Skip to content
CampeloLabs

Blog

Practical AI guides for founders

Short, useful reads on growing your startup with AI, safely. From the team behind AI Operating System for Startups. See also AI startups from Y Combinator.

A founder turning away from a wall of glowing leaderboard rankings to read one small private scorecard in her hand

September 29, 2026 · 11 min read

AI Benchmarks: What the Leaderboard Hides

Founders reach for AI benchmarks at one moment: when they have to pick a model and want a number that makes the decision for them. It will not. Vals AI co-founder Rayan Krishnan says Llama 4 underperformed on his team's held-out private benchmarks while looking excellent on the public ones, where the questions and rubrics are open source. Benchmarks also saturate, go stale, and have shrunk to small samples with long rubrics. Here is how to read a leaderboard, and the only test that picks your model.

Read more →
A founder at a control desk opening one throttle wide while metering several smaller ones

September 28, 2026 · 13 min read

Tokenmaxxing: What It Is, When It Pays Off

Tokenmaxxing is the practice of deliberately maximizing what you spend on AI tokens on the logic that the output is worth far more than the bill. Large companies spent 2026 walking it back: Uber burned its annual AI budget in four months. USV general partner Mike Mignano argues the retreat is correct for incumbents and wrong for startups, because the math genuinely differs at 20 people and at 50,000. Vals AI ran the experiment on comped access and burned $1.5 million worth of tokens in a month, 10 times its salary bill. Here is where the line actually sits, and the test that tells a real edge from burn.

Read more →
A founder at a workbench choosing between a glass case holding a glowing research artifact and a working machine wired into a factory floor

September 27, 2026 · 15 min read

Neolabs: What They Are and Which Ones Last

A neolab is an AI company funded on a research thesis rather than a product, and the category now holds an extraordinary amount of capital with very little shipped. Factory co-founder and CTO Eno Reyes thinks 80 to 90% of them stop making sense as independent businesses inside 18 months, and his reason is not that the research is bad. It is that most of them are not attached to a workflow that survives the next model. Here is his test, why verifiability decides the winners, and what it means if you are choosing between building a lab and building an application.

Read more →
A founder handing a finished glowing component to a crowd of builders while keeping a more intricate mechanism on the workbench

September 26, 2026 · 14 min read

Open Source Business Model: What Actually Pays

Give the code away and you have a distribution channel, not a business model. The business is whatever you build beside it, and founders who skip that distinction end up with a popular repository and no revenue. Nvidia agreed in September 2026 to buy Hugging Face for about $12.9 billion, a company that got there by giving its most famous asset away. Hugging Face co-founder Thomas Wolf, Keras creator François Chollet and OpenClaw's Peter Steinberger have between them run this play and paid for it. Here is what open source buys, what it charges, and the four ways it converts.

Read more →
A founder assembling a machine from modular parts around a glowing core, next to a stack of finished units

September 25, 2026 · 16 min read

AI Agent Infrastructure: Build or Buy

Nobody sits down and decides to build agent infrastructure. It accumulates: a retry here, a queue there, a sandbox somebody stood up for a demo. Then the work starts running while nobody is watching, and the pile becomes the product. Factory CEO Matan Grinberg expects 90 percent of tokens to be asynchronous within 12 to 24 months. Here are the seven components that sit between a model call and a finished task, what each one has to do once nobody is waiting, and which ones are worth building.

Read more →
A founder watching a row of machines pull products off a shelf and pay at a counter with no other people in the room

September 24, 2026 · 12 min read

AI Agent Economy: Defaults Are the Market

When an agent picks the database, the email API and the transcription model, the winner is not the best product. It is the best represented one. Agents choose a default from what they read, then stop looking, which is why Y Combinator's CEO watched his own coding agent reach for a model OpenAI has since scheduled for deletion. Here is what actually sets the default, the one number that proves the buying has already moved, and where a challenger gets in.

Read more →