美客多如何用 1.8 万工程师、每天 3 万次发布，撑起拉美最大公司

Sebastian Barrios leads the technology team behind Mercado Libre, which he describes as a completely vertically integrated e-commerce marketplace. The company owns its distribution network, including trucks and airplanes, and it also runs a fintech operation that is almost the same size as the commerce side, with accounts, credit cards and loans. He says about 18,000 developers work there and the number is climbing, though they do not all report directly to him.

The scale numbers he gives are large. The company delivers more than 5 million packages a day and expects to reach over 100,000 employees. It is valued at over $100 billion and has around 100 million users, customers and sellers on the platform. Barrios says engineering headcount at that level is hard to find outside big tech or huge banks.

Delivery routes drawn on a map trace out Latin America, reaching far beyond the major cities. Barrios adds that the company does not focus much on its stock price. He also says it is not better known mainly because it does not operate in the US.

Barrios says it is hard at Mercado Libre to tell where engineering stops and product begins. A small product organization exists, but its ratio of PMs to engineers, under 1,000 against about 18,000, is far smaller than at other tech companies. Who owns a product is decided by who is best for the role, and in most cases that turns out to be engineering leaders and tech leads who understand both what is technically possible and what users and the business need.

He frames the benefit as being close to the problem, the technology and the users in one head. He allows that two coordinating people can work as well and that Mercado Libre may be an exception that is hard to replicate. Hiring tests product skills but leans heavily on engineering depth, and there is little tolerance for someone who has to check with others to understand basic technical details.

Host Lenny Rachitsky notes that many companies try the no-PM approach and find engineers tire of writing specs. Barrios credits a top-down factor: the leadership is very technical and very involved. Executive meetings can go down to the wording of a button or why a pixel sits where it does, and the CEO's first question about any feature is how users will experience it, before revenue.

Asked how AI affects this way of working, Barrios says it accelerates what people can do on both sides, and he sees no competition between them. He is happy when product-minded people build demos themselves, because turning an idea into something you can touch speeds up everything. On the coding side he describes a progression from advanced autocomplete to agentic tools, while noting that security, compliance and production readiness still take time.

Mercado Libre leans on platform thinking to scale, and it has an internal development platform that handles scaling, security, building, testing and compliance. It is taking the same approach to AI with an internal platform that abstracts away questions such as where data comes from and how access is authorized. As models improve, that platform now supports agentic frameworks that take actions and handle longer tasks.

The most experimental result is skipping new code altogether. The team can extract the functionality of every microservice and let agents combine pieces of different services into new end-to-end features with a UI. Barrios extends the idea to apps whose interface is entirely generated, predicting what a user will do next and showing a personalized main screen.

Barrios's first way of working is reducing the fear of failure. No one gets fired for shipping something the market was not ready for, although there is much less tolerance for poor quality or outages. The company also does not use cascading OKRs, because the market is highly competitive and regulations vary across countries. Leadership sets high-level objectives and some areas to explore, and everything else is left to the teams.

He pairs that freedom with tight feedback loops and hands-on leadership for the most important projects. The risk of full distribution is a product made of great parts that is poor as a whole, so leaders also supply a holistic vision. How much freedom a project gets depends on the project, and he says no single rule fits all of them.

His other principles are to put users before revenue and to observe users more than talk to them. He says user research sessions show assumptions collapsing, for example when people type both their name and email into a field labeled for one of them. Watching what people actually do, he argues, often beats asking them what they want.

To keep teams aligned, Mercado Libre relies on nothing magical, Barrios says: regular design and product reviews in which the leadership is candid and honest, though always cordial. Sometimes the review reveals that the vision was wrong, and the company can pivot quickly. He admits that with around 30,000 changes to production a day, which include configuration changes, database updates and some automated changes, nobody can check everything.

He describes Brazil as one of the most competitive e-commerce markets on the planet, with local, international and Asian players, some funded by video game profits. Competing with the best pushes the company to improve. Rachitsky points out that the top startup in a market tends to attract the best talent, and Barrios agrees that this flywheel is hard to start but a clear advantage once it runs.

On building tech outside the Bay Area, Barrios says talent is distributed everywhere, but the US has a very high concentration of it. Outside the big hubs you may be the only top company, which makes recruiting harder but also means you set your own path and build things that were not built before. He adds that competitors try to recruit from Mercado Libre.

Asked how to make a fear-of-failure message credible, Barrios says culture is not what is written on the walls but what leaders do. It shows in performance reviews, in public praise, in who is promoted and in how failed projects are discussed. He gives a clear line: a system outage is an unacceptable error, but a bold bet that did not work should not be penalized and may even be rewarded with a promotion.

