He was the founding CTDO of Kenvue, the $40B+ company behind Tylenol and Listerine.

When Kenvue split from J&J, he found it had outsourced so much software work that it no longer knew how to build.

Outsiders held what he calls “the soul of the business,” and IT had become a contract manager.

Now, when a startup sends forward-deployed engineers, his team spends its own time teaching them how the business runs.

If they never leave, Kenvue depends on outsiders again.

“If the forward deployed model has an end, I like it.

They should teach the companies how to fish for themselves, and go away.”

If you sell AI to enterprises, put two things in your proposal: what their team will run on its own, and the date your engineers step out.

In the episode, Bernardo also talks about whether the executive backing your pilot can actually buy, what makes him bet on an early-stage startup, and how he used VCs to find startups to buy from.

[YouTube] → [Spotify] → [Apple Podcast]

Also in this episode:

[46:38] How he used VCs to find startups to buy from

[47:44] What makes him bet on an early-stage startup

[58:28] Whether the executive backing your pilot can actually buy

He founded Facebook’s Applied Machine Learning team and built the AI platform behind its products.

Then he joined OpenAI, where he led Preparedness, the team that tests frontier models for catastrophic risks, and later ran recruiting and people.

I asked him what founders should do differently now that AI is everywhere.

1. Spend on tokens, not headcount

The old advice was not to raise too much before product-market fit, because the extra cash ends up in hires you don’t need.

Joaquín still believes in small teams and focus, but he thinks the share of a startup’s budget that goes to tokens instead of salaries has to grow a lot.

Because burning compute is what lets a small team test its ideas faster than anyone else.

2. Don’t do research just to impress investors

More and more startups are building research teams early, and he thinks most of them are “investor bait.”

Research that moves a field, like what came out of Bell Labs, needs years of funding with no pressure to show results, and no startup can afford that.

What a startup can do is applied research that makes its own product better, like cutting the cost of inference by 10x.

3. Stay close to the work, whatever your title

As a company grows, the founder usually ends up further from customers and from the technology.

At Facebook, he left his job leading applied machine learning to go back to writing code.

At OpenAI, when leadership offered him a much bigger role, he asked to be an intern.

He says careers aren’t ladders, they’re helixes: you keep circling back to hands-on work, and every time you return, the tools have changed.

In the episode, we also talk about why the labs warning loudest about AI keep racing to build it, and why he’d rather speed up safety research rather than slow down the frontier.

[YouTube] → [Spotify] → [Apple Podcast]

Also in this episode:

[12:00] Why careers are helixes, not ladders

[37:25] Why the labs warning loudest about AI keep racing to build it

[39:20] Why no lab can tell its investors it’s slowing down

Most people building with AI agents think more context is the fix.

Nazli Tan (cofounder & CEO) who previously led growth at Getir ($12B) thinks the opposite.

Context usually means someone writes down what a “good customer” or a “risky customer” is and the agent runs with it.

But those definitions are mostly human guesses. They’re not mathematically rigorous.

Agents should never get to define the problem. Ask one the same question twice, and you can get two different answers for which customers are about to leave.

Nazli’s answer is that the definition is already sitting in your data. Years of customers acquired, retained and lost.

Hilbert trains deep learning models on at least 3 to 4 years of that history, and those models decide who’s a good customer and who’s about to leave. Same customer, same answer, every time.

Their agents only work inside those guardrails. They aren’t even allowed to do math beyond rounding.

Why so strict? Her customers use those answers to decide where hundreds of millions in marketing go. If the answer changes every time you ask, nobody will trust it with that budget.

Today, I had Nazli on the pod and she went deeper into where agents break, why anything that touches the money runs on DL models not LLMs, and why she rejects the traditional Silicon Valley growth playbooks.

[YouTube] → [Spotify] → [Apple Podcast]

Also in this episode:

[42:19] Why “more context” makes your AI agents worse

[45:03] How to price AI when a 1% improvement is worth millions

[52:58] Why Silicon Valley growth playbooks don’t work for 99% of enterprises

That’s the gap Matt Kropp keeps finding inside the biggest companies in the world.

