Airbnb ships 80% more features than it did a year ago, and Brian Chesky gives most of the credit to AI.
Then he says the line that matters more: “AI is the ultimate democracy.” Airbnb has the same tools your team has, so the gap sits with the person at the keyboard.
Chesky, Airbnb’s co-founder and CEO, sat down with the Silicon Valley Girl podcast seven years after its host ran up to him at LAX. In 34 minutes he covers who AI amplifies and who it leaves behind, plus the notebook he fills every day.
Go-to-market works the same way. Your competitors buy the same software you do, so the gap shows up in whichever part of the funnel is weakest.
together with HubSpot for Startups:
Most founders can tell you when growth slowed. Far fewer can point to the part of the funnel causing it, because one or two soft spots drag everything else down and they’re hard to see from the inside.
HubSpot for Startups built a free GTM audit for exactly that, using what it learned helping 35,000+ startups grow:
You drop in your URL, and it reads your website and pricing, plus your social presence, tech stack and competitors, before you type a word. Then you answer a few questions about your channels and pipeline, and where things feel stuck.
▫️ A score for every dimension of your GTM, so you can see which one is holding the rest back
▫️ A prioritized fix list built for your stage and tech stack
▫️ A 90-day roadmap that tells you what to fix first
It takes about 10 minutes, and it’s free:
1. Everyone Has the Same AI Tools, So Why Is the Gap Growing? (The Amplifier Effect)
Everyone had the same tools. A few people still pulled far ahead.
“AI is like an amplifier. It makes the good people better, and the people that weren’t as good, it’s less of a multiplier for them.”
Chesky learned this as a designer. Everyone had the same tools back then, he says, and the very best people still got the most out of them.
AI repeats the pattern with higher stakes. Codex and Claude Code sit in every company’s toolbox, Airbnb’s included, and he says the performance gap widens because the tool rewards curious people who can iterate without limit.
He offers himself as evidence: he couldn’t code when he started out 19 years ago, and today he can because he can prompt. That is the gap widening among builders inside 1 career.
Execution got cheap. Taste and judgment did not, so the question moved from whether you can build it to what you build, whether you should, and whether it matters.
“AI allows an idea to get out of your head into the world faster, but it doesn’t necessarily give you better ideas.”
2. How Airbnb Ships 80% More Features With AI (Why Token Leaderboards Backfired)
A new service used to take Airbnb 1 year. Now it takes weeks.
“We look at product velocity. How fast are you shipping? How many features are you shipping? How quick are you shipping them? Are they high quality? And are they successful?”
Chesky credits 2 causes for the 80%. AI ended hand-coding as the default, and the culture changed with it, because people have to use the tools and use them correctly.
Early this year, he says, CEOs measured AI adoption by tokens burned and posted leaderboards to shame the low scorers. He says leaderboards “might have been a good thing to do for a couple months” to get people using AI, then calls them a long-term strategy that was not going to work. Then everyone asked why the AI bill kept climbing.
His replacement metric is velocity: what shipped, how fast, how good, and whether it worked. Effort is the easiest number to inflate, and customers never see it. OpenAI’s product lead makes a related case: AI-native companies out-build teams 10x their size.
Steal the 5 questions and drop them into your next sprint review.
Planning to ship at that speed? Start here:
▫️ Ship your first AI agent in a day
▫️ Sam Altman: 3 months of work now takes 7 minutes
▫️ How to 8x Your Code Output Using Context Engineering
3. Agent vs Chatbot: What Airbnb Built Instead of a Chat Box
Airbnb’s new AI search feels like chat and works like an app.
“A chatbot is a user interface. And I think because most agents are chatbots, we conflate the two.”
Chesky separates 2 things most of us blur. An agent is an application that understands you and knows you, and a chatbot is the interface for talking to one. Most agents ship as chatbots, so people blur the two, and he calls agents the future of Silicon Valley.
Travel exposes the gap in 3 ways: text is hard to compare, shoppers want to browse instead of type, and a typical group of about 3 travelers needs to share results, message the host, and watch listings change when a filter flips.
Airbnb’s answer is an AI search toggle. You describe a weekend in Paris with a certain view and architecture, and it prefilters the results while the familiar filters stay in place. Summaries written for each guest, a listing page you can question, and a comparison tool sit around it, and hosts will get an app that coaches them on pricing.
The product lives in the application around the model, the case behind 6 decisions for agents. Ask whether you need an agent at all first, then see how 3 founders shipped managed agents.
Build the application first and let chat be one door into it.
4. Chat Is a Jet Engine on a Bicycle (What to Build Instead)
Plenty of AI products put a jet engine inside a chat window.
“A lot of entrepreneurs are trying to fit the most powerful technology in human history into a messaging platform. It’s like putting a jet engine on the back of a bicycle.”
Chesky calls Instinct and Muse the beginning of consumer and agentic AI, then rejects messaging as the right interface for most work. His advice to founders: try audio, or go further out, because most people are visual and TikTok beats Twitter for that reason.
