what AI desktop applications could access. It became a local-first development workbench built around explicit project scope, brokered credentials, controlled execution, external knowledge, MCP services, remote connections, and durable project memory.
General-purpose tools become expensive when their unused flexibility creates more operational work than the narrow problem they solve. Agentic workflows make purpose-built alternatives increasingly practical.
Complex products rarely arrive fully formed. The trick is not to prevent change, but to build a workflow that lets the product evolve without turning every discovery into a rewrite.
On CER Games, engineered confusion, and turning a livestream community into the game itself.
The tools matter, but the real advantage comes from turning business requirements into durable context, executable plans, verified work, and shared project memory.
What happens inside a product team when leadership does not make clear decisions, and how a non-technical founder can become the source of clarity.
Investors expect ambition. What they do not expect is to discover that the founder cannot distinguish ambition from evidence.
AI has made software dramatically easier to produce. It has done almost nothing to make people want what gets produced.
The bug may be real. The failed deployment may be real. The missed deadline may be real. But the cause often began long before anyone opened the code.
AI-generated code often looks more sophisticated than the thinking behind it. That does not mean the thinking has improved.
A year of building a home with Andrea, caring for three small dogs, getting lost around Oregon, restoring old things and recovering parts of myself I had allowed to become much too small.