The Coding Agent Harness
A coding agent is only as good as the harness around it. This series covers the structure that makes agent-written code reviewable and safe to ship: specifications instead of vibes, the guards and checks that catch what review misses, what cheap code does to technical debt, and how procurement changes when building is no longer the constraint.
5 articles
- Part 1Guides, Guards, Checks and Sensors: A Framework for Trustworthy Coding AgentsMost coding-agent failures aren't model failures — they're harness failures. A four-part framework for the context, constraints and feedback loops that make agentic engineering safe at enterprise scale.9 min read
- Part 2Your Legacy System Is a Witness, Not Just a LiabilityMost South African replatforming efforts fail because nobody can say, precisely, what the old system actually does. A disciplined approach to using coding agents for reverse engineering — before you write a line of the replacement.10 min read
- Part 3Why 'Vibe Coding' Doesn't Scale in the EnterpriseConversational prompting feels productive right up until it isn't. A practical look at spec-driven development — the discipline of using specs, contracts, examples and decision records to constrain what coding agents build, instead of hoping they guess right.9 min read
- Part 4Keeping Systems Fit: Technical Debt in the Age of Cheap CodeCoding agents make change cheaper — but not all change equally. A framework for deciding which technical debt to pay down now, which to defer, and why the architect's job has quietly changed.9 min read
- Part 5Buy, Build or Grow? Rethinking Procurement When Code Is CheapCheap code is quietly rewriting the buy-versus-build decision — and the procurement processes South African organisations built around it. What 'Grow' means, and how to price exit cost instead of just entry cost.10 min read
Working through any of this in your own organisation?
These articles come out of delivery work rather than theory. If the problems in this series look familiar, a short scoping conversation is usually the quickest way to work out what is worth doing first.
Talk to CloudNala