My coding agents got faster. I became the bottleneck.
Building Wildo with AI agents taught me to delegate implementation, verify outcomes and stop making myself the approval step for everything.
The Wildo journal
Ideas & engineering
Read the product thinking and engineering decisions behind Wildo.
Building Wildo
Wildo is coming this October
After a year focused on development, a first look at what Wildo has become. The website is live, with launch planned for October 2026.
Building serious software
Why applications need an engine
B2B applications keep rebuilding the same mechanisms. An application engine needs a shared business language—and room for the parts that make each product different.
Writing code is easier. Building serious software is not.
Good code is not a full product. Why capable AI agents still need a coherent foundation to build B2B software—and leave room for what makes it yours.
Research and deep tech
Beyond retrieval: giving application knowledge operable meaning
How domain-specific specifications make application knowledge checkable and usable: a research perspective on semantics, context reduction and coordination in AI-assisted development.
What should an AI coding agent still have to decide?
How domain foundations combine inherited implementation, operable application meaning and custom code, and how to judge their value through change.
Inside the features
Your permissions need to work everywhere your application does
Teams, APIs, attachments and AI assistants all need clear access rules. How Wildo connects enterprise authorization to the resources and actions it protects.
Change the interface without rebuilding the application
From Tailwind and headless components to Wildo's component injection: the quieter changes that made a flexible, connected frontend possible.
Engineering
Labels and emails deserve better
AI can write application copy. The harder part is giving it enough context to keep labels, emails and translations meaningful as the product changes.
Files and MCP: keep the bytes out of the conversation
Large files do not belong in an LLM's tool arguments. Wildo uses consumable tokens to upload the bytes separately, then gives the agent a file reference.