We realised that vibe coding gave us quick results, but these were unmaintainable over time. So we swung to the other extreme: stepping through every line the agent delivers. Good quality, but you become the bottleneck. The next step is building the harness: the constraints, skills, hooks, and conventions that

A founder of a growing startup told me this recently. They're under pressure from investors to reach a certain ARR, but the team can't keep up with the feature requests. The fast pace of development has turned the product into a complex system. Jumping on the

The AI wave has split software development into two speeds. On one side are developers running multiple Claude Code instances in parallel, experimenting with voice interfaces, and figuring things out as they go. Their workflows might be held together with duct tape, but they're building a lead that

In 2010, hosting was a solved problem. You no longer needed to set up servers. A cloud account, and you were good to go. AWS, Azure, and GCP kept growing, and suddenly, services could be provisioned with a single click. Free credits were handed out to anyone who asked. Then

Throughout software development history, we have kept trying to hide complexity behind black boxes. JavaScript frameworks abstracted away how browsers render pages. WYSIWYG editors promised early on that you would never have to touch HTML. Layer after layer of abstractions was put in place to hide what is actually happening

I was cautious about flooding a code repository with documentation. There should be documentation close to the code, but in a limited and intentional form. A README with context about the application, a high-level architectural overview, and clear setup instructions to get a development environment running. Some teams add ADRs

Founders need to focus on shipping new features. Revenue depends on delivering more value to users, so the pressure to deliver is real. What often gets overlooked is the state of the systems those features rely on. Legacy modules that need attention, old shortcuts that slow teams down, and parts

Quick feedback is one of the strongest levers in product development. Teams slow down the moment feedback becomes delayed or hard to get. It starts on the developer’s machine. Engineers need to run the application locally, test changes, debug issues, and immediately see the effect of their work. If

Rebuilds have an expiry date. Every day a team isn’t shipping the new version, they move one step closer to declaring bankruptcy on the effort. Teams sometimes need to rebuild a feature because too many shortcuts were taken, documentation is minimal, or key people have left. A rebuild can