He runs the experiment on Airbnb itself. Social networks became broadcast platforms, he argues, so Airbnb is building a travel graph where you add friends and family, see who visited Chicago and where they stayed, and ask what’s good. People want recommendations from the ones who know them over celebrities and writers.
Interfaces are up for grabs everywhere. Zuckerberg is placing a $14B bet that glasses replace the phone, and Satya Nadella’s answer is a computer for every agent, where most builders hand theirs a chat box.
Chat handles a question and an answer well and strains when the job is browsing, comparing, or deciding with other people. Ask what your user would rather see or hear before you ship a chat box.
5. How a Masseuse in Tulum Gets a Global Audience on Airbnb
Airbnb sells small businesses the 1 thing they can’t build alone: an audience.
“This photographer would never have known how to broadcast to all these guests from around the world.”
Chesky builds the case with 2 examples he flags as hypotheticals, down to the made-up location. A masseuse in Tulum lives on word of mouth from locals, until a listing puts her in front of every traveler headed to town. A Paris photographer who shoots families at the Eiffel Tower sells professional photos in 30 minutes, at a price he describes as reasonable.
The menu is already long: chefs, photography, car rental, airport pickup, luggage storage, baby gear, laundry, ski rental, and Instacart groceries that know your address and guest count. He expects dozens, then hundreds, of services to plug into a booking.
The logic lives in the long tail. Ride sharing has Uber and food delivery has DoorDash, yet most services lack a global brand, and the few with an app sit outside Airbnb’s system.
Airbnb supplies the audience, and a small business or 1-person company supplies the service. Own the service, rent the audience.
6. Airbnb Tests Every New Service in 1 City First (How It Decides What to Launch)
Airbnb usually tests a new service in 1 city before it goes anywhere else.
“We usually do a test in 1 market, 1 city. If it works, then we bring it to other markets.”
Dozens of services sit in test at once. Ski rental is in test in Italy, by Chesky’s recollection, and surfing gear is the next idea he floated.
The interviewer asked for a launch bar, maybe 10% of guests using a service. Chesky skipped the number and described intake: a contact button at the bottom of the page, an apply button for services hosts, and a promise to take every lead. Floral arrangements for a romantic trip are moving through that pipeline now, and if Airbnb needs more time on your idea, it launches eventually.
Call it validation by distribution: let supply walk in, test in 1 city, then widen. That is an MVP with a boundary, and product-market fit shows up in what the test teaches.
It spares you the $40K of features nobody wanted and tells you early when the market is rejecting the idea. I’d take 1 city and an answer over 10 cities and a guess.
7. Airbnb’s CEO Has 100,000 Ideas a Year and Calls 99,000 of Them Bad
Airbnb was 1 idea in a pile of bad ones. He still builds the pile.
“When you critique your ideas, you edit your thinking and you’re afraid. And bad ideas sometimes are one step away from a good idea.”
The interviewer credits “the best way to have a good idea is to have a lot of ideas” to Linus Pauling, and Chesky stretches it to drawings: the best way to make a good one is to make a lot of them. The design firm IDEO used to run an exercise where everyone sticks Post-it notes on a wall and nobody critiques anything at first, devil’s advocates included.
By his own count, “not exaggerating,” he has around 100,000 ideas a year, and 99,000 are bad. Outsiders see only what ships, a textbook case of survivorship bias.
Chesky cites Eminem as someone who wrote constantly. Rick Rubin, who produced Eminem, built a career on taste alone, and taste needs a pile to pick from. Start yours with 100 agent ideas or 70 startup ideas from YC.
8. The Notebook System Airbnb’s CEO Uses to Filter 99% Bad Ideas (Bold, Review, Repeat)
99% of what he writes is useless. A 4-step filter finds the rest.
“At the end of every day, I bold the things that jump out at me.”
Chesky fills a notebook with about 1,000 words a day, roughly 100 lines of 1 or 2 sentences each. The system runs like this:
Write down every thought, unfiltered.
At the end of each day, bold the 5 to 10 lines that jump out.
At the end of each month, have AI read the bolded lines.
Let AI also read the lines you dismissed, spot patterns, and build a short list to iterate on.
The count wobbles mid-answer: 100,000 ideas a year becomes about 30,000 after the notebook math, and the 99% ratio holds. A good idea returns, he adds, because a month later you write it down again.
His notes read as borderline nonsense to anyone else, and he accepts that because only the survivors get seen. Treat the setup as a small self-improving loop: volume feeds the filter, and the filter picks what earns another round.
“Good ideas don’t die.”
Building your own notebook-plus-AI loop? Start here:
▫️ Garry Tan Runs YC at 400x His 2013 Output. His AGI Is a Folder of Markdown Files
▫️ I built a second brain in 10 minutes. Here is the exact stack
▫️ Your Obsidian Vault Finally Has a Model That Can Read All of It
9. Cap Your Feed at 10 Minutes a Day and Do Deep Work on a Big Screen
By Chesky’s estimate, a phone holds your attention for about 15 seconds.