He advises Fortune 500 boards on AI at BCG, and lately he hears the same thing in every room.

“I’m not seeing ROI from this thing”.

Take a bank that hands 20,000 engineers a tool like Claude Code. On paper, adoption hits 70%.

Yet under 5% of them actually write code with it. Most logged in once and went back to working exactly how they always had.

On Matt’s team, write code by hand, and you’re fired.

But that’s in San Francisco.

Almost everywhere else, most software is still typed out by someone who got handed an AI and never changed how they work.

None of it is a software problem.

What actually lifts that 5% is coaching: someone who sits with each person, day after day, until using AI on real work becomes the habit and not the exception.

Getting one engineer to switch is easy.

Doing it across 20,000 is where almost everyone stalls.

In the episode, we talked about:

Coaching skeptics into real usage, why the biggest AI spenders grow the fastest, and what running a company entirely on AI agents taught him about where it breaks.

[YouTube] → [Spotify] → [Apple Podcast]

Also in this episode:

[13:35] How he cracks open an AI skeptic CEO in the first five minutes

[22:00] Why AI-native startups keep outrunning companies 100x their size

[22:40] Why “adoption” numbers lie, and what actually gets people using AI

[51:50] What running a company entirely on AI agents taught him about where AI still breaks

Zeno Rocha bet his company on one of the most crowded markets in tech: email.

He started Resend in 2023, and they’re now at 4M users and 100k paying customers.

I asked him what he saw in it.

Every other developer category had been rebuilt. Auth0 led auth until WorkOS and Clerk showed up. Heroku led hosting until Vercel.

But in email, SendGrid, which he loved, was bought by Twilio for $2B and the product decay. The rest got acquired too, and stopped improving.

His bet was care: dark mode, animated icons, details nobody in the category had bothered with. “We’re gonna out-care everyone in this industry.”

Then AI turned email into a hot market again, and what he had to care about changed.

Agents don’t care about the interface, they can’t use it! Resend’s pricing page has a slider, but an agent can’t drag it.

And those had just become his fastest-growing customers.

He rebuilt the product for a customer that can’t see: every page readable as plain text, every action callable without a click.

In the episode, Zeno gets into how out-caring an industry became his strategy, and why incumbents getting acquired is what opens the market

Presented by Ashurst Perkins Coie. Paul Navarro works with some of the best founders in Latin America, many of them guests on the podcast. I always ask my guests who I should interview next, and the only time anyone recommended a lawyer, it was Paul. Glad to have them as a founding sponsor.

[YouTube] → [Spotify] → [Apple Podcast]

Also in this episode:

[15:00] Why the giants in your category are more vulnerable than they look

[15:55] The advantage of knowing nothing about the market you enter

[24:10] What changes when your fastest-growing customer is an agent

Jumpstart is the immigration service I wish I’d had when I moved to the US. Bianca and Fabiano have helped 1,250+ founders get O-1, L-1 and EB visas with a 98% approval rate, and if it doesn’t come through, they DON’T charge you.

He refused. Selling software to law firms didn’t fix the system and incentives behind expensive legal services.

The average small-firm lawyer spends 2/3 of their time not practicing law (selling, hiring, insurance), so they bill few hours and charge a fortune for each one.

That required changing the traditional ownership of a law firm. After months of research, he found an Arizona license that lets a law firm have non-lawyer owners.

That let him split the business in two:

Manifest OS, a venture-backable technology company, sitting on top of Manifest Law, the firm itself.

Manifest OS handles customer acquisition and the back office, while its AI prepares cases for lawyers to review. Lawyers set the strategy and review cases.

The bet underneath it: AI is the leverage that finally makes a services business as good as SaaS.

When each lawyer is 10x more productive, the company stops scaling with headcount. Dan’s benchmark: the largest law firm in the world does $7B in revenue with 1,200 attorneys.

The price is software-like margins on services revenue, which bills at higher fees than software ever could.

The first proof point is immigration law, but Manifest plans to scale to all law and build a $100B business.

In the episode, we talk about his AI-Native Services playbook.

