This is a running list of articles, threads, and posts on AI-assisted development, automation, and software craft that I've read and thought were genuinely useful. Not a firehose — just the things I'd actually forward to a colleague.
I run Claude Code across four machines and kept losing track of which session was still working and which had finished and was waiting on me. Rather than adopt one of the many existing dashboards, I built my own as a learning exercise — a small always-on-top panel, one color-coded row per session, click to jump to its terminal, all machines in one dock. It's Windows-first (most of the work was wrestling with how Windows Terminal and Claude Code's TUI interact) and it's now public on GitHub, MIT-licensed.
Read the article →Now that AI can generate working software cheaply, the obvious conclusion is that SaaS is doomed — why pay for a tool you could just build? This piece argues the opposite, and the analogy it hangs the argument on is genuinely clarifying: people have always had flour, water, and an oven, yet they still buy bread. What they're paying for isn't the recipe, it's convenience, consistency, and someone accountable when it breaks. AI lowers the cost of the flour; it does nothing about the far harder work of building and maintaining real business infrastructure. A sharp counter to the "everyone will just build their own" narrative.
Read article →Most AI productivity writing traffics in impressions — "it feels faster," "I ship more." This post does the harder work of measuring actual outcomes across eight months of real-world use. The numbers give you a realistic baseline for evaluating whether AI-assisted development is worth the investment, and the methodology is honest about what it doesn't capture. A useful counterweight to the hype.
Read article →This stands out from almost everything else written about AI development tools because it's grounded in how Anthropic's own engineers actually work — not a tutorial, not a vendor pitch, but a window into evolved practice. The framing of the future as "orchestrating intelligence rather than writing code" is the kind of formulation that genuinely shifts how you think about your own workflow. Unusually well-crafted for the genre.
Read article →Written around Anthropic's December 2025 announcements, this piece does something most AI commentary fails to do: it explains not just what changed but why it matters architecturally. The move toward stateful agents isn't an incremental improvement — it reframes what AI-assisted development actually is. If you want a clear marker for where the field stood at the end of 2025, this is it.
Read article →There's a threshold in AI-assisted development where prompting skills stop being the bottleneck and architectural decisions take over. This post identifies that threshold clearly. Once you're orchestrating multiple agents, you need to think about state management, failure modes, and agent boundaries the same way you'd approach distributed system design. This is where "just ask Claude" ends and real engineering begins.
Read article →While most agent discussions in 2025 stayed theoretical, this thread showed practitioners building real pipelines: forms intake, CRM updates, document generation. The specifics matter — not "AI can automate business workflows" but here's what actually broke, what held up, and what the integration seams look like in practice. An early and useful map of the territory for anyone considering forms-to-CRM automation.
Read thread →This thread cuts through the hype with something harder to find: honest longitudinal experience from someone who's past the honeymoon phase. The central observation — that raw speed is the wrong metric for AI-assisted development — is one that most tutorials and vendor marketing actively obscure. There's also a canonical warning buried here about over-reliance that most developers ignore until they've learned it the hard way themselves.
Read thread →One of the most honest accounts of what working with Claude actually felt like for an experienced developer in 2025 — the initial skepticism, the moments it genuinely surprised, the friction when it didn't, and the gradual recalibration of expectations. The emotional arc here is real: nearly every serious developer went through some version of this in the past year. Worth reading not for the conclusions but for the recognition.
Read article →This thread surfaced early in 2025 when most developers were still treating AI as a sophisticated autocomplete tool. What makes it worth revisiting is that the community was already articulating something that would define the rest of the year: the interesting question was never about raw capability, but whether you could trust an AI to act autonomously on your behalf. The discussion got there before the consensus did.
Read thread →