Looking forward to meeting some of you at OpenAI DevDay today!

This is one of the secrets I've mostly kept to myself. I've been doing this with LatchLoop since February. Its part of how I'm effectively using about 30 million tokens per day now. One problem with Cursor's implementation from this video though. These plans shouldn't be

If your agents haven’t made more commits than you have this year, I’d encourage you to evaluate your workflow. Next year this time people will be like, “What do you mean you wrote a line of code? You actually typed it like a caveman?!”

This isn't talked about enough, but we use AI in customer support to give us more work! Our AI support agent will escalate conversations to us that we would want escalated, even though the customer didn't try to ask us for help directly. This lets us serve our customers at a

90% of my “code” is me asking background agents to do things. I almost never touch an IDE anymore. I did the math on what makes it to production and currently my background agents change 33.2 lines of code every hour in my codebase.

DHH time used screen keyboard mattress hoodie kitchenware … … car

What people don't realize is how wild the coding agent landscape is about to become. You might think, "This is impressive, but Ramp is valued at $32 billion, so it makes sense they could build this." What about the small bootstrapped team? Here's a screenshot of LatchLoop. The

It used to be controversial to state that there exist 10x engineers. People would get all worked up. I have news for you. AI has created 100x, possibly 1000x engineers. Time to update those hiring plans.

It seems that enough people now realize that software development will become fully automated this year. The future of software development is not an IDE or CLI, and what surprises me is that there aren’t more alternative UIs for directing coding agents.

What the codex desktop app will look like 12 months from now:

So interesting to see similarities between Entire and LatchLoop The future interfaces we'll use to direct coding agents are being revealed, and they look much more like this than CLIs or IDEs.

Deep research in ChatGPT now works off a todo list and shows this in a UI, similar to what coding agents have been doing for the past 10~ months. This likely isn't only about performance/UX, but also about allowing ChatGPT as an agent to perform more actions during each of

GPT-5.3-Codex-Spark is now in research preview. You can just build things—faster.

jack friks a lion doesn't concern itself with token counting

Of course that's your contention. You're a first-time SaaS bear. You just got finished listening to some podcast, Dario on Dwarkesh, probably. Now you think it’s the end of white collar work and seat-based pricing is screwed. You're gonna be convinced of that til tomorrow when

Noticed we had 120k impressions over 2 days for a phrase in Google Search Console. We ranked position 2. We got 1 click... It is an old article. Human written. Search console shows an 11,821,200% increase in that query's performance. Any SEOs know what is going on?

I’ve been working like this for almost exactly a year now with my own cloud coding agents. It changes everything.

This is the most important stat about the GPT 5.4 release. Everyone was impressed when 5.2 won almost 50%. Now we are already at GPT 5.4 winning 70.8%. When do we consider it AGI?

People are building agent orchestrators but using off the shelf harnesses. There is so much untapped potential in building the harness first.