About

We're at a turning point. AI tools no longer suggest code — they write it, test it, commit it. This isn't a change of degree. It's a paradigm shift in how we work.

postcursors documents what this shift means in practice, from the perspective of a production workflow. What actually holds up on projects running in prod — not impressive demos or benchmarks without context.

What this site is not

Not an AI news aggregator. Not a beginner course. Not a sponsored tool list — there's no advertising here, no affiliation, no deals with the vendors of the tools reviewed. Opinions are based on real usage, and flaws are documented as thoroughly as advantages.

The "postcursors" in the name

Cursor IDE was the first visible disruption — the first mainstream tool where AI wasn't an add-on but the central interface. Since then, several waves of tools have followed: Claude Code, OpenCode, Kilo Code, dozens of others. We're post-Cursor in the sense that this disruption is behind us — it happened, the tools exist, the question now is how to work with them.

It's too late to be a precursor, but still too early to be obsolete. The devs who come after — those arriving now in a world where AI writes code — need concrete reference points. That's what this site documents.

Founder context

Full-stack dev. PHP/Symfony backend, TypeScript/React frontend, GKE in prod. Terminal-first workflow: Neovim, herdr as local multiplexer (tmux stays on the servers), SSH. I've been using AI in code since the GitHub Copilot technical preview in June 2021 — autocomplete, then agents, then full agentic workflows. Currently on Claude Code (Max subscription) in daily use — after leaving it and coming back, a round trip documented in the dispatches.

The practice doesn't stop at my terminal: I evaluated and rolled out a coding agent for a team of 4 devs — onboarding time measured, conventions and commands versioned in Git, model choice documented and corrected when it broke adoption. Structuring AI practice so it outlives individuals is this site's common thread: adoption, costs, tooling governance.

This context informs every piece of content on the site: the Symfony examples are valid, the cost comparisons are based on real usage, the recommendations apply to production projects — not demos.

Frequency and format

No marketing cadence here: the priority is evergreen content — guides, dated dispatches, tool reviews, and TILs as discoveries are made. Every piece carries its last-updated date, and content that can no longer be defended gets corrected or removed.

There's an RSS feed to stay up to date.

How this content is produced

A site about agentic practices that stayed quiet about AI in its own production would have a problem. So, plainly: part of this content was written with an agentic harness — source research, outline, first draft, proofreading, translation into English depending on the piece. Every technical claim is verified and rewritten by hand before publishing; that's the condition, not a formula.

Since July 2026, the pieces concerned carry the mention at the bottom of the page: versioned harness, model, exact scope of the intervention. For earlier content, that per-article traceability doesn't exist — usable session history doesn't go back further, and part of the work was done from another machine. Rather than reconstructing plausible mentions after the fact, this paragraph stands as the declaration for that period: yes, AI assistance, on tools that changed over the months (OpenCode, Kilo Code, then Claude Code); no, no per-article detail, for lack of evidence.

The rule applies to this paragraph too: when the data is missing, we write that it's missing.