← writings

writings · 01 Jan 2025

no longer the gatekeeper

ai · development · architecture · tooling

// ai-assisted development · ongoing


In mid-2023, I made a deliberate bet and stopped reaching for Google, using AI for everything instead. The tooling was immature (copy-paste), workflows were non-existent, so it wasn't an obvious move. I made it because it was clear what was coming was not just hype.

I'd made the call before, with Ruby on Rails when it was still niche and Node.js before the ecosystem caught up. What looks like trend-chasing is really a habit of noticing early when something is going to reset the baseline, then building the capability ahead of the demand for it.

By 2024 the question had moved from the tools to the team. Everyone was already using AI privately, almost apologetically, as though it were cheating. It took a deliberate conversation with management to name what was happening. AI-assisted development was nothing to hide. Every organisation now has to lean into it or fall behind. That conversation changed how the team worked.

By 2025 the frontier had moved again, toward agentic coding, autonomous workflows, and AI wired directly into the build and review pipeline. The tooling is extraordinary. Capability has stopped being the hard part. What's hard is to get deterministic outcomes from components that are inherently fuzzy. That problem is genuinely hard, and most people haven't hit it yet.

Two years of deliberate practice taught me that judgment matters as much as the tools. AI can do things that were impossible five years ago, and in the same breath, it is confidently wrong about things people assume it has covered. Telling those two apart is the whole skill.

Software is no longer the developer's domain alone. Founders can get further than ever without a technical co-founder, and developers are no longer the bottleneck between an idea and a shipped product. That is a good thing, provided you are honest about what it means for how you work.

You would expect the stack choices to get easier... instead, they have grown harder, and bad ones are punished by pace. The competition to build good software has never been more intense, nor more open to anyone willing to step in.

back