10 lessons I've learned about handling React state over the last 7 years... (thread) #react #reactjs

Just met with someone who is unhappy with AI. Me: "What are you using?" Them: "Copilot with Sonnet" Me: There's your problem.

Just met with yet another developer won’t use AI because he “enjoys the act of typing code” and “can’t trust it”. I get it. I felt that way too. Problem is, that ship has sailed. AI isn’t not optional anymore. If you’re unwilling to do agentic coding, you’re a liability.

It's wild how much more enjoyable my work day is with Opus 5.5 vs previous versions. The output prose feels human again.

Just told Opus 5.5: "Clean up the prose in this doc. It was written by an older Opus version." So. much. simpler. Here's some before -> after results: What is moving, and what isn't -> Scope Completed pre-work -> Status They are not ours to edit -> We can't edit

Opus 5.5 just released. Comparable to Fable 5.1 and 40% cheaper to run?! Wow. The pace of improvement is wild.

Jev is compelling for making quick, cheap decisions. I specify a set of potential options, and Jev instantly provides its answer and its confidence level. Way faster and cheaper than an LLM. This vid uses Jev to quickly classify emails by category, identify scams, and rank emails by urgency. youtube.com/watch?v=o1CogAtW…

No. It’s a boring implementation detail. In 2026, I care about results.

I’m eager for the “post AI” dev world. What will software devs talk about after AI is a boring thing we all use and take for granted?

“A luxury, once sampled, becomes a necessity.” AI is the latest example. First we liked it. Now we require it.

In a family, communism works. Each person works and is “paid” according to their abilities and needs. This works because we love our family enough to accept the lopsided transaction. And a family is small enough that centralized planning is practical. Outside of a family, communism fails. Why? The free rider problem. Humans are natural energy savers. They need incentives to work. If I can’t keep the fruits of my extra effort, then I’m far less motivated to work hard. This is why East Germans tried to sneak into West Germany. This is why no one wants to immigrate to North Korea. Incentives matter. In a family, mercy for the freeloader is expected. In business, mercy for the freeloader is cruelty to the hard worker.

Finally!

Just spent the last week preparing a detailed plan to migrate a team from ADO to GitHub so we could use Copilot code review. Woke up today and Copilot reviews for ADO are now available in preview. Trying to decide if I should celebrate or cry

One of AI's most annoying tendencies is needless specificity. AI creates docs and comments with specific counts, even when the specific number doesn't matter, or changes often. I regularly remove these values to streamline reading and maintainance.

2027: Senior devs will get paid well to "harden" apps written by amateurs. Senior devs can see what's broken, notice what's needlessly complex, and have a rich vocabulary for prompting effectively. And they'll be wildly effective because they’ll use AI to accelerate their cleanup work.

if something is hard, do it more often

There's an epidemic of teams drowning in needless process. Why? A lack of trust. Trust takes a long time to build, and it's easily broken. The moment trust is broken, the org wants to add "safety" through manual inspection, more environments, external approvals, less frequent deploys, and more planning meetings. But, these things are generally counter-productive. They create a false sense of control and safety, and ironically often reduce stability and quality. The actual solution? Automate quality checks and deploys. Merge daily and use feature flags to merge incomplete work. Ship often. Implement observability so you know immediately if things go sideways.

My grandma passed away this year. Today my mom handed me a card from my grandma. She made me a birthday card months in advance, knowing she wouldn’t be here. One last unexpected message from her was a wonderful gift. So thoughtful. That’s love.❤️

If I don’t know what “right” looks like, I can’t tell when AI does it wrong

Problem: AI is often a sycophant. It praises in an insincere way. Solution: Don't ask "Sound good?". Ask “What are the tradeoffs?”