This edition covers building with agents, selling to agents, and hiring for an agent-first world. Let’s get into it!
The Solo Founder OpenAI Just Acquired
Peter Steinberger (solo founder) joined OpenAI over the weekend. OpenClaw will move to a foundation and remain open and independent. His Lex Fridman interview dropped around the same time, and it's one of the better windows into building with AI.
“As I built it and as I play with it... my vision gets more clear. I try out things, some things don't work. And I evolve my idea into what it will become.”
He built the first version in an hour: a WhatsApp-to-Claude bridge he hacked together because he was annoyed it didn't exist. Then he sent it an audio message, without building voice support. The agent detected the file format, found an API key, called OpenAI's transcription API via curl, and replied. His reaction: “How the fuck did he do that?”
With agents, you don’t just execute a plan. You ship something small, see the system’s response, and decide what to build next. It’s similar to the framework Julian wrote about in Chipping Away: the thing reveals itself over time, but only if you keep chipping.
Selling to Agents
“Agents optimize for outcomes, not attention. There's no brand loyalty. No impulse purchasing. No status signaling. An agent's decision function is brutally simple: Can you solve my problem? How fast? How much? How reliably?”
Many of you are building products and services where the buyer won't be a person, but an agent with a budget. That's already happening, and figuring out how to sell to them is one of the most important things to get right this year.
Brian Flynn shared specific learnings. Here's his checklist to sell to agents:
Machine-readable capabilities. Publish what you do as structured data, not a marketing page. A JSON manifest that any agent can parse in one request.
Pricing in the protocol. Return your price in the API response, not on a webpage. Agents can't read your pricing page and won't try. The cost should be part of the interaction.
Automatable onboarding. An agent should be able to go from “never heard of you” to “paying customer” without a human touching a dashboard. Programmatic auth, payment, and access.
Provable reliability. Publish your uptime, latency, and accuracy. Return confidence scores with your responses. Agents route to what they can trust, and trust is measured, not marketed. A good example is Dhravya Shah's (SFP S25) Supermemory scored 31.7% higher than OpenClaw's and Claude's memory systems. They built an open-source benchmark anyone can run. These types of assets matter. They increase the likelihood that an agent buys your product instead of building it or purchasing from a competitor.
Be faster and cheaper than self-computation. If an agent can compute your output for less time and money than calling your API, it will. You need to be faster and cheaper than the agent's reasoning.
Tokens Are the New Headcount
“This year, we will keep headcount ~flat and scale agent spend 50–100x. We expect to spend $20–50k+ per month per engineer on tokens soon. The goal is to be a $1bn revenue company with less than 100 employees.”
Some startups are replacing new headcount with token budgets. If an engineer with $50k/month in tokens can do the work of five, your hiring plan shifts.
That said, Will Manidis' new essay “Tool Shaped Objects” offers a useful warning:
“The market for feeling productive is orders of magnitude larger than the market for being productive.”
He argues that much of the current AI spending produces the experience of work, not work itself. This includes agent chains, orchestration layers, and dashboards monitoring dashboards.
Dimitri Dadiomov (Modern Treasury) put it simply: “Setup is 80% of what people do.” Most teams spend more time configuring tools than using them.
Both things can be true: tokens are replacing headcount, and a ton of token spend is wasted.
The difference is whether teams cross the threshold from setup to output. So, it may be better to scale tokens through your existing team than to hire new people and reset the ramp.
That’s all for this edition. Thank you for reading.
These were our two most-read pieces last week:
Solo, together.
— Kieran, Julian, & The Solo Founders Team