“The bigger the screen, the longer your attention span.”
He confesses to a habit he says nearly everyone watching shares: too much time on social media, watching what other people ship. In his words, “we’re all like kids chasing soccer balls,” copying fast fashion as new launches pile up.
His budget is 10 minutes a day on competitors and the latest AI developments, instead of the 2 to 3 hours, or 17 to 25 trips into X, YouTube, or Instagram, most of us burn. His own ideal, which he says he follows less than he wishes, is 1 session of 20 to 30 minutes on the platform, then done, then uninterrupted time with the phone off.
The phone, he argues, trains your dopamine to multitask. A whole IMAX film, by contrast, holds people in their seats.
Our piece on phones, comfort, and disconnection shows how technology turned freedom into paralysis, so put your deep work on the biggest screen you own.
10. It Is Easier to Build a Company Now Than 20 Years Ago (And 50 to 100 More Anthropics Are Coming)
Chesky wishes he were 26 again.
“It is so much easier to build a company now than it was 20 years ago.”
Outside Silicon Valley, he says, AI feels like something happening to people, changing their jobs and displacing them. He concedes some of that fear is true, then says everyone has the power now and anyone with an idea can make something.
His forecast is 50 more Anthropics, maybe 100, plus millions to tens of millions of small companies, because 1 to 3 people can now start a whole business. If AI is a window at all, he says, it lasts 1,000 years, since technology that doubles every year for 10 years is a 1,000x.
Chesky doesn’t know what he would build at 26, but he would look at education, storytelling (you could build an “AI Disney”), and health. Andrew Ng, who taught 8 million people AI, calls AI terrible for learning, so read his case before you bet on education.
If AI still feels intimidating, he says, jump in. He felt nervous about Claude Code, asked a strong employee to tutor him, and points to YouTube for anyone without one. Here is how to use Claude like a power user.
“The water is not cold. It just looks cold.”
Ready to build? Start here:
▫️ How to Build Three AI Agents Without Writing Code
▫️ The 5-agent sales team you can build this weekend
The Amplifier Playbook
AI hands every company the same tools, so the volume of your ideas and the focus you protect decide who AI amplifies.
▫️ Founders: Run every new idea as a 1-city test and take every inbound lead while it runs. Judge the test on features shipped and how many succeed. Pick the city this week.
▫️ Investors: Ask each team how it measures AI: tokens spent, or features shipped and how many worked. Chesky’s 5 velocity questions double as a diligence list. Add them to your next call.
▫️ Operators: Retire the token leaderboard if you run one, and put the 5 velocity questions in its place. Pair every tool rollout with training on using it correctly. Start this quarter.
▫️ Everyone else: Keep a notebook and bold 5 to 10 lines each night. Cap your feed at 10 minutes a day and do deep work on your biggest screen. Start tonight.
The 5 Principles to Steal
Measure shipped value. Count features shipped and how many succeed, ahead of tokens burned. A token count tracks effort and velocity tracks results, so ask Chesky’s 5 questions in every sprint review.
Match interface to task. A chat box is 1 option; test whether your user wants to browse, compare, or talk. Airbnb kept its filters and controls and added AI search on top, so copy that move before you replace anything.
Test small, then scale. 1 market, 1 city, then widen. Let the test teach you what to fix, and take every inbound lead while it runs, the way Airbnb does for new services.
Write more, filter daily. Bold the best lines every night and let AI sort the month. About 99% of the notes get cut and the good ideas return on their own, so keep the volume high.
Protect your attention. Cap the feeds and give deep work the biggest screen. Budget 10 minutes a day for news, then do the work on a computer, where attention span stretches.
Same tools, different habits.
Pick the habits.
Watch how Chesky decides what Airbnb launches, from the Tulum masseuse to the 1-city test, in the full interview.
If this breakdown saved you 34 minutes, send it to one founder or investor who needs it.
Interfaces, agents, and product speed
▫️ OpenAI’s Product Lead Reveals the 4-Part Framework for AI Product Strategy
▫️ Stop Blaming the Model. Fix the Architecture
Ideas, notes, and focus
▫️ Your voice is the only AI moat that compounds. Here is how to clone it into Claude in a weekend
▫️ How to 10x any AI skill using Karpathy’s Autoresearch method
▫️ What Stealth Startups Know That Others Don’t
Building a company in the AI era
▫️ How Lovable hit $400M ARR in 14 months with 146 people and almost zero paid ads
▫️ The Billion-Dollar Startup Formula: Why AI-Driven Small Teams Are Beating Giants
My superhero screenplay "AI Man" is now available for you to read for free.
https://drive.google.com/file/d/1gZ2aL8NG6qBCJjMmBIywJ6gvLqafX4rh/view?usp=drivesdk
I don't have money or resources to make this film.
If you help me make this into a feature film. I would split the profit with you. I will take only 2% royalty fee.
But are those 80% more significantly better or differentiators that transform what they do or I just a check mark that they use AI more than others