Presented by Ashurst Perkins Coie. Paul Navarro works with some of the best founders in Latin America, many of them guests on the podcast. I always ask my guests who I should interview next, and the only time anyone recommended a lawyer, it was Paul. Glad to have them as a founding sponsor.

[YouTube] → [Spotify] → [Apple Podcast]

Also in this episode:

[23:15] The ownership innovation that made Manifest venture-backable

[43:10] The case for and against the hype of AINS

[43:40] Whether AI services can hold quality and margins as they scale

Their harder problem is determining which data will actually make their models better.

José Nunes built Vetto to solve that problem. Instead of waiting for a lab to specify the data it needs, Vetto studies where a model is underperforming and designs the work required to improve it.

That changes Vetto’s role.

Most data suppliers receive guidelines from the labs and find people to execute them. Vetto works one step earlier: its researchers design the guidelines themselves.

To determine whether those guidelines work, they build benchmarks and train small models on each dataset. The results show the labs how the model performed before and after training.

If the model does not improve, the data is not good.

This ability to demonstrate improvement is also what drives the business. The research gives a lab a reason to buy before a traditional sales process begins.

Today, José is 25 and Vetto is doing eight figures a year. It tripled revenue several months in a row, with no sales team or paid marketing.

As José told me:

“What you do with the data is as important as having the data.”

In the episode, José explains how Vetto became a research partner to the AI labs and why deciding what data to collect is more valuable than collecting it.

Presented by Ashurst Perkins Coie. Paul Navarro works with some of the best founders in Latin America, many of them guests on the podcast. I always ask my guests who I should interview next, and the only time anyone recommended a lawyer, it was Paul. Glad to have them as a founding sponsor.

[YouTube] → [Spotify] → [Apple Podcast]

Also in this episode:

[03:15] His journey from a small town in Brazil to Silicon Valley

[16:20] How Vetto competes with the labs in research

[18:05] Why the most valuable part of AI data happens before collection

[19:15] How Vetto recruits researchers it cannot outbid the frontier labs for

[57:20] Why researchers hired in Brazil can perform alongside people from DeepMind

Today he’s playing the “main league” of Silicon Valley: building an AI company in SF, against the smartest, best-funded people in the world. In his words, “you’re playing against the Pelés and Kakás.”

That company became Firecrawl, the biggest web-scraping repo on GitHub. 170k stars from a team of 20.

Every founder is dealing with the same question: if AI can clone your product in a weekend, what makes yours worth using?

Firecrawl does one unglamorous job: it reads webpages for AI. It takes a page full of code and turns it into clean text a model can use. Almost every AI agent needs that and agents now browse the web more than people do (according to Cloudflare).

Nicolas and his cofounders built it without connections or a big raise. They open-sourced it, developers picked it up and it spread from there. Tobi Lütke, the founder of Shopify, invested without even taking a meeting.

In the episode, Nicolas explains why the most boring part of the stack is often the most valuable and how open source became his sales team.

Presented by Ashurst Perkins Coie. Paul Navarro works with some of the best founders in Latin America, many of them guests on the podcast. I always ask my guests who I should interview next, and the only time anyone recommended a lawyer, it was Paul. Glad to have them as a founding sponsor.

[YouTube] → [Spotify] → [Apple Podcast]

Also in this episode:

[02:55] At 12, he had to choose: drop out of school or quit soccer

[31:25] The indie hacker test: if it takes a weekend, it’s a feature

[35:45] The best product is the one your users never see

[50:05] What he means by building the eyes for AGI

Those have been the best and highest-growth years of my life. Today his episode goes live, and I finally get to tell his story, because most people don’t know the arc.

He landed in the US at 18 with almost no English and spent his first year mostly in the library, learning the language and what he calls the OS of American culture.

In the 2000s, he ran Penguin Holdings out of Miami, a circle of young LatAm founders betting on the next generation.

Alongside Wences Casares, Micky Malka and Verónica Serra, they wrote one of the earliest checks into a small Argentine startup called Mercado Libre.

