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 inspecting the internal layers of a robot foundation model while a robot arm holds a task mid-motion

October 14, 2026 · 12 min read

Robot Foundation Models Are Not Black Boxes

A robot foundation model takes an open-ended instruction, looks at what the robot sees, and outputs the actions that move the machine. That generality replaced the explicit kinematics, dynamics, and control of the classical stack, and the easy assumption is that the trade bought you an unreadable model. The first framework for interpreting and steering these models through their own internal representations, out of Claire Tomlin's lab at Berkeley, shows otherwise: project the feed-forward activations onto the model's own vocabulary, cluster them by concept, and you can turn behavior up or down at inference with no retraining. Here is what that means for the questions you should be putting to whoever sells you the model.

Read more →
A founder reviewing retailer deductions and demand planning for a consumer brand on an agent dashboard

October 13, 2026 · 14 min read

AI in CPG: What Consumer Brands Actually Buy

Search for AI in CPG and page one sells you shopper personalization and concept testing. The contracts consumer brands actually sign are for the back office: cash application, deduction disputes, trade promotion math, and how much raw material to buy. Confido built an agent workforce for that work, announced a $55 million Series B led by Insight Partners, and counts divisions of Mars and Unilever as customers. The founders also spent a year in what they call pivot hell before finding the idea, then validated it with cold emails that strangers at Coca-Cola answered with two free hours. Here is what AI in CPG means inside a real brand, and why an industry that does not buy software does buy agents.

Read more →
A founder reviewing scoped access grants for a fleet of software agents on a permissions board

October 12, 2026 · 16 min read

Agent Identity: The Million-to-One Problem

Most founders meet agent identity by accident. Someone wires an agent into Slack and a production database to unblock a workflow, it borrows a human credential or a shared service account, and nobody revisits it. Howard Ting has worked in identity since 2000 and now runs Opal Security, and his argument is that the agent problem is not a harder version of the human one, it is a different order of magnitude: humans get a permission for 30 days and use it a thousand times, while an agent might get access for one minute to do one task. Here is what that multiplication does to your access model, and the six fixes to make before you have agents in production.

Read more →
A founder reading a power energization schedule while modular fuel cell units and a data center hall rise behind it

October 11, 2026 · 16 min read

Data Center Power Is the Binding AI Constraint

Ask a founder what gates their AI roadmap and you hear chips, money or hiring. Ask the people building the physical layer and you get one answer, and it is electricity. K.R. Sridhar co-founded what became Bloom Energy in 2001, after running NASA-funded Mars research at the University of Arizona, and he told 20VC that a greenfield data center takes 12 to 18 months minimum to build while his company can stand up the power faster than the building goes up. That inverts the usual mental model, in which power is a utility bill and GPUs are the hard part. Here is what data center power actually constrains, and the five planning assumptions it should change.

Read more →
A founder aligning a model layer, a software layer and a silicon wafer into one stacked column

October 7, 2026 · 12 min read

Hardware Software Co-Design: AI's Real 100x

Ask where AI performance gains come from and you get one of three answers: better chips, better kernels, or better models. Dylan Patel, who founded the semiconductor research firm SemiAnalysis, told Sequoia the question is wrong. Three layers each improving 2x separately give you 8x. Tuned against each other they land closer to 100x, and that is why the same model can be excellent on one vendor's silicon and mediocre on another's. Here is what hardware software co-design means in practice, and what it changes whether you build above the model layer or below it.

Read more →
A founder handing a scoped spending mandate to an AI agent at a payment terminal while a credit card sits unused

October 6, 2026 · 14 min read

Agentic Payments: The Agent Pays, You Still Pick

Most agentic commerce demos are built on the assumption that an agent will decide what you buy. Max Levchin, who co-founded PayPal and now runs Affirm, and Alex Rampell, who co-founded Affirm with him and is now a general partner at a16z, spent an hour on payments and closed on the opposite: taste stays human, and agentic payments is the part that gets delegated. Stripe's Will Gaybrick agrees on the direction and adds the uncomfortable part, which is that the first rails shipped in March 2026 and he still calls the field short of primitives. Here is where that leaves a startup.

Read more →