Andrew Wilkinson joins Shaan and Sam to talk through which AI tools are actually transforming how they work — Lindy, Fyxer, Fathom, Howie — and how to build a personal AI evaluation system. The episode also covers why software may be a worse business going forward as vibe coding lowers barriers, the IREN stock thesis, the “charisma discount” trap in investing, and why Andrew’s best financial discipline is making money illiquid before he can spend it.
Speakers: Andrew Wilkinson (guest, Tiny Capital founder), Shaan Puri (host), Sam Parr (host)
AI as a New Continent [00:00:00]
Andrew: The way I’ve been thinking about AI lately: imagine you discovered a new continent that already had ten billion highly educated, highly motivated geniuses living on it — and they were willing to work for you for free. That’s what’s happened. The continent has been there the whole time. We just found the boat.
Shaan: That’s a good frame.
Andrew: The problem is most people are standing on the dock going: “This seems risky. What if the continent is dangerous? What if I don’t like it there?” And a smaller group of people are just getting on boats and sailing. The ones sailing are going to have an enormous advantage over the ones on the dock.
Sam: And you’re in which camp?
Andrew: I’m mostly on the boat. Still learning to sail. But I’m on the boat.
The K-Shaped Future [00:06:00]
Shaan: There’s this economic idea of a K-shaped recovery — where one group goes up and one group goes down after a shock. Do you think AI creates a K-shaped economy?
Andrew: Yes. And it’s already happening. The people who learn the tools and build the habits — who start every work task by asking “can AI do this?” — are getting dramatically more productive. The people who resist, ignore, or use AI superficially are falling behind at the same rate.
The scary thing: the gap isn’t linear. It’s compounding. The person who figured out AI prompting two years ago has built workflows, intuitions, and shortcuts that are now second nature. The person starting today has to catch up to a moving target.
Sam: What’s the single thing someone should do if they’re trying to get on the boat?
Andrew: Pick one task in your work that you do repeatedly and that you hate. Spend two hours trying to get AI to do it. Not one hour — two. Because the first hour will be frustrating and you’ll be tempted to give up. The second hour is where it clicks. Do that for one task, and you’ll start doing it for everything.
The Tools Stack [00:14:00]
Sam: What are you actually using?
Andrew: Let me go through the stack. Fathom is the foundation — it records every meeting, transcribes in real time, creates a summary and action items afterward. I haven’t taken a meeting note by hand in over a year. I just focus on the conversation.
Shaan: Fathom is the one that changed my behavior the most too. I used to spend mental energy in meetings trying to remember what was being said. Now I’m fully present because I know Fathom is catching everything.
Andrew: Exactly. It removes the cognitive tax of documentation.
Second one: Lindy. This is a full AI agent builder — think of it as Zapier but the agents can actually reason. I’ve built meeting prep agents that pull in everything about a person before a call: LinkedIn, their company, recent news, any prior conversations. Takes ten seconds to run. I walk into every call knowing things I never would have had time to research manually.
I also built a restaurant reservation agent. When I’m traveling, I give it my dates and preferences and it calls restaurants and makes reservations on my behalf. Via voice. It’s not just filling out a form — it’s having a phone conversation.
Sam: That works reliably?
Andrew: Mostly. Occasionally it gets confused if the restaurant has a complicated phone system. But it works probably 80% of the time, which means I’m delegating 80% of the task.
Shaan: What else?
Andrew: Fyxer for email. It drafts replies in my voice based on my email history. The quality is good enough that most drafts I edit rather than rewrite. Maybe one in ten I discard and write from scratch. That’s a massive time saving.
Howie for scheduling. It handles all my calendar back-and-forth. I give it rules — no meetings before 10am, no back-to-back calls, 30-minute buffer before anything important — and it manages the inbox coordination on its own.
Sam: How do you evaluate whether a tool is actually working?
The AI Evaluation Framework [00:26:00]
Andrew: I built a simple system. Every three to six months I sit down and write out ten real tasks from my actual work. Not test tasks — tasks I actually have to do. I give each tool a grade: did it do this task at the level I’d accept from a junior employee? Yes, no, partially.
If a tool is doing six out of ten tasks at that level, it’s earned its place in the stack. If it’s doing three out of ten, I cut it or downgrade how I’m using it.
