How to start an AI company from zero, from a three-time builder
Adam Robinson · Jan 2026 · 2 min read
If I had to start a brand new AI company from absolute zero, knowing everything I know now from building 3 software companies to over $35M in combined ARR, this is exactly what I would do:
— This is a long post, buckle-up 1200 words [bookmark this] —
Why Most Founders Fail
Founders fall in love with their idea instead of their customers’ problems.
They build something for 12-18 months before talking to anyone. That’s why 90% of startups don't make it to $1M ARR because all they had was an idea, locked themselves in a room with their co-founders and then showed up waiting for the world to buy it.
But that’s not how this works.
——
Discovery Before Everything
Before writing a line of code, I'd commit to 50 discovery calls minimum. Your only goal is to find people who will pay you for something that you haven't built yet, and I'm talking about actual payment, not just “Yeah, that sounds interesting.”
My favorite story about this is Barkbox, which is now publicly traded.
Matt Meeker owned a Great Dane and lived in NYC. All the dog stores sold tiny toys for tiny dogs. (This was 2010, pre-Amazon Prime)
Matt took a Photoshop image, showing a monthly delivery of 5 large toys for $29. He showed it to every large dog owner. When they said they were interested, he swiped their card for $29 with a Square reader.
He got 50 purchases the first day before he had a website.
——
Getting Those First Calls
I'd start with my network. People who match my ICP.
Andrew Mewborn uses this message for his company: “Hi [first name]. I was at Outreach, and now I'm building a Highspot and Seismic competitor, 3x cheaper and 10x more user-friendly, and we have around 4,000 active users that love us. Are you be open to giving feedback?”
He gives them with a $100 gift card (and yes, this still works)
You could also offer paid advisory roles to director-level people in your ICP or go to industry events with just a notepad.
Alina Vandenberghe, founder of Chili Piper (now $40M ARR), would go to sales meetups and find the most influential people. She'd approach them with a drawing of her product and ask, “Would you buy something like this? If not, what would we have to change?”
How do you find influential people? She’d look for who’s getting hugged. The person receiving the hug is the one you want to talk to.
——
What I’d Ask On The Calls
I'd lead with my sizzling one-liner that makes people's eyes light up.
For RB2B, it was “person-level website visitor identification.” Before us, there was only company-level. When people heard person-level, they got excited.
Then I’d ask 6 questions:
1. Do you understand the problem I'm trying to solve? 2. Do you have this problem? 3. How are you solving it today? 4. Would you pay for this solution if I could get it to you tomorrow? 5. What else would it need for you to pay for it?
I'm thinking of charging X. Would you pay that?
The goal is to understand their pain so deeply you can solve it better than anyone.
CONTINUED👇
Written by Adam Robinson and shared publicly; kept here in full, in his memory.