Andrew Wilkinson on finding great startup ideas, AI workflows, and why money doesn't buy happiness

Executive overview

Most founders chase ideas that are popular, competitive, and exciting — and lose. The real opportunity lies in boring, under-fished niches where margins are real and competition is thin.

Andrew Wilkinson has started or been involved in 75+ businesses and now runs Tiny, a holding company with nearly $300M in revenue built entirely without venture capital. His framework: find where the fish are, not where the fishing boats are.

The core mistake is optimising for what excites you, not for what has structural advantages — low competition, recurring demand, and an unfair edge you personally bring.

Finding good startup ideas

• Avoid ideas everyone has: cafes, restaurants, project management tools. High excitement = high competition = compressed margins.
• Fish where the fish are (Charlie Munger): find a small pond with lots of fish and no other anglers, not the crowded lake.
• Boring ideas win: pest control, government form-filling software, exhaust vent cleaning. Nobody romanticises them; few compete.
• A form-filling SaaS helping people claim disability grants was making $30M/year — a business almost no one thinks to start.
• Your first business should be simple and give you an immediate win. A quick win builds the narrative that you can do this — critical for persistence.
• Wilkinson's first business (web design agency Metalab) worked because the model was simple: find client, send invoice, do work. That win carried him through later failures.
• Match the idea to your unfair advantage: a unique skill set, domain knowledge, or access others don't have.

Avoiding bad ideas and business models

• If there are lots of dead bodies in a space, something structural keeps killing entrants — assume you're not the exception.
• Wilkinson's Flow (Asana-before-Asana): lost $10M competing bootstrap vs venture-backed incumbents. "It was like Fiji deciding to invade the United States."
• He also failed at a pizzeria, designer cat furniture, an online DJ school, a skin cream business, a local news paper. Common thread: ignored structural disadvantages.
• High competition doesn't just mean harder work — it means lower prices and lower margins by definition.
• The test: would it be easier to start this, or an enterprise SaaS that solves a narrow, unsexy problem? Usually the latter.

Lifestyle businesses vs venture scale

• Bootstrap does not mean small. Tiny bootstrapped to ~$300M in revenue across 40+ businesses.
• The only real difference from VC: tolerance for burning money before the model works. In non-hyper-competitive markets, that cash burn is often unnecessary.
• The "Things" app (task manager) is Wilkinson's example of a lifestyle business done right: small team, probably $5–25M revenue, bootstrapped, founder does what he loves. The VC measure of success says it "lost" to Asana; Wilkinson says it won.
• VC money only makes sense if you need to light capital on fire to capture a winner-takes-all market before others do.
• Don't compete with the trawlers if all you need is enough fish to feed your village.

Keys to a great business model

• Moat is the primary filter. A moat means competitors can't easily take your customers.
• Types of moats: brand (pricing power, loyalty), network effects (each new user increases value for all), high switching costs (less desirable but effective).
• Elon buying X was a live stress test of network effects: changed the brand, lost 80% of staff, but the network held.
• Letterboxd (owned by Tiny): why would a film fan join a competitor's network when all their friends are already here?
• Serato (largest DJ software): moat via passionate user base plus deep hardware integration.
• Businesses Tiny buys should be "hard to mess up." Most businesses are held together with dental floss.
• Tiny's default post-acquisition move: change nothing. Leave management in place. Only intervention — replace the founder if they want to leave.

Hiring and people

• "There are no problems, only people problems." A great team handles any crisis; one bad actor poisons everything.
• Hire for what you need now, not for potential. You can't mentor someone into capability they don't have.
• Romantic-relationship mistake: "I'll hire them and change them." Doesn't work.
• Heuristic: if you've asked "should I fire this person?" even once, fire them immediately. Superstars are the ones you can't imagine losing.
• CEOs are elephants; you're the rider. They go where they want. Interview carefully, listen to what they say they'll do, and only hire if you'd nod along.
• If you tell a CEO to do X and they want to do Y, they'll sabotage X — consciously or not. Hire someone already aligned.
• Wilkinson spends ~20% of his time filtering people to ensure he's only working with those he enjoys.

