B B B B Blog
Notes, experiments, and insights from real-world development work
This section is a space for documenting development experience, experiments, and practical insights gathered across projects and workflows.
It includes technical notes, breakdowns of processes, lessons learned, and explorations of tools and approaches—covering both successful implementations and trial-and-error iterations.
Some content is authored directly from hands-on experience, while other posts are generated or assisted through my AI-augmented workflow, reflecting how knowledge is captured and refined within my development system.
SDLC & Process
The Spec Is the Interface: How I Keep AI Agents from Forgetting the Goal
Spec-driven development for AI agents: goals, stories, and decisions as structured YAML the agent reads at every task start. The spec is the source of truth.
SDLC & ProcessFeaturedI Built a Machine-Checked SDLC for AI Agents — Then Measured It: 82% of My Tasks Were Force-Closed
I built a machine-checked SDLC for AI coding agents, then measured it: 82% of tasks force-closed. What the data disqualified, and the redesign that followed.
SDLC & ProcessAI-Augmented Dev
Schema First, Automate Second: Turning Repetitive LLM Tasks into Verified Pipelines
A model quietly upgraded "villa" to "pool villa" — valid JSON, plausible, wrong. The schema, validator, and hook method that turns LLM tasks into automation.
AI-Augmented DevThe MCP Server I Didn't Ship: LLM Tool Calling Across Go, Python, and Svelte
The security review said no public MCP server. What shipped instead: tool contracts as data — published by Go, enforced by Python, toggled in Svelte.
AI-Augmented DevThe Prompt Fix That Never Propagated: Prompt Management as Linked Documents, Not Strings
A prompt fix that never reached its diverged copies taught me the unit of prompt reuse is the section. Prompts as linked documents, with zero-spend evals.
AI-Augmented Dev