Then he spent 12 years as a founder himself, building two video game studios with Latin American talent and selling to the US, back when everyone said it couldn’t be done from there.

The first was acquired by Activision after a wall of nos, and the next one sold too.

In 2015, he co-founded Reach. The fund just closed its fifth fund at $265M.

The wildest part? He’s too humble to tell you any of this himself.

So I got it out of him on today’s episode. This was a special one for me. Hope you enjoy it!

[YouTube] → [Spotify] → [Apple Podcast]

Also in this episode:

[09:40] How Argentina’s first VC circle found and backed Mercado Libre

[45:20] Why Reach invested in Replit at pre-seed in 2016

[49:20] Why AI is raising the bar for technical cofounders

[51:55] The “bloodbath” Esteban expects as companies cut AI tools

[55:05] Who should move to Silicon Valley and how everyone else should spend time here

He just came back a unicorn.

Zero to a $1.2B valuation in three years. HappyRobot raised a $150M Series C led by Prism Capital and Eurazeo.

DHL, Repsol, Koch on the customer list now. Four industries, three continents.

And they got there by doing what the Silicon Valley playbook says not to.

• “Focus on one vertical”. He went multi-vertical: logistics, energy, telcos, insurance.

• “Go global late”. He’s already deep in Europe and LatAm, because his customers are global and he’s European, so why wait.

• “Sell pure software; services don’t scale”. They do both on purpose. Sending FDEs onsite is their biggest growth lever and moat: they shipped in two days what a rival quoted in three months.

The best companies don’t copy the playbook. They write their own.

A Spanish founder built a US unicorn and now he’s taking it back into Europe and LatAm, the markets the Valley tends to overlook.

Unlike any tech wave before it, AI is diffusing and getting adopted faster than we’ve ever seen.

So the next generation of great companies won’t wait to go global. They’ll build that way from day one.

In the episode, Pablo breaks it all down, starting with why he almost quit HappyRobot to go work as a freight broker.

[YouTube] → [Spotify] → [Apple Podcast]

Also in this episode:

[04:35] When he wanted to quit his own company to become a freight broker

[12:29] Why DHL has more in common with a telco than a regional logistics shop

[32:55] They deployed in 2 days what a competitor quoted at 3 months

[44:35] Why he went global this early.

5M users across 150 countries. In January, the company raised a $21M Series A led by Matrix.

But Anada Lakra started BoldVoice with a problem that felt much smaller.

She moved from Albania to attend Yale after studying English for almost 10 years. On paper, she was fluent. In person, people constantly asked her to repeat herself.

Those daily “paper cuts” slowly changed how confidently she spoke.

Years later, Anada realized millions of non-native speakers had the same problem.

They already knew English, but their accent was hard to follow

Solving that required more than another language learning app. Existing speech recognition could identify what someone said. It couldn’t diagnose pronunciation, cadence and word stress with the precision of a speech coach.

So BoldVoice – or how she calls it – “an audio AI lab masked as a consumer company” built that capability

What looks like a simple coaching app is powered by proprietary AI models trained to hear how people actually speak English.

That technical foundation became the moat behind Boldvoice.

[YouTube] → [Spotify] → [Apple Podcast]

Also in this episode:

[18:18] Why BoldVoice built its audio AI in-house

[18:45] What conventional speech recognition misses about non-native speakers

[26:03] How a seven-person team reached $10M+ in revenue

[27:58] How BoldVoice turns AI research into a better coaching product

Not even a vibe coder like me: we mostly live in Claude Code, Cursor, and Codex.

These players are at war to own the coding agent, maybe the biggest AI market there is.

Luis Bitencourt runs GitHub Copilot. So I asked him straight up: is Copilot losing the race?

His answer:

Everyone’s fighting on two layers: the harness and the model. Luis thinks neither is winnable.

No model has stayed on top more than three months. And the harness, in his words, has “no huge secret sauce” long term.

What doesn’t move is the platform underneath. Every one of those tools runs on GitHub.

So instead of locking developers into GitHub’s own agents, they let Claude and Codex now run as agents next to Copilot.

Both harness and model are replaceable. Luis is betting the place where humans and agents come together to ship software is much harder to move.

