I started 2026 with a work-related resolution: don't start in the editor, ask an agent to write code first.

Near the end of last year, my experience with AI was limited to using Claude as a glorified template generator, or tab completion on steroids. I never really boarded the vibe coding train. I didn't have an issue with agentic development, but I didn't feel compelled to use it either.

It's difficult to break out of 15 years of experience. Sure, tools have evolved in the time I've been doing this, but they've always revolved around an editor, a browser, and a terminal. I'm not the adventurous type to jump on the latest tooling hype train before it's been tried and tested by others. Muscle memory always kicked in before I even thought of anything else than opening PhpStorm. Claude never stood stood a chance.

There was a vibe shift in December with the Opus 4.5. People were really excited about the capabilities. More importantly: people I respect in the field were excited, not just the loudest voices on the internet.

In comes my New Year's resolution: whenever I need to write write code, I'll ask an agent to do it first. If I didn't force myself to break away from old habits, I'd never give it a proper shot.

Fast forward three months, and Claude Code & Codex are my daily drivers when coding. I don't even have a PhpStorm license anymore.

I'm still early in the process of learning what works and what doesn't. But the best way to fully understand the potential of a tool is to keep pushing it beyond its limits.

One thing I'm sure of: I don't want AI to replace my thinking, and this is easier said than done. Thinking now requires self-discipline. It's easier to effortlessly prompt the machine and forwarding the output as is. I don't want to aim for good enough. I want AI to be a lever for better work.

• As a follow-up to those thoughts, Rasmus Andersson hit the nail on the head on X: "There's only one playbook that remains: make the best product. […] You can’t blame timelines for janky scrolling or broken text editing anymore. Build something that’s meaningfully different, something that you can be proud of a decade from now."
• AI enables us to do things faster, and sometimes better than we’ve been able to before. But it has its limits. And as we learn those limits, the work we do will shift to avoid them. Some great olive oil & sewing analogies by Robin Sloan in Flood fill vs. the magic circle.
• Reality has a surprising amount of detail is a perfect follow-up to Robin's article. "The more difficult your mission, the more details there will be that are critical to understand for success."
• I wrote a post on how I write feature specs. Process: collect all input, consolidate a cohesive list of requirements, review and edit. Main tools: Granola, Codex, Monologue, Spiral, iA Writer.
• I'm looking forward to trying Glaze from the Raycast team. They have good taste and I trust them getting this right. Looking forward to diving into the worst idea possible: building my own todo app.
• I'm a sucker for a good cooking analogy. Liam Hammett's Two Soups Two Cookies is about the difference between "it works", and "it feels right". You can't really condense this article into a single quote. You're gonna have to read it to feel it.
• Matt Pocock shared a /grill me agent skill: interview me relentlessly about every aspect of this plan until we reach a shared understanding, and if a question can be answered by exploring the codebase, explore the codebase instead. Incredibly simple, but the times I've used it so far it has been incredibly effective.
• I've recently switched to Current as my main RSS reader. It has a unique "river" feature: articles arrive, linger for a time, and then fade away. You can configure how long they stay. Breaking news becomes stale before greenfield essays do.
• If I were to share some thoughts on making a meaningful impact at work, Ben Kuhn's Impact, agency, and taste would be it. Don't rely too much on permission or encouragement. You can just do things. The hard part is being honest with yourself: sometimes we avoid making the plan explicit because we're afraid it will surface failure.
• Henrik Warne in praise of –dry-run. I've added --dry-run to scary destructive stuff I'm planning to run on prod, but not as often as I've wanted. This is the kind of thing I want AI to give me the breathing room to do.
• I love digital tools that have some analog look or feel to them. Speaking as an Excalidraw super user, I'm going to have to find an excuse to use MarkerHighlight.js.
• I’ve seen a lot of strong opinions about disclosing whether code in a PR was written by hand or generated by AI. I don’t really care. The author owns the code in the first place. The author and reviewer have a shared responsibility for what happens on production. I like Vercel's view on using agents responsibly while multiplying their shipping cadence.
• RollerCoaster Tycoon, a classic game from 1999, is still worth dissecting for its performance to this day. An interesting lesson on nudging requirements to align with the grain of our programs.

It sure feels good to be writing again! I'll be sending these out every two weeks. Thank you for reading, and talk to you soon!

— Seb