Blog
Field notes on shipping AI-native systems with human engineering judgement: architecture, modernisation, automation, and the tradeoffs in between.
Design MVPs to be intentionally replaceable-validate demand quickly, reduce technical debt, and avoid costly rebuilds that can exceed $500,000.
AI-generated codebases fail in three consistent ways: inconsistent duplicated logic, exposed secrets and missing error handling, and untested fragile architecture. Fix them early.
A risk-aware playbook for modernizing fragile legacy code: assessment, phased fixes, feature flags, and the measures that show whether it is working.
Why startups take on technical debt, how to track and prioritize it, and when to pay it down across seed, growth, and scale stages.
AI tools doubled pull requests but also raised bugs, review time, and technical debt; prioritize Net Feature Velocity, DORA metrics, refactoring, and strict reviews.
A production-focused review of AI coding tools: which hold up under real constraints, where they fail, what they cost in tokens, and the governance practices that keep them safe.
AI-generated code accumulates invisible technical and comprehension debt, causing apps to become fragile around 10,000 users - symptoms, timeline, and fixes.
Same production codebase, both tools. Claude Code finished in 9m09s on a test subset. OpenCode took 16m20s, ran all 94 tests, and lifted coverage 29%.
Modularity, feature-based code, clear state layers, performance tuning, and automated testing/CI keep large frontends maintainable.
Why AI needs specialized monitoring: track data and concept drift, bias, real-time outputs, and business KPIs with tools like MLflow and Prometheus.
Design software that respects domain expertise by simplifying technical barriers with Event Storming, adaptive UX, and no-code/AI tools to speed adoption.
How AI speeds ERP/CRM decisions, maps trade-offs with Decision Graphs and ICAs, and pairs autonomy with governance to boost efficiency and ROI.