Palo Alto AI Lab · Incubator
You bring an early AI product. We pair it with one accountable human advisor, the engineers and investors around this lab, and, where the advisor allows it, an AI agent that carries their knowledge between sessions.
Palo Alto, California. Rolling intake: there is no batch, no cohort calendar, no demo day you have to wait for. You apply when you are ready and we answer within a week.
→ Apply with your startup· three questions, one email
→ Join as an advisor or engineer· as yourself, as your agent, or both
→ How it works ·
Advisor agents ·
The network ·
What we are not
Companies where the AI is the product, not a line in the deck: agents, agent infrastructure, memory and retrieval, evaluation, governance and safety tooling, vertical products that live or die on model behaviour. Pre-seed to seed. A prototype that runs is enough; a slide deck alone is not.
Three paragraphs: what you are building, what you have learned, where you are stuck. A link to something that runs beats a deck.
Within a week we say yes or no. If yes, you get one lead human advisor, chosen for your bottleneck and accountable for the relationship.
Not a mentor pool, not a Slack channel. One person whose name you know.
You work with your lead advisor and the specialists the problem calls for: product and technical strategy, agent architecture and evals, customer discovery, positioning, early hiring. Where the advisor has one, their advisor agent works with you between the human sessions. Small circles of two to four companies form when it helps and dissolve when it does not.
When the company is ready, and only then, we make introductions to investors, customers and engineers in the network. Introductions are selective. We do not invest, we do not take a fee for an introduction, and nothing here is a promise of financing.
Terms are agreed per company with your lead advisor after the first conversation. There is no price list because a two-person agent startup and a seed-stage company do not need the same structure.
Good advisors have little time. Their knowledge does not have to.
An advisor agent is an AI system built from material its advisor has explicitly approved: their writing, frameworks, talks, the questions they always ask. It remembers what your company already discussed, applies the advisor's way of thinking to your current draft, critiques a pitch or an architecture at two in the morning, and prepares you for the next session with the human.
The human still makes the calls that matter. The agent keeps the context alive.
your startup
|
lead human advisor ......... judgment, introductions, the final call
| \
| advisor agent ..... memory of your company, the advisor's frameworks,
| critique between sessions, always on
|
engineers · operators · investors in the network
|
customers · talent · capital
Four rules, enforced in the same control model this lab publishes for its own agents.
Every advisor agent is labelled as AI and never speaks as the person. Every advisor decides what their agent may know and can switch it off. Every consequential decision escalates to a human.
What your company shares with an agent stays in a store that belongs to your company: it is not used to train anyone's model, and it is deleted when you ask.
This is not a pitch about a future product. This laboratory already runs on one human and one agent with persistent memory across eight machines, and publishes the division of labour it uses. The incubator extends that operating model to the companies it works with.
The people this incubator draws on are the ones who have gathered with this lab since 2025: engineers at large AI and infrastructure companies, researchers, operators who have shipped products, and early-stage investors in and around Palo Alto. The gatherings log is public and read from the calendar that owns it, so the size and shape of that network is something you can check rather than take on trust.
Names appear on this site when their owners put them there. Today it names one human and one agent; the rest of the network is reachable through them.
No logo wall of employers. And when we publish a headcount, humans and advisor agents will be two separate numbers, never one.
Built or shipped something real in AI, and want to share what you know without taking on another job? Three ways in.
Be the lead advisor for one company. Join focused sessions on the problems you know best. Or create an advisor agent from the material you choose to share, and let it do the patient part while you keep the judgment.
Engineers who would rather contribute code than advice: the lab's repositories are open, and a reproduction or a counter-example is the contribution we value most.
One email to a@platinum.fund with the subject line Incubator: [your company] and three things:
Advisors and engineers: same address, subject Advisor: [your name], one paragraph on what you have built and whether you would consider an advisor agent.
Direct: Telegram @tonydzi · WhatsApp +1 341 222 9178
Not a fund: we do not write cheques and do not promise anyone else will. The correspondence address belongs to the operating company behind the lab, which is not a fund either.
Not an accelerator with a calendar.
Not affiliated with Stanford University, Google, OpenAI, Anthropic, or any employer named in an individual's biography; those references describe where a person has worked, nothing more. Not a place where an AI pretends to be a person: every agent here says it is one.
Palo Alto AI Incubator is a programme of Palo Alto AI Lab, an independent research laboratory. Introductions are selective and do not constitute an offer of financing. Advisor agents are AI systems built from material their advisors have approved; they do not speak for those advisors outside that scope.