If you’re building anything on top of AI right now, that’s the reframe: you might be fighting for the layer that resets every 3 months, not the one that lasts.

I lived a mini version of this: my podcast team moved all our work to a GitHub repo this year. I resisted it, tried to make it work in Notion, and I couldn’t.

Where he sees the next battle field of agents:

“The same way that you think of PRs, you think of GitHub. Can we get to a point where you think of sessions, you think of GitHub?”

[YouTube] → [Spotify] → [Apple Podcast]

Also in this episode:

[16:20] What Rappi and Loft, two messy physical-world businesses, taught him about humans in the loop

[28:45] Why coding agents are just the beginning: code review, SRE, debugging

[38:25] Why Microsoft’s distribution buys a head start of months now, not years

Fifteen years later he was leading the engineering behind Instagram’s feed, the personalization machine billions of people open every day.

His take on AI agents: they’re making the mistake Instagram almost made. They optimize for what you want right now instead of what you actually want.

At Instagram the easy path was showing people whatever kept them scrolling. His name for it: candy for lunch. Amazing the first time, miserable by week two. The hard part was never the ranking model, it was choosing an objective function that paid off long-term, even at the cost of engagement today.

Agents take the candy. They validate whatever you say and optimize for the next reply. Feels great in the moment. It’s not the relationship you wants with agents.

Then add memory. Everyone’s racing to bigger context windows, assuming more context means better answers. Past a point it’s the opposite: the extra data makes the agent worse.

Rodrigo says forgetting isn’t a flaw in memory, it’s what makes memory usable. The win is knowing what to throw away, and that’s the app’s job, not the model’s.

Rodrigo left Meta to do something about it, co-founding Vetto to build the research-grade data frontier models train on, betting the next leap comes from better signal, not more of it.

[YouTube] → [Spotify] → [Apple Podcast]

Also in this episode:

[21:30] Why Instagram’s early feed almost fell into the same trap

[33:20] Why a bigger context window can quietly make your agent worse

[35:30] Why forgetting isn’t a bug, it’s what makes memory work at all

[43:20] His lessons from Facebook’s AI Research on building product & research orgs

The room was full of founders spending a few weeks in the Bay trying to crack the city.

So Santiago pushed me on the one thing I get asked most: how do you actually break into Silicon Valley?

Here’s what I think now after 5 years in SF:

• Skip the big events (SF Tech Week, SaaStr, Salesforce). They’re built (mostly) for tourists, but the locals are who you came for.

• San Francisco is a city that happens indoors. Deals close at dinner parties. Friendships start at birthday BBQs and soccer pickup. You have to hack your way into the indoors.

• Don’t network. Just live a life you enjoy. The most successful people I’ve met in this city, I met playing soccer. Some of them run top venture firms in the Valley.

• Stop asking what you can get from SF. The investors and founders worth knowing are exhausted by coffees from people they’ll never see again. Show up with something to give and the city opens.

• Capital follows the circle of trust. SF investors will pass on a business they love if your LinkedIn shows zero connections in common. The YC and Stanford logos are shortcuts into that network, not substitutes for it.

The real reason to come isn’t the capital or the talent. It’s what this city does to your mindset.

You don’t realize you needed it until you leave and notice you came back a different person.

Santiago and I kept going from there. We talked about AI, the founders I’m betting on now, and what content survives in the AI era.

[YouTube] → [Spotify] → [Apple Podcast]

Also in this episode:

[10:35] Why San Francisco Runs on Abundance, Not Scarcity

[14:55] Silicon Valley’s Real Advantage Is the Circle of Trust

[22:15] Why VC Conviction Is Becoming More Psychological Than Rational

[34:15] Why AI Makes “Informative” Content a Losing Game

[43:50] Why San Francisco Happens Indoors

But the story isn’t the product. Carlos arrived from Colombia with no network in the US insurance industry. Latin American VCs told him directly that a Latin team couldn’t run commercial processes in the US.