The reason I run this regularly: AI tools improve fast. Something I tested eight months ago and dismissed might now be excellent. And something I’m using might have been surpassed by a competitor. The evaluation schedule keeps me honest.
Shaan: Most people either hype everything or refuse to engage. You’re being very empirical about it.
Andrew: The hype is useless and the resistance is useless. The question is: does this tool do real work at a standard I’d accept? That’s it.
E-Commerce and Inventory [00:34:00]
Sam: You run physical-product businesses. How are you using AI there?
Andrew: Two big areas. Product photography is the exciting one — we’re probably 90% of the way to AI-generated product photography that’s indistinguishable from a professional photo shoot. For catalog-style photography, a lot of e-commerce companies are already not hiring photographers. It’s not perfect yet — complex lighting, texture-heavy products, fashion — but for simple products on clean backgrounds, it’s effectively solved.
The second area is inventory forecasting. We have businesses with large SKU counts and seasonal demand. We were using fairly basic spreadsheet models for inventory. We’ve migrated to AI-assisted models that incorporate external signals — weather, competing product launches, social media trend data. Forecast accuracy has improved meaningfully. The cost of overstocking and understocking goes down, and in e-commerce that’s a margin story.
Shaan: What’s still not working?
Andrew: Customer service quality control at scale. I want AI to monitor all customer service interactions and flag anything that’s handled badly — wrong information given, customer left unhappy, policy applied incorrectly. Technically this should work. In practice, the false positive rate is still too high and you end up with a supervisor spending more time reviewing AI flags than they would have just reviewing calls manually.
Software as a Worse Business [00:44:00]
Shaan: You had a take before we started recording that software is a worse business going forward. Expand on that.
Andrew: Here’s the logic. For the last twenty years, if you wanted to build vertical niche software — say, software for dental practices, or software for pool service companies — you needed a real engineering team. Real money. Real time. The barrier protected the incumbents.
Now, with AI-assisted coding, someone can spin up a basic version of that vertical software in a weekend. The moat that protected a lot of SaaS businesses — the engineering required to build it — is evaporating.
Sam: So what survives?
Andrew: Enterprise software survives. The sales cycle, the procurement process, the integrations — those aren’t replaced by good code. You still need relationships, you still need compliance and security certifications, you still need the three-year contract. The barrier there is distribution and trust, not code.
Social networks and communities survive. The product is the people. You can’t vibe-code a user base.
Data-moat software survives. If a business has accumulated years of unique proprietary data that trains its models or powers its analytics, that data doesn’t get commoditized because someone can now write code faster.
What doesn’t survive: the vertical SaaS company in the middle. Good product, reasonable niche, but no real moat beyond having been first. Those businesses are going to get competed down.
Shaan: That’s a scary take if you own one of those.
Andrew: If I owned one of those, I’d be looking very carefully at what the actual retention driver is. Is it the software? Or is it the data, the integrations, the switching costs? If it’s genuinely the software — I’d be worried.
The IREN Stock Thesis [00:56:00]
Sam: You mentioned a stock pick. What is it?
Andrew: IREN. Spelled I-R-E-N. It’s a Bitcoin mining company that’s pivoting to AI data centers. Here’s the thesis.
Bitcoin mining companies and AI data centers need the same thing: massive amounts of cheap power, cooling infrastructure, and purpose-built facilities. The build-out is almost identical. IREN built their infrastructure for Bitcoin mining, which means they got their power contracts and facilities at Bitcoin-era prices, not AI-era prices. AI data center demand is outstripping supply dramatically.
The market still prices IREN like a Bitcoin miner — cyclical, commodity-dependent, lower multiple. But the AI data center transition means their infrastructure is worth significantly more than the market is giving them credit for.
Shaan: The “miscategorized” play.
Andrew: Exactly. The market puts things in boxes. If you were a Bitcoin miner last year, you’re still a Bitcoin miner in the market’s mental model. But the underlying asset has changed. The repricing happens when analysts and institutional money catch up to what the business actually is.
Sam: How much have you put in?
Andrew: Meaningful position. I think there’s a 2-4x from here over a few years if the AI data center transition plays out as expected. It could also go to zero if Bitcoin crashes and they don’t execute the pivot. Asymmetric bet.