AI stack and workflows

• Primary tool: Lindy.ai — builds automated workflows and agents. Wilkinson is an angel investor but discovered it organically.
• Email agent stack (five layers):
  • Does Andrew even need to see this? Archives irrelevant threads — cuts inbox by ~20%.
  • Is this time-sensitive (24 hours)? Labels it separately so no emergency is missed.
  • Is this a simple decision? Agent emails a multiple-choice prompt; reply with a number and it drafts and sends the response.
  • Replaces a full-time email assistant.
• Calendar agent: research perplexity on every meeting contact 30 minutes before, pulls email context, texts a briefing.
• Travel agent: spots upcoming trips, cross-references CRM locations, suggests who to meet in that city.
• Calendar cosmetics: emojis auto-added to every event by type.
• Replit for vibe-coding: builds web apps and sites solo. Claude 4 backend. "Things that previously required a team of five."
• Limitless (wearable recorder): queries conversations as an LLM. Most useful for relationships — post-argument analysis, "how could I have been a better dad today?"
• Claude for writing, Gemini 2.5 for large-context tasks (entire medical records), ChatGPT for general use and voice mode.
• ChatGPT deep research: replaced a $400–500/guest researcher for podcast prep. O3 Pro version meaningfully better.

AI and the future of work

• We are in the Palm Trio phase: power users can see what's coming, but it's not yet accessible to everyone. The iPhone moment is still ahead.
• Jobs already being displaced: translators, researchers, admin, executive assistants.
• Knowledge work broadly is at risk if model capabilities scale as predicted. Dario Amodei (historically conservative) said models will exceed all PhDs by 2027.
• The Lindy/agent stack currently replaces tasks Wilkinson used to need a full-time assistant for — for ~$200/month.
• Future inflection: when you can open ChatGPT, describe your job, and it stands up a digital employee without requiring you to build the agents yourself.
• Advice for new grads: get very good with these tools now, use them to build wealth, diversify into compute and energy.
• People tend to overestimate short-term change and underestimate long-term. Opportunity window remains open.
• Wilkinson's kids are 5 and 8. He's not stressing about their career path yet — too multivariate.

Money, happiness, and mental health

• The anxiety loop doesn't go away with more money. At $15M revenue, he was stressed. At $300M, same stress, same thoughts.
• Met multi-billionaires still upset Jeff Bezos could buy a super yacht. The comparison never ends; the peer group just shifts.
• Reframes that helped: redirecting 90%+ of wealth to a philanthropic foundation so wins feel externally meaningful, not just personal accumulation.
• Owning more stuff made him less happy. More houses = more people managing them = owned by possessions.
• Stays relatable: drives a normal car, meets new people in cafes, doesn't show the house early.
• Biggest intervention: SSRIs. Treated lifelong anxiety. Within days, "someone had turned down the volume on the nasty voice." Four-plus years on an SSRI. No amount of external success had done what that one medication did.
• Cut pill into 10 pieces out of fear; took 6 months to work up to a full dose. Now advocates openly.
• Note on side effects: 23andMe or similar can show how you metabolize different SSRIs — choose one you can metabolize well.
• ADHD diagnosis followed. 30% of entrepreneurs have ADHD vs 5% of the general population. Explains: loving new things, inch-deep-mile-wide, describing oneself as unemployable.
• ADHD medication took his brain "from Times Square to a quiet library."
• Practical impact at home: repeating the garbage example — forgetting isn't lack of care, it's executive function. Diagnosis gave him empathy for himself and vocabulary for his partner.

Books, products, and closing advice

• Recommended books: The Laws of Human Nature by Robert Greene (compendium of personality types and psychological patterns); How to Get Rich by Felix Dennis (candid warning that wealth is probably not worth it, told by someone who did it anyway).
• Favorite motto: "Easy choices, hard life. Hard choices, easy life." — Jerzy Gregorek. Shutting down a project, firing someone, walking away — almost always right.
• Favorite product: Matic robot vacuum (former Google engineers, machine vision, never gets tangled).
• Key business philosophy: "Ask for amazing and you'll get something great." Wilkinson emailed Apple PR asking to interview Steve Jobs at 17. Didn't get Jobs directly, but got a private tour — and ended up next to Jobs the whole time anyway.