He ignored them. In November 2024, he bet everything on a single design partner in Miami — no charge, five months — while in parallel warming conversations with other prospects. They launched in April. Eight clients in three weeks. 80% demo-to-paid conversion.

What looked like speed was months of invisible work.

Braven raised $6M from Collide Capital, Fiat Ventures, Broom Ventures, Aito Capital, and a16z scout.

Carlos didn’t have Stanford on his resume or a warm intro to a single US insurance company. He showed up in person when everyone else sent emails. He called when nobody calls. All while everyone told him a Latin team couldn’t crack US commercial insurance.

This is one of the most honest conversations I’ve had about what it actually takes to break into a traditional American market from the outside.

[YouTube] → [Spotify] → [Apple Podcast]

Also in this episode:

[03:30] Moving to the US Gradually vs. Going All In

[08:55] Selling AI Into a Traditional Industry

[15:50] The Sniper Approach to Getting Your First Customers

[28:15] Raising From US VCs as an Outsider

[41:10] Why San Francisco’s “Boring” Culture Can Be a Superpower

Every other investor passed. “The market is too small”, they said.

That company was AuditBoard. In 2024 it sold for $3.1 billion, one of the biggest software exits of the year, on just $43.5M ever raised.

Alejandro didn’t learn to spot that in venture. He learned it in music. His early career was A&R, the people who find talent before anyone signs it. He came up under the man who discovered Linkin Park, a band every label had passed on.

He reads founders the way a scout reads a raw artist. Not by the resume. By whether they’ve lived the thing they’re building.

That instinct isn’t an accident. Alejandro grew up fifteen minutes from the Tijuana border. Son of a mechanic, first in his family, no network.

He backs founders who don’t fit the pattern because he never fit it either.

He’s forty, and he just moved to San Francisco to go all in again. Founders do that. VCs his age don’t.

Full conversation below.

[YouTube] → [Spotify] → [Apple Podcast]

Also in this episode:

[06:15] - Growing up 15 minutes from Tijuana: how it shapes how he reads founders

[16:10] - The bet that changed everything: $4M fund, week one, two accountants with no traction and why he said yes

[19:00] - "TAM's too small" and why every other investor passed on AuditBoard

[27:25] The Linkin Park analogy for how real pattern matching works

[38:45] Act One's AI super analyst trained on 10 years of their own memos

[53:35] Why 3x growth no longer impresses anyone in 2026

[01:00:20] How a $73M fund accidentally became a native AI fund

His name is Victor Cardenas, founder of Slash.

$250M in revenue. 65 people. Profitable.

2% of all Facebook ad spend runs through Slash. More than 5,000 businesses spend almost $10B a year on its corporate cards.

A year ago, Mickey Malka of Ribbit Capital, the world’s #1 fintech investor (Robinhood, Nubank, Coinbase), said, “fintech is dead.”

Today, Ribbit is leading Slash’s $100M round at a $1.4B valuation.

Why?

Because Victor is executing something no traditional bank can copy.

More than 50% of Slash’s engineering hours go into building internal software. Agents that automate what banks solve by hiring hundreds of people: processing documents, managing disputes, and responding to bank partners.

15% of PRs going to production today are initiated by non-technical people on the team.

Meanwhile, less than 5% of companies in the US use a fintech. Everyone else is still on interfaces that, in Victor’s words, “were last updated in 2003.”

That is the opportunity.

Victor wants Slash to be this generation’s JP Morgan. When you hear Victor’s story and how Slash is being built, you’ll believe him.

[YouTube] → [Spotify] → [Apple Podcast]

Also in this episode:

[6:45] - How to adapt and compete in Silicon Valley’s high-performance environment

[13:25] - How AI is reshaping startup strategy and execution

[15:05] - How to scale niche vertical fintech products using AI

[22:05] - How to build defensibility when AI commoditizes everything

[32:05] - Why Venezuela could become a major business opportunity

Listen the full interview here 👇

Techstars founder David Cohen was an angel investor in Uber. He also had the chance to invest in Lyft — and passed.

Same market. Same timing. Two completely different outcomes in his portfolio.

