WATCH THE VIDEO. LISTEN TO THE BASS. FEEL THE SPEED. 3,000 TPS is the new standard and we are the only ones doing it at @Nullshot_ai

More likely story: The "jailbreak" wasn't super serious (a situation anyone who has ever received bug reports is familiar with), Anthropic thought the demand to halt model was absurd, and Fed Gov used opportunity to punish and humiliate Anthropic for the prior sins of not bending knee. Anthropic has more credibility on such topics than Washington

This is a huge moment for AI. What I would describe as beyond intelligent, just got shut down. I had ~5 sweet days with it. It's pure madness. And let this be a reminder that models must be open.

Deodorant sales going up?

Interactive product timelines, made simple.

Most founders pivot too early. Got a rule: every new idea goes in the back pocket for three months before I do anything with it. Most of them die on their own. The ones that survive are usually worth chasing.

Been realising how important closed loops are. Set the AI up properly, it runs, it finishes, it reports back. Done. When the loop isn't closed, you're babysitting. Which is just doing the work yourself with extra steps. Working towards more loops that actually close.

No one is talking about it?

Let’s Talk about Agents x.com/i/broadcasts/1wGWjjByd…

The longer I do this, the more I think data is overrated. Conversations are where the actual shifts happen.

I can plead not guilty on this one.

Someone at the @cursor_ai meetup last week asked how I trust AI-generated code I haven't read line by line. Honest answer: you don't trust the code. You trust the loop. Different models reviewing each other catches more than any single pass would.

If you haven't heard it yet, $XAVA is building at xavadao.com

Who's the best builder you follow on here who has under 1000 followers? Drop them below.

Hiring is the most undervalued thing a founder does. Everything else comes from it. A lot of founders treat hiring like an interruption, something to get through so they can get back to the real work. The real work being the product, the customers, the fundraising. Hiring is the thing in between. I think it's the other way around. The product is only as good as the team building it. The customers are only as happy as the product. The money is only as good as everything underneath it. The team is the thing that makes the rest of it work. The thing I keep coming back to is how much you can tell from the first conversation. Whatever you notice in someone the first time you meet them is usually who they are. The things you tell yourself you'll fix don't usually get fixed. They just get louder. The other one: people are best at interviewing when they're trying to get the job. Six weeks in is when you actually find out who someone is. By then you've already made the bet, and undoing it is way more expensive than slowing the call down would have been. A better process doesn't really solve this. More questions, more take-homes, more references. What actually helps is being honest with yourself about the gut feeling you had after the first conversation, and not talking yourself out of it because the CV looks good or you need the role filled. Excitement is a useful signal. Talking yourself into someone is usually a sign you've already got your answer.

Everyone in AI is saying prompting is the main blocker now. Models are good enough, the limit is humans. I don't think that's right. Anyone can build. The skill ceiling has come way down. What's actually blocking people is the cost ceiling. The price war is squeezing rate limits, tiers keep shifting under us, everyone's chasing GPUs, Codex is now ahead of Claude Code on limits... We went from "is this possible" to "is this affordable" really fast.

Easiest advice I can give other founders: talk to your users every week. Everything else gets easier once you do.

Speaking at a @cursor_ai meet-up tomorrow. First time. Been using Cursor for 15 months. Strange to be the one doing the talking now. The talk is about multitasking in Cursor, how to run multiple things at once without losing the thread. Then 15 minutes to build as many features as I can, live, before the timer goes. One thing I'm curious about. The way you use a tool when nobody's watching is different from the way you use it in front of a room. The shortcuts you skip, the prompts you half-write, the moments you sit and think, all of it shows up in a live demo. Going to find out what my actual workflow looks like. Recap to follow.

Turns out the thing that matters most when building with AI isn't what everyone says it is.

Most of my week is split between building and talking to people. Both matter. But the talking is where the shifts happen. Had one of those last week. A mentor I trust, the kind of person who's been through this a few times and sees the shape of it before I do. Lost sleep over it after. The conversation was about clarity. How well the team actually knows what to prioritize, how confident I am that we're spending time on the things that move the business, how much of that is real vs how much I'm assuming. Spent the weekend rebuilding how we're running priorities this week. New tooling, new rhythm, clearer signal on what each person's working on and why. By Sunday night I was grateful the conversation happened, even though it cost me a Saturday. The thing nobody tells you about running a company is that the dashboard never tells you the real story. It tells you what's happening. People tell you why. And until you understand the why, you can't actually make a decision worth making. That conversation gave me the why. The weekend gave me the how. Better Monday because of both.

How many tokens can you afford to burn before you find the right answer? Feels like that's the actual ceiling right now.