Founder Topic

How are founders actually building AI startups?

The practical consensus: build the layer where your proprietary data makes you the best in the world, buy everything else (Josh Foreman's InDebted rule after 90+ days deep in voice AI). The real bottleneck is trust, not capability — TruthSystems found nobody bought AI guardrails until they sold them to the people who actually held the pain. "GPT wrapper" is a marketing slur, not an analysis — Hung Bui reframes a model call as "a function in the code that allows your platform to be intelligent." And the thing incumbents should fear isn't tech debt but organizational debt — the legacy processes AI-native companies simply never accumulate.

Half the archive's recent guests are building with or against AI — YC AI companies, a 10-year-old fintech rebuilding itself AI-native, and founders on both sides of the "GPT wrapper" debate. This hub collects what they're doing in practice, not the discourse.

Quick answers

Should you build or buy your AI stack?

Josh Foreman's honest answer: "we've done both" — early on "the humble answer is we just have no idea," so test the market while building in parallel. Where InDebted landed: build the layer where proprietary data makes you the best in the world (negotiating collection outcomes), buy the layers where you never will (text to speech).

Is being a "GPT wrapper" a real business?

Hung Bui thinks the bad rap is unfair: "you don't have to train your own models to be different." A model call is just "a function in the code that allows your platform to be intelligent" — differentiation comes from customer understanding and distribution, same as ever.

What's the real bottleneck for AI adoption?

Trust, not capability. Nam Nguyen says it was "quite obvious" — TruthSystems' early guardrail products made sense on paper but nobody bought until they sold to the people who actually held the pain. Jevon Le Roux's version at Keeyu: roll out agentic workflows five at a time with the customer watching, and expand only once they're comfortable.

Is agentic AI overhyped?

Finnlay Morcombe publicly called it "robotic process automation on the cloud" — and has since reversed: "I don't agree with it anymore. I think the opposite's now true." His read: people are "quite flippant about the current capabilities" and undervalue what's coming. Josh Foreman adds the incumbent's angle: the thing to fear is organizational debt — the legacy processes AI-native companies never accumulate.

What founders in the archive say

EP 32
Josh Foreman · InDebted

$350M Fintech Founder Steps Down as CEO to Rebuild With AI

Ten years into building InDebted into one of the biggest debt collection software companies in the world, Josh Foreman stepped down as CEO to rebuild it from its core — because the competitor he fears most is an AI-native version of himself, starting today with nothing to protect.

$12 → $1M/weekfrom the first revenue Josh remembers "like it was yesterday" to over $1 million a week today
"Now I'm so confident on the ability for the business to compound over the long term that I have to think about a world where the business exists without even me in it."— Josh Foreman, on why ten years in changed his thinking (03:34)
↗
EP 23
Finnlay Morcombe · Fluency

The 25-Year-Old Who Raised $6M From Early Facebook Investors

At 25, Finnlay Morcombe built Fluency — a platform that maps how work actually happens in Fortune 500 organizations — raised 6 million from Accel within weeks of landing in the US, and hard pivoted into the product within one week of going to market.

6 millionSeed round Finnlay raised from Accel, before he'd been in the US more than a few weeks
"we went to market with it and within one week we're like okay we're hard pivoting because the reception was completely different"— Finnlay Morcombe, on the one-week hard pivot (15:48)
↗
EP 24
Nam Nguyen · TruthSystems (YC S25)

How He Turned 3 YC Rejections into a $25M AI Governance Company

Nam's co-founder was 19 years old. Law firms told them "come back in 5 years." They applied to YC four times. When they finally got in, their $4M round filled in 48 hours. TruthSystems is now the AI governance layer sitting inside law firms in real time.

$4MSeed round filled in approximately 48 hours after YC acceptance — Sunday night to Tuesday night.
"We had firms tell us, "come back in five years. Come back when you are Microsoft-sized." When you're that young you actually don't get a lot of benefit from the imagination. We realised we had to remove imagination from the equation — less vision, but actually more of the product."— Nam Nguyen, on selling AI governance as a 21-year-old (05:23)
↗
EP 21
Jevon Le Roux · Keeyu

He Ran One of Australia's Biggest Athleisure Brands - Then Built Software to Fix Its Biggest Flaw

Jevon Le Roux ran retail brands like Surf Stitch and PE Nation before co-founding Keeyu — and he knew the product mattered when a customer, asked what she'd do without it, said "I just wouldn't come to work tomorrow."

1,000Customers during the Covid warehouse sale who paid but weren't going to get what they ordered
"there is that bit of angst that what they've ordered is not gonna arrive on time as promised"— Jevon Le Roux, on the problem Keeyu solves (01:50)
↗

Related topics