Full-stack engineer orchestrating teams of people and agents. Writing about agentic development, developer tools, and MCP. Australia.
Last week I attributed a measured performance improvement on a platform I work on to a specific sequence of technical components. That sounds unremarkable until you know I had been circling the same question for years.
There’s a pattern showing up in how teams talk about agentic coding. The agent writes code fast, faster than any of us could alone, so the next move feels obvious: ship it fast too. Out come the words that used to mark a mature engineering org. Trunk-based development. Feature flags. Continuous delivery, many releases a day. The reasoning goes that if generation is no longer the bottleneck, the…
I just published pixel-llm, a small autoregressive transformer that generates 32x32 pixel art sprites of reef sea creatures. About 2.9 million parameters, a 64-colour palette, runs on consumer hardware. Built end to end through agent sessions, with me steering rather than typing.
At a quarterly meeting in our group last week, senior leadership made the expectation clear: everyone should be using agentic tooling to build apps that solve problems they actually have. Not just engineers. Everyone.
Does it actually matter when you post your Show HN? And does a front-page run translate into GitHub stars? I scraped 188,085 Show HN posts and cross-referenced the top 500 with their GitHub star histories to find out.
Martin Fowler’s latest fragments post collects several ideas about how AI is reshaping software development. The one that stuck with me is Ajey Gore’s argument: as coding agents take over execution, verification becomes the premium activity.
I typically have half a dozen Claude Code sessions running at once, spread across different terminals and monitors, some hidden behind other windows. The visual “done” indicator is easy to miss when you’re not looking at the right terminal.