The difference? At Uber, he met Ryan Graves (employee #1) at Techstars. “I really clicked with him.” He invested without ever meeting the Uber founders.

At Lyft, Matt Van Horn was a Techstars mentor. They were friends. But the idea seemed stupid: “Intercity transport for college kids? I didn’t get it.”

That same team pivoted, fast-followed Uber, and built a multi-billion dollar company.

After 20 years evaluating startups, the pattern behind his misses is always the same.

The Miss Framework: Why “Bad Ideas” Create Unicorns

The problem: We judge founders by their initial ideas, when ideas are what change the most.

The framework: There’s a Venn diagram that explains everything: “Seems like a bad idea” vs “Actually is a good idea.” The tiny overlap in the middle is where real money is made.

Why it works:

1. Uber and Airbnb were “stupid ideas” — No one wanted to ride in a stranger’s car. No one wanted to sleep in a stranger’s house. Today they’re worth hundreds of billions.

2. The idea blinds you to the team — When an idea seems so bad, you never get to properly evaluate the people. And the right people will pivot to something that works.

3. The market matters more than the idea — Cohen now asks: “Do you love the market? Because they’re probably staying in that market.” The specific idea will change. The market won’t.

The tactical shift:

Before: “Is this idea good?”
Now: “Do I love this team? Am I interested in this market?”

The lesson: Great teams can fix bad ideas. Great ideas can’t fix bad teams.

“The commonality in my passes is that they were ‘bad ideas.’ And there’s a Venn diagram I always remember: what seems like a bad idea, what actually is a good idea, and that tiny overlap in the middle — that’s where people make fortunes.”

This is the simplified framework, but David goes much deeper on how AI is changing founder evaluation (spoiler: technical talent matters less, market insight matters more), why he believes we’ll soon see “zero-employee unicorns,” and his contrarian take on why you DON’T need to be in San Francisco to build a world-class company.

🎧 Listen to the full episode: → [Spotify] → [Apple Podcasts] → [YouTube]

Also in this episode:

[15:50] – Why AI creates a “barbell effect” for startups — the middle gets stuck

[26:00] – How to scale networks without destroying them — segmentation is key

[31:30] – Techstars’ internal meme: “Replace ‘AI’ with ‘software’”

[38:35] – Why AI agents will do venture capital — and probably better than humans

[40:10] – Jack Bogle’s philosophy applied to VC: “The finance industry just subtracts value from society”

Instead of freezing, Freepik became one of the first major image platforms to integrate generative AI. Today, Freepik is one of the most widely used AI creative apps, according to a16z.

This is exactly how they did it.

The Playbook: Action First, Strategy Later

Most incumbents respond to disruption with endless analysis, committees, and 18-month plans. Joaquín did the opposite: he acted before having clarity.

1. Accept that you don’t have a plan (and act anyway)
When DALL·E 2 launched, Joaquín had COVID. Isolated, he spent 24 hours a day looking at AI-generated images. He came out of that period without a plan — but with one clear conviction: the existing business model was dead.

“When you don’t know what to do, do something instead of standing still. Every action teaches you something.”

Rather than waiting for clarity, he made a simple next move: put the technology in users’ hands and see what happened.

2. Use your distribution as a laboratory
Freepik had 600,000 subscribers. Instead of protecting the legacy business, they turned that audience into an experiment:

• They integrated Stable Diffusion 1.5 the day it launched

• They were the first major platform to do so

• They listened closely to what users asked for (history, visual styles, thumbnails)

• They iterated on what they could control and waited on what they couldn’t

Their existing distribution became a competitive advantage — not because it defended the old model, but because it accelerated learning about the new one.

3. Strategy emerged from action
Only after acting did Freepik truly understand its position:

• AI models quickly commoditize (many LLMs competing with each other)

• The real value sits in the wrapper: how the model is exposed to the world

• Freepik already had distribution and deep knowledge of creative users

“Strategy came after action. We started by reacting to users, and only then did we develop the strategy.”

4. Avoid self-deception about your ‘advantages’
Adobe, Shutterstock, and Getty fell into the same trap: they built comforting theses (“only we have copyright-safe images”) that turned out to be wrong. Their models were weaker because they restricted their training data.

