Spent years leading product teams. Now I enjoy building alone, orchestrating AI to help businesses. Still enjoy hacking organic search with grounded content.
Los Angeles, CA
Joined March 2010
- Tweets2.2K
- Following1.1K
- Followers824
- Likes555
Gev Balyan retweeted
Alright, I've made the shadcn/ui skills really good at 𝙳𝙴𝚂𝙸𝙶𝙽.𝚖𝚍.
Ask for a design system and it picks a preset, then builds everything: colors, tokens, type scale, elevation, styles the primitives and blocks.
Start with a solid foundation and build something unique.
8/ Options come bare. Claude marks a recommended choice and explains what each one will actually do. Codex lists options and leaves you to rebuild the context.
9/ Auth flips between chats. Claude CLI is authenticated in one Codex chat and "not authenticated" in the next. Push back, and it admits it was wrong.
10/ None of this is about intelligence. I need both models: one builds, one tears it apart. It's also a lot of fun.
5/ No next step. Claude ends with the best next move and pre-fills it in the input box: Tab, Enter, done.
Codex says "done" and leaves you to figure out what's next. My global AGENTS.md tells it to "offer a useful, specific next step". Still nothing.
6/ Decision points are buried. My AGENTS.md says to ask before pushing.
Claude: "Ready. Want me to push?" with "push" pre-filled. I don't reread anything.
Codex doesn't ask. I have to read the whole message to work out whether it's even ready.
1/ I'm a product guy, not an engineer. I can't review AI-written code. Yet I ship production payments code for real clients.
The trick: two AIs check each other. Claude Code writes, Codex reviews.
Last week I flipped them. The models are close. The apps aren't. 🧵
3/ Codex the model is genuinely smart. The app is sleek and feels native.
But with Codex as the author, it took me two days to ship my first ready PR. Every reason was app experience, not model quality:
2/ Codex is what makes it work. It reviews every plan, spec and line Claude writes and catches what Claude misses. Without it, my quality would collapse.
When my two $200 Claude plans ran dry, Codex became the author and Claude the reviewer.
Gev Balyan retweeted
Claude has discovered a previously unknown enzyme system hidden in the DNA of bacteriophages. Beside the enzyme’s gene sits a long array of repeating DNA—a structure that looks somewhat similar to CRISPR.
We don’t yet understand what this system does, but only a handful of known systems share its features, and all of them are able to cut, copy, and paste DNA. Historically, the discovery of such programmable systems has helped revolutionize medicine. CRISPR, for instance, is now the foundation of genetic medicines. But it will take much more work to learn what this system does, and whether it can be put to similar use.
Read more: anthropic.com/news/claude-di…
Russian espionage, as always 🤯
We're publishing our most detailed threat intelligence report to date.
It covers how people tried to misuse Claude—for cyberattacks, influence operations, surveillance, biology, and building weapons—and how we found and stopped them.
We disrupted every operation in the report, and used the lessons from them to strengthen our safeguards. Where appropriate, we also shared what we found with authorities and other AI companies.
These cases are not typical: we’re highlighting some of the most sophisticated misuse we’ve seen. But they’re especially important to discuss, because they show us where AI misuse is headed, where our safeguards work, and where they need to improve.
We’re publishing this report so others can spot the same activity on their own platforms, and so we can give the public a clearer view of how emerging threats develop.
Read the report: anthropic.com/threat-intelli…
Today I downgraded my Claude $200 subscription. I will still use it, but much less, since Codex started to understand me better, do things much better, I mean much much better. And I am sad today :(
Gev Balyan retweeted
I'm officially done reading AI-generated code.
It's been two weeks since I looked at any of it.
I think the IDE is officially on its way to the graveyard. The job is no longer about "writing code," so we need new tools that better reflect this new reality.
While reviewing the code, I realized my only complaints were stylistic, and I wasn't finding any obvious bugs anymore.
The more code I generated, the harder it became to keep track of every line. I found that my time is better spent designing ways to verify that the overall system works than looking at the code.
State-of-the-art coding agents are better at writing code than I'd ever be, and I'm going to stop pretending otherwise.
I still think these coding agents can't go too far without an experienced human guiding them, but we're past the point where we need to check every line of code.
I also find GPT 5.5 so reliable and accurate. It usually does its job much more grounded compared to Claude Opus, and Codex for desktop app so fun to use! Good job guys, seriously. I never could imagine I will come back to GPT, but it happened😎
Do we still need the Karpathy Skill to be installed if Opus 4.8 is out and kind of promises to take similar or close to Karpathy approach while reasoning on coding?
