I write about system design, engineering decisions, and lessons from building real products. Some pieces are complete guides. Others are shorter notes that I will keep improving over time.
We ate our own dog food by building a real app on Yada, and it bit back: twelve sore spots, three flaky tests, a race condition quietly shipping since 0.1.0, and a CI check that held our pull requests hostage.
Meet Yada, our open-source AI agent harness, now live inside TalonHR. It's a standalone TypeScript project for bringing conversational, tool-using agents into existing applications.
Part 3 of the TalonHR AI harness series: what it takes to move an agent from working to production-ready, covering observability, evaluations, durable workflows, cost budgets, permission checks and safe rollouts.
Part 2 of building an AI harness around TalonHR in Go: the execution loop, tool registry, state, approvals, partial failures and testing. Most of the real work is making the software around the model predictable.
Notes from building an AI harness around TalonHR in Go: tool discovery, the agent loop, auth, context, approvals and failure handling. Turns out the language model is only one small part of the problem.
Kenya celebrates its startup funding headlines, but early-stage founders face compliance, capital and procurement barriers built for mature companies. Regulation should scale with risk and size, giving young businesses room to try and fail.