The Charisma Discount [01:06:00]
Sam: You talked about a “charisma discount” in investing. What is that?
Andrew: When you’re evaluating an investment — a founder, a company, a deal — charisma scrambles your judgment. You meet someone who’s articulate, confident, has great stories, makes you feel good about yourself. The charisma produces a feeling of “this is a winner.” But that feeling is not correlated with business quality.
The charisma discount is the premium you pay — or the return you sacrifice — because you were charmed. You invested at too high a valuation. You didn’t push back on due diligence questions you should have asked. You ignored red flags because the person was compelling.
Shaan: What do you do about it?
Andrew: I have a rule now: if I feel unusually good about a founder after a first meeting, I add skepticism, not enthusiasm. That feeling is a signal to slow down, not speed up. I specifically try to find the person who understands the business well but isn’t charming — the operator, the CFO — and have a long conversation with them instead.
Sam: Who’s the most charismatic person you’ve met that turned out to be a bad investment?
Andrew: I’ll stay off the record on specifics. But the pattern is consistent: the most charismatic founders I’ve met have had the widest range of outcomes. The best business outcomes I’ve been part of have often come from people who were competent but unremarkable in a pitch meeting.
Missing Slack Equity (Again) [01:14:00]
Shaan: You’ve brought up Slack equity a few times. Describe what that decision was like in the room.
Andrew: We were designing Slack. This is early — Stewart Butterfield’s team had pivoted from a failed game, Glitch, and were building a workplace communication tool. They needed design work. We quoted a price, they said: “We can give you equity instead.” Or they offered a mix.
At the time, Metalab needed cash. We had payroll. We had rent. The equity in a startup is worth zero until it’s worth something, and usually it’s worth zero forever. So we took cash. I think we got paid $100,000 or something in that range. Slack was eventually worth over $26 billion.
Sam: What’s the lesson?
Andrew: The lesson isn’t “always take equity.” Most startups fail and the equity is worthless. The lesson is about pattern recognition: when you’re doing work for a company that seems to be doing something genuinely new — where the failure mode isn’t obvious — you should be willing to make an asymmetric bet on them. The expected value of equity in genuinely novel companies is higher than it appears.
The other lesson: at some point in your career, if you’re in a position to take equity and still eat, do it. We were too close to the edge financially to think clearly about optionality. Building more financial runway would have let us take more of those bets.
Forced Savings and the Illiquid Trick [01:22:00]
Sam: How do you think about personal finance at your scale?
Andrew: The thing that’s worked best for me: make money illiquid before I can spend it. I’m a good spender. If there’s money in my account, I find ways to use it. Yacht? Sure. Another house? Obviously.
The discipline I’ve imposed on myself is automatic investment. Every time money comes in — dividends, distributions, sale proceeds — a significant chunk goes directly into private investments or long-term holdings that I can’t easily access. They’re not in my brokerage account. They’re in vehicles where getting the money out would require effort and delay.
It’s the same principle as a 401(k) pre-tax contribution. The money never enters the account where I can see it and decide to spend it. I just live on what’s left.
Shaan: The pay-yourself-first thing.
Andrew: But taken further. It’s not just saving — it’s making the savings inaccessible. I genuinely cannot get to most of my net worth on a Tuesday afternoon. That’s intentional. The Andrew who is sober and thinking about compound interest made the decision. The Andrew who sees a really nice boat doesn’t get a vote.
What’s Next [01:30:00]
Sam: Last question. What’s the thing you’re working on that you’re most excited about?
Andrew: Honestly: figuring out what Tiny looks like in five years with AI fully integrated. Not just “we use AI tools” — but what businesses in our portfolio can be fundamentally restructured around AI? Which ones get dramatically better margins? Which ones get replaced?
I’m in the middle of a portfolio review where we’re going through every company and asking: what does this business look like if AI gets as good as we think it will? Some businesses look better. Some look the same. And a few — we’re having honest conversations about whether we’re in the right business.
The honest answer is: I don’t know. But I’d rather be asking the question aggressively than waiting to find out the hard way.
Shaan: That’s the whole pod.
Sam: Go buy IREN. Not financial advice.
Andrew: Definitely not financial advice.