The AI-Augmented Architect
How AI stopped being a novelty and became infrastructure — from context-loading in brownfield codebases to a full workflow that follows you everywhere.
AI generates plausible code. In a greenfield project, plausible is often good enough. In a large brownfield codebase, plausible is quietly dangerous — calibrated to the median of the internet, not to your conventions.
The fix isn't better models. It's better context.
Open-source Claude Skills for design, architecture, and developer workflows.
Once the context problem was solved, the next question was obvious: what else can be encoded? Skills turn methodology into infrastructure. Every invocation gets the same rigour without re-explaining the framework.
Five world-class designers reviewing every design decision from five philosophical lenses. Not because AI replaces taste — but because the tension between conflicting verdicts surfaces what a single reviewer would miss.
Context-loading. Encoded skills. Planning agents. Code review agents. The pieces assembled into a workflow that starts with a late-night WhatsApp message and ends with shipped, reviewed code — regardless of what device you're on.
How to generate blog cover images that actually reflect your content — using a three-tool pipeline built around visual metaphor, not stock photo logic.
Even the cover images follow the same principle. The pipeline reads the post before it draws the picture. Not keyword matching — visual metaphor derived from the actual content. The pattern is always the same: read first, generate second.
Context, skills and workflow all assume one model doing the work. Once the work fans out to a fleet, the question becomes which model deserves which decision. A month of that, priced to the cent: frontier judgment on the plans, cheaper tiers on the volume, and a month a single model could not have run at all.