Aron Atkins

I just finished the 2026 Agentic Coding Trends Report. Anthropic makes it clear. We’re not moving from “developers” to “AI replacing developers.” We’re moving from implementers to orchestrators. The real shift isn’t that agents can write code. It’s that they can: • Coordinate in teams • Run for days • Ask for help when stuck • Ship entire systems with checkpoints That changes how you build. If you’re still thinking in “feature tickets,” you’re behind. Builders in 2026 will: ✔️ Define the problem clearly. ✔️ Decompose it well. ✔️ Set guardrails. ✔️ Let agents execute. ✔️ Review what matters. Speed won’t come from typing faster. It’ll come from designing better workflows. The gap won’t be technical skill. It’ll be orchestration skill. Builders who learn that now will compound. The rest will think this is just “autocomplete 2.0.”

37

29 Comments

LLMs are good at generating code that they're bad at modifying later. It'll be fine for a while, but as the code base grows beyond the usable context limit - orders of magnitude smaller than advertised - the unnecessary complexity, duplication and lack of separation of concerns will kick in. Anyone claiming otherwise has evidently never gone beyond that point using the tool. Or just hasn't noticed all the stuff that's broken. They're telling you how to run your marathon based on their experience running the 400 m.

108

23 Comments

Update on new Listify features, and experiments with Wap and ChatGPT 5: Warp is running a $5 ChatGPT promotion (https://www.warp.dev/), integrated with their Agentic terminal/dev environment. I've been using Gemini Pro and the Gemini CLI so far to build Listify (https://lnkd.in/gmU5YdUE), my hobbyist AI+flutter powered list management app, but thought I'd give ChatGPT 5 a chance for comparisons sake and to see how smoothly the LLMs could hand off the codebase. Observations: - Handoff was seamless, GPT5 had no problems understanding the code base and working with it. - I really didn't see a big difference in coding capabilities between GPT 5 and Gemini Pro 2.5, but it's true this is not an extremely challenging app. - GPT5 has much better testing discipline than Gemini Pro; it suggested adding a number of tests whereas Gemini never even considered to offer. - The Warp terminal is definitely better than Gemini CLI due to the greater power that Warp offers; e.g. formatting is richer, you can upload images and easily specify files as context, and the collapsible sections provide a better experience for looking at diffs and log output. - I am still uncertain if the agentic environment of the future will look more like a browser, an IDE, or a terminal. Toggling back and forth between terminal mode and agentic mode in Warp is awkward. While you can give commands like "git push" to the LLM instead of the shell, it is terribly slow to respond. Since I last wrote, I've added the following features: - an AI-powered autosort and group feature. For example, if you have a shopping list, it will be break it into labeled sections like "produce", "dairy", "snacks", where each section contains things that are likely to be nearby in the store. This makes shopping faster. - an "Export to Google Tasks" feature to the app for people who like tahe scanning/parsing features of Listify but want to live in Google Tasks. (The backend but not UI integration was done so let GPT5 take over the UI. - more optimistic updates to make the UI feel snappier. - I also added a feature that lets you add a section header by long pressing on add item. Obigatory plug: Please LMK if you'd like to be a tester for the Android version of Listify, I still need to get to 12 testers to get it listed on the main app store for convenience of download. The web link above gives you an idea of the feature set but the mobile experience is much better. I still also need to get the iOS version up but haven't jumped through all the certifcate hoops.

3

Many of the teams I talk to are either all-in on real user monitoring or overly reliant on synthetic. But frankly, both approaches leave blind spots. RUM tells you what actually happened to your users, while synthetic tells you how your pages perform in a controlled environment. The real magic is in how you use them together. That’s what I’m diving into with my #webperf BFF Cliff Crocker in this upcoming live session 👉 https://lnkd.in/devccuqw We’ll be getting into: → When to use RUM vs synthetic → How synthetic and RUM fit into your CI/CD pipeline to catch regressions before and after they go live → What's possible with the latest diagnostic capabilities in RUM (Long Animation Frames, LCP subparts, INP attribution, and more) → How to connect your performance data to real business impact, not just dashboards If you’ve ever looked at your performance data and thought, "Um, cool... but what do I do with this?” you’re not alone. This session is all about turning that confusion into clarity. Join us! https://lnkd.in/devccuqw #ux #o11y #sitespeed #pagespeed

20

1 Comment