Latest Blog Posts

• The Current State of Agentic Development

  AgileTechnicalPractices AgenticCraftsmanship AI CodingAssistants

  A survey of what the published evidence actually shows about AI-assisted and agentic software development as of October 2026, drawing on five industry reports, controlled experiments, and code-level research. The short version is that the reports agree on more than they appear to and that most of their disagreements come down to how the measurement was taken.

  Oct 14, 2026

• Running Laguna S 2.1 on a GMKtec EVO-X2

  AgileTechnicalPractices AI CodingAssistants LocalLLMs

  A walkthrough of getting poolside's Laguna S 2.1 coding model running locally on a GMKtec EVO-X2 under Windows, from building the ROCmFPX fork of llama.cpp through to serving the model to OpenCode. It covers the toolchain gotchas, the speeds I measured at different context sizes, and the OpenCode configuration.

  Oct 14, 2026

• An Argument For Agentic Craftsmanship

  AgileTechnicalPractices AgenticCraftsmanship AI Thoughts VibeCoding

  Handing the coding to an agent takes away three things from the engineer: ownership, understanding, and pushback. This article works through what each loss costs a delivery team, why the workflow frameworks, skill collections, vibe coding tools and software factories built so far do not close the gaps, and what a practice that does would have to look like.

  Sep 17, 2026

• Three Levels of Spec-Driven Development

  Thoughts AI SDD CodingAssistants

  Spec-driven development now means at least three different things, and the tools that wear the label are quietly very different from one another. I argue that only the lightest of the three, spec-first, is genuinely useful, and that spec-anchored and spec-as-source repeat mistakes we have already made with code comments and model-driven development.

  Jun 19, 2026

• Sideways, Not Up: Why AI Is Not a New Abstraction Layer

  Thoughts AI CodingAssistants Abstraction

  There is a popular claim that AI coding assistants are the next rung on the ladder of abstraction, sitting above high-level languages just as those languages sat above assembly. I take the strongest version of that argument seriously, then argue it does not hold. A compiler is a deterministic, specified, reproducible abstraction you can stop thinking about. An LLM is none of those things. It does not move us up the ladder so much as sideways into nondeterminism.

  Jun 18, 2026

• Pair Programming, Properly Understood

  AgileTechnicalPractices PairProgramming XP

  A practitioner deep-dive into what the data actually says about pair programming, why most teams "do pair programming" and get few of the benefits, and what the discipline looks like when it is done well.

  May 28, 2026

• Contract Testing in Microservices: Beyond E2E

  AgileTechnicalPractices TDD Microservices Testing

  A practitioner's view of where contract testing belongs in the test pyramid, why it should replace most (but not all) end-to-end tests across a microservice estate, and the organisational discipline most teams underestimate when they adopt it.

  Apr 28, 2026

• Maintaining Code Quality in Modern C# Codebases

  dotnet quality

  Explore the evolution of code quality tooling in the dotnet ecosystem and get practical best practices for using modern Roslyn based analysers effectively. Use centralised configuration, build enforcement, and incremental adoption help teams maintain consistent, high quality C# codebases as they scale.

  Jan 06, 2026

• Spec-driven Development with Kiro

  AgileTechnicalPractices CodingAssistants AI Kiro

  Following on from my Getting Started with Kiro blog post, we start investigating the spec-driven development flow with Kiro.

  Dec 04, 2025

• Getting Started With Kiro

  AgileTechnicalPractices CodingAssistants AI Kiro

  Exploring Kiro, Amazon’s spec-driven AI development tool, by bootstrapping a greenfield project.

  Sep 19, 2025