Joaquín recognized the pattern: companies look for AI narratives that favor them instead of facing reality.

Freepik accepted that its millions of images were not a sufficient moat — and that freed them to compete on product, not protection.

“AI often forces you to rethink everything from scratch. And that’s very hard to do.”

This playbook isn’t about having all the answers. It’s about moving fast enough to find them before your competitors do.

Joaquín goes much deeper into how he handled conversations with his team (some needed psychological support), why he believes humans will remain essential in creative work, and the framework he uses to think about what AI will never be able to replace.

🎧 Listen to the full episode: → [Spotify] → [Apple Podcasts] → [YouTube]

Also in this episode:

[10:00] Why launching AI without a clear strategy beat staying frozen

[23:00] The “Humans vs AI” framework and what machines will never replicate

[31:00] The hardest conversations with artists who thought their careers were over

[45:00] The strategic mistake that made Adobe, Getty, and Shutterstock fall behind on AI

[58:00] Why distribution is a temporary advantage, not a real moat

[01:00:00] How building Freepik from Málaga—not Silicon Valley—became a competitive edge

Manny’s path took him from a shrimp farm in Ecuador to the heights of Silicon Valley, where he built Outreach into a $4.4B SaaS. But in 2023, a conversation with the CEO of DocuSign handed Manny a “Red Pill” moment: if AI agents do the work of humans, companies will hire fewer people, and the seat-based SaaS model will collapse .

He realized the current infrastructure—giants like Salesforce, Netsuite, and Datadog—was “conspiring against him.” They could track closed deals or API calls, but they couldn’t distinguish between an agent’s cost (tokens) and its value (reasoning) .

Manny founded Paid to solve this: a way to interpret agent behavior so builders can stop selling cheap software and start selling valuable work. Paid has raised +$30M from investors like Sequoia, Lightspeed and EQT.

Spotify | YouTube | Apple Podcasts

🗒 What we covered:

[13:00] – How the tech recession forced Outreach to reinvent itself

[18:30] – Why seat-based pricing is dead and what replaces it

[34:00] – Why the Agent Economy is impossible without a new metering layer

[45:30] – Why generalist agents lose to specific domain experts

🧠 Big Ideas

#1 Tech Is a Great Starting Market, But a Terrible Place to Stay

Outreach grew by selling seat-based SaaS to tech companies. But when the market crashed and those companies cut staff, Outreach’s revenue stalled overnight. He realized tech gives you PMF, not durability. Founders must ask: Is my business built on early adopters, or on industries that survive downturns? If you want a long-term company, your revenue has to cross the chasm long before you think you need to.

#2 The Question from DocuSign’s CEO That Proved Seat-Based SaaS Is Dead

DocuSign CEO told Manny: “If your agents work, why would I hire more people next year?” That single sentence exposed the flaw behind the entire SaaS model. If headcount shrinks, seats shrink. If seats based pricing shrink, SaaS dies. Founders building AI products must stop pricing like old software. The future is not per-seat, it’s per outcome.

#3 The 10× Opportunity: Ultra-Vertical Agents That Replace Salaries, Not Seats

Manny argues that the real AI opportunity isn’t general-purpose agents, it’s vertical agents that perform specific workflows so well they replace human headcount. And when an agent replaces five SDRs or automates a mortgage workflow, it isn’t a $49/mo seat, it’s a slice of payroll. That shift from IT budgets to labor budgets is what makes the agent economy a 10× larger market than SaaS.

😲 My Favorite Quote:

“At Outreach I was paying fortunes to Zuora, Salesforce, and Netsuite, yet it felt like those systems were working against me. The status quo wasn’t neutral; it was my enemy. It was in my face, laughing, taking money out of my pocket, and reminding me I couldn’t do anything about it.

And when you get that kind of punch in the mouth, something changes. That taste of blood wakes you up. You stop thinking in terms of market opportunities and start thinking in terms of refusing to feel powerless again. You promise yourself you won’t get hit like that anymore, and you begin fixing what those systems will never let you solve.”