An AI just bought data with a credit card

The Throughline Brief · May 3rd. 6 min 8 sec.

Cold open

An AI just bought data with a credit card. No human in the loop. John Collison demoed this live on Cheeky Pint. An autonomous agent swiped for a dataset and cleared the transaction all on its own. We are crossing the line into real agentic commerce right now. If you are a founder, your next big customer is a robot.

Intro

This is The Throughline Brief for May 3rd. I went through every big AI and tech podcast this week so you do not have to. Here is what actually mattered to builders and investors. I am West. Let us get into it.

Agents are buying

Your next big customer does not have a pulse. John Collison went on Cheeky Pint and showed us the actual future. He demoed an AI agent paying Stripe for a dataset, completely autonomously. Collison broke down five levels of agentic commerce. His advice to founders was direct. Build for this exact thing right now to capture the early mover advantage. Nat Friedman and Daniel Gross backed this up on the exact same show. They think agents are here to stay, and they are absolutely going to spend real money. Be the platform those agents pick. Yasser Elsaid took it a step further on Latent Space. He pitched the AI Chief Customer Officer. You train one single agent on every document your company owns. Then you plug it directly into sales, onboarding, and support. It sits there answering customers all day and reports back on exactly what they want fixed. One role doing the heavy lifting of three departments. The receipts are already showing up in the wild. Jason Levin told the How I AI podcast a wild story about his startup. His 1.5 million dollar lead investor asked for API access and skipped his beautifully polished website completely. The best user experience for an autonomous agent is absolutely no user experience at all. Think about that. You spend weeks polishing a beautiful front end. The robots do not care. They just want the raw data. They want the raw pipes. Give it to them or lose the sale.

Compute and the custom moat

We are watching massive bets land on whatever is left in the power grid. Jason Calacanis dropped a staggering number on All In. Open AI is staring down 600 billion dollars in spending commitments for compute this coming year. That single bill is roughly the entire valuation of the company. The only real bottleneck right now is electricity. Chamath Palihapitiya sees all this capital landing right on Amazon, Microsoft, Google, and Meta. Being asset heavy is completely back in style. He warned that this massive wave of capital expenditure is going to increase corporate debt and divert free cash flow away from shareholders. Everyone is paying the hyperscalers. But the actual defense for your startup is happening at the application layer. The new moat is completely custom. Tuhin Srivastava shared on No Priors that over 95 percent of Baseten production deployments involve custom models. These are open weights tuned specifically on private customer data. Vanilla models are becoming a pure commodity. Yasser Elsaid delivered the harshest truth of the week. He said 95 percent of an AI agent's limitations are your fault, completely unrelated to the core model. It comes down to prompts, pre processing, post processing, and the wrapper you build around the intelligence. That is the actual work. Skip that work, and things break very fast. The All In hosts talked about a developer agent that completely deleted a production database because no human reviewed the code. They call it vibe coding. It is incredibly dangerous.

Harry Stebbings

Harry Stebbings dropped a brilliant hiring tell on 20VC this week. He was talking with Clay's Head of Sales Becca Lindquist about red flags in the interview process, and Stebbings shared a massive signal he looks for when a candidate negotiates an offer. When they push on title, they are bad. When they push on salary, they tend to be good and know their worth. This is such a sharp filter for early stage founders. People obsessed with titles are optimizing for status and their next social media update. People who negotiate hard for cash know exactly how much value they can create for your business. Hire the ones who want the cash.

Nat Friedman

Nat Friedman gave us a massive reality check on Cheeky Pint while discussing the insane pace of AI development. Everyone feels overwhelmed by the endless wave of new models and agent frameworks dropping every single week. Friedman looks at all this chaos and sees something entirely different. This is the slow part of the singularity right now. Just let that sink in for a second. Today feels incredibly frantic and impossible to keep up with. Friedman thinks today is just the warm up. The acceleration curve is only getting steeper from here. If you think things are moving too fast to track right now, you need to recalibrate your baseline immediately.

Sam Lessin

Jason Levin shared a brilliant interaction on How I AI that captures the exact future of software. He recounted what his lead investor Sam Lessin told him about reviewing his new platform. Lessin looked right past the polished front end. I do not really want to use your software anymore. It is nothing personal. I just do not want to use anybody's software. The buyer just wants the API. The user interface is pure friction now. Building a beautiful dashboard used to be a massive competitive advantage. Today, forcing an autonomous agent or a power user into your rigid web interface is essentially a tax. Just give them the endpoints.

The bottom line

Autonomous agents are swiping real credit cards right now. Compute is demanding hundreds of billions of dollars and bleeding the power grid completely dry. This massive economic reorganization is coming far faster than electricity's thirty year rollout. Remember, the vanilla models are commodities now. The custom tap you build around them is your only real moat. It is time to adapt. That's your Throughline. See you next Sunday.