Blog
Thinking on specification management, testing workflows, and AI-assisted engineering.
Loop Engineering: Your Loop Is Only as Good as Its Exit Condition
Loop engineering has arrived quickly and the writing about it is mostly about machinery: worktrees, hooks, schedules, sub-agents. Almost none of it asks the question that decides whether the loop is useful or expensive. What is the loop checking itself against?
Keep the Approaches You Tried, Not Just the One You Picked
An architecture decision records the option you chose. The approaches you tested and rejected, and why, usually vanish. Approaches are now first-class in Specsource: candidate solutions with a lifecycle, promoted into a reusable org library, and versioned as the platforms underneath them change.
Mermaid Diagrams in Specsource
Specsource now renders Mermaid diagrams of any type in the Data Flows editor, from a lightweight DSL or raw Mermaid. What they are, examples, and where to find them.
What to Include in a Software Specification
There is no agreed standard for what goes in a software specification. Here is a practical anatomy: the sections that actually remove ambiguity, from context and constraints to C4 and Mermaid diagrams.
Why Code Is the Cheap Part
Anyone can write code quickly, and an agent can clone an app in an afternoon. So the value has moved to the part that is portable across platforms and survives every rewrite: the articulated problem, the decisions, and the architecture.
How to Write Architecture Decision Records
An architecture decision record captures the reasoning behind a technical choice at the moment it was made. Here is the format, what to include, and where most teams go wrong.
How Specsource Actually Works
Capture context, synthesise specifications, and unlock a chain of automation that improves everything downstream. Here is the full picture.
What Is Spec-Driven Development?
Spec-driven development is suddenly everywhere, from GitHub Spec Kit to Amazon Kiro, and plenty of engineers suspect it is just waterfall in Markdown. Here is what it actually is, how the workflow runs, and where the sceptics have a point.
Why Agencies Struggle with Existing Testing Platforms
At every agency I have worked at, testing gets deprioritised. Not because anyone thinks it is unimportant, but because there is always something more urgent. Here is what actually needs to change.
Is Your AI Agent Flying Blind?
The real bottleneck in AI-assisted development is not the model. It is the context you never gave it.
From Scattered Feedback to Test Suite
Client feedback arrives as Slack messages, email threads, and meeting notes. Here is how to turn that noise into a structured test suite that actually protects your releases.