Firing the leader of a risky project that failed would show that the stated values are not real. Leaders also have to find people who will propagate the culture into their teams, because that is how it scales. Barrios says the top management's participation in product reviews and its direct feedback set the tone.

On radical candor, Barrios says it is even more important in Latin America and parts of Asia, where hierarchy runs strong and people are polite and find it hard to say no. Argentina, where the company was founded, is an exception, with a direct culture that Mercado Libre has exported. He says new managers learn that people value honesty, and that feedback on the product or vision is not feedback on the person.

He rejects the idea that companies should be families, because families lack an expectation of high performance. The analogy he prefers is a high-performance sports team, where people can enjoy working together but results speak for themselves.

Barrios writes a weekly email to himself and to his boss, the CEO, and sometimes to his team, covering what worked, what did not, what he learned and where he needs help. He does not expect his people to send him the same. The email frees up the weekly meeting with the CEO, because they can pick out what is relevant. Rachitsky's own version, which he calls the state of Lenny email, turns out to be a likely source of the idea.

On hype cycles, Barrios says he does not ignore new things. He bought and mined Bitcoin early, and he likes the technology behind crypto, including distributed consensus. But when he looked at the numbers, such as throughput, he saw the limits of what could be built on it at the time. Mercado Libre has its own cryptocurrency, and he says crypto is useful in countries where access to US dollars is harder.

His alarms go off when someone claims everything will run on a blockchain, and he applies the same lens to claims about AI taking over the world. His method is to understand the fundamentals and how the technology could be applied usefully, helped by experience in real businesses and deep research tools.

Barrios tells the story of an early app, built at about 16 or 17, that drained a phone's battery by switching on every sensor at once. He built it partly to learn about the sensors and partly because of the then-popular advice to run a battery completely down. At the time Apple's executives, including Steve Jobs, were personally calling developers whose apps were rejected, to avoid angry press coverage.

Jobs called from a blocked number and told him the app could not stay in the store. Barrios pushed back that no rule banned battery draining, and Jobs told him to check again because a new rule had just been added. He did not realize who he had talked to until minutes later, found Jobs pleasant and relaxed on the call, and kept building apps.

At 19 he made an app that alerted callers when they were close to the five-minute limit on a free-calls plan, so they could hang up and call again. He did it for his girlfriend, now his wife, and priced it at $1 to cover the $100 App Store fee, and it became a top-selling app in Mexico and other countries. Companies and governments then asked him to build apps, which led to a mobile development company he later sold in order to focus on products.

One of those products was a taxi-hailing app for Mexico City, modeled on Uber's early network. The first test simply had staff call taxi companies on behalf of passengers, and demand was so large that they had to build a call center to call other call centers while they recruited drivers. Once drivers and riders joined, the network effects began, and the company competed with Uber.

Barrios drinks only water, with no alcohol, coffee, tea or juice. He says seeing heavy drinking among teenagers in Mexico, where the legal age is 18, taught him early that he wanted control of his own state. He also does not listen to music while programming, since his father is an orchestra conductor and he grew up in concert halls. He enjoys X for news and tech updates but avoids news organizations and TV.

His mother once put a sign on the family TV saying everything you see here is a lie, which he thinks encouraged independent thinking. He tells people that his routine is not strict, and that following his curiosity has worked well because curiosity usually aligns with where he can add the most value. He adds agency as the second ingredient: being curious and then going to get what you want.

His mother wanted the children to handle any situation and said she wanted to train them like spies. She woke them one morning to go camping with no gear in a survival course, and she dropped them in the middle of Mexico City to find their way home on public transit. He adds that the family also did a lot of skiing, biking, trekking and climbing, and sees this as formative for how he approaches problems.

For books, Barrios recommends High Output Management for people in the company, the Odyssey for its enduring appeal, and The Dream Machine for the history of computing and the internet. He also loves science fiction such as Dune. His recent favorite movie is Everything Everywhere All at Once, and he recommends the children's show Bluey.

He recommends a reading program for young children that he used with his five-year-old son, framing it as building a strong foundation in a base skill rather than reading early for its own sake. He also mentions Beast Academy for math, protein bars, and 3D printers, and says he built a telescope with his son. His life motto is that life and the world are malleable, so you should be careful what you aim for because with enough agency you can probably get it.

Asked about the Tatami Project, he recounts that he wanted a room for tatami mats, so he bought construction gear and diamond cutters and cut a large hole in his childhood bedroom floor. He added a staircase and turned the room into a two-floor loft, and his parents were happy with the project. He says this fits his point that taking risks and making mistakes was always encouraged at home.