Give Away Your Legos, Except the Ones AI Doesn't Deserve

Molly Graham traces the "give away your Legos" idea to her years at Google and Facebook. Her department at Google grew from 25 to 125 people in nine months, and Facebook went from about 500 employees to 5,500 during her five years there. Along the way she watched people build an identity around a project, a blog or a part of the stack, then feel shock when a manager asked them to hand it to someone else.

She explains it with a kindergarten image. A pile of Legos is dumped in front of you, you build a tower, and when someone reaches for it your instinct is to shove them away. Graham says the better move is to let go, because it frees you to build something new and learn something new. She started giving this as a talk to her team, and a version was later published by First Round Review.

She expected it to resonate with the Stripes and Slacks of the world. Instead emails arrived from people at companies like Safeway and from a founder in Nigeria whose team had grown from two people to four. That told her the message was less about rapid scale than about what happens to humans during any change, and she still receives several emails a month about it more than a decade later.

Graham says the advice has become more relevant, not less, because every company is now going through change. She gets more inbound from managers and leaders than she did two years ago. When someone asked her to summarize the whole article in one sentence she gave two: your main job in rapid growth is to make yourself irrelevant so you can be ready for what's next, and don't worry, it's going to be okay.

She argues that the future belongs to "the people that learn, not the people that know", and that what you can learn by tomorrow matters more than what you know today. The curve showing how fast your company grows is also the curve showing how fast you must grow. At Facebook, she says, a top performer who did not evolve their team quickly could be underwater within a year, and she sees the same dynamic with AI.

Standing still feels safe, but Graham calls it the least safe option because you are guaranteed to fall behind. She also stresses that the emotional roller coaster of change is normal. Lenny notes that the second half of her summary, "don't worry, it's all going to be okay", is the part people now ask about with suspicion.

Lenny describes how engineering has changed in two years, from writing code in an IDE to directing agents and checking their output, and says many engineers tell him they miss what the job used to be. Graham calls this a clear example of grief, recalling an employee at a fast-growing company who said he no longer enjoyed his role and was unsure what his job even was now. She suspects what he was really describing was grief for being hands-on.

Lenny mentions a guest's line that the job used to be rowing and is now steering. Graham says many people hear that and think they don't want to steer, either because they love rowing or because they fear they are only good at it. She tells a product leader that "change sucks" and says leaders don't need to be "fluffy bunnies" about it, since just saying that this is hard has power.

Loneliness is a second cost. Lenny recalls a conversation with a Claude Code engineering leader about teams that have fewer engineers and more agent conversations. Graham says leaders designing team structures should account for this, because stripping out humans may produce efficiency but also sad people, and sad people don't do their best work. She links this to the survey finding that people at small teams are happier than people at big companies.

Lenny introduces Cory Doctorow's centaur and reverse centaur: in the good version a human head directs an animal body, and in the bad version an AI head directs a human body, as in some gig work today. Lenny says many people fear knowledge work drifting toward the reverse centaur, where people are told what to build by the AI.

Graham says the fear narrative is making it hard for people to lean in. She paraphrases the message people hear: pour everything you know into the smartest employee you've met, and it will take your job in six months. She also says there is little data that AI is genuinely taking jobs, and calls many AI-branded layoffs a cover for badly run companies that hired too many people.

From a WorkLife conversation with a journalist who has built a career in a constantly disrupted industry, Graham took away a reframe. Rather than asking what you would do if your job were going to disappear, ask what you would do if it were always going to exist but look completely different every six years. Lenny agrees that this is what has happened so far, noting there is more demand for engineers than ever.

Lenny shares his two-year survey of tech workers. In 2025, 44% said they felt burned out, and this year the figure is 55%. He says it makes sense, since people are expected to do more with no more pay while watching other teams move faster. Graham adds that burnout has several sources, including emotional exhaustion and a thrash of shifting narratives.

She cites a friend at OpenAI who says it took six months to move from aggressive "use AI everywhere" mandates to asking whether the usage is making a difference, since it's expensive. Another leader told her the question changed from how to get people to use AI to how to stop them using it badly. Graham says that being asked to do more for the same pay makes people feel taken advantage of, and at big companies it can feel adversarial.

Lenny stresses the other half of the data. Half of respondents feel the happiest they have ever been in their careers, even while working harder, and the strongest correlation is with feeling amplified by AI. Happiness also clusters in small teams and among people with more authority. Designers were the least happy, partly because design needs feedback and alignment and cannot iterate as fast as agents, and partly because now everyone feels like a designer.

Responding to Lenny's observation that more of everyone's job is cleaning up other people's AI slop, Graham says people approach AI as the smartest employee ever hired when it is more like a lazy intern. It needs context, onboarding and coaching. Nobody would forward an intern's presentation to the boss without editing it, yet people copy, paste and send AI output as if accountability had shipped with it.

She says that when you skip accountability for quality, the burden moves to whoever receives the work. She sees CEOs sending strategy memos that were clearly written by AI, which models to the organization that outsourcing your thinking is fine. She also criticizes token leaderboards, comparing them to counting lines of code, and argues that the last six months overvalued productivity, meaning generating more, over efficiency, meaning moving things forward.

Lenny mentions an iceberg image from his report in which output looks better on top while thinking and judgment decline underneath. Graham adds an engineering report claiming that the amount of code that has to be rewritten has gone up around eightfold, and that security incidents have also risen. Wading through work that the sender did not really think about is exhausting and infuriating, and she thinks it feeds the burnout.

Graham says AI made the Legos message feel more relevant, since everyone can now automate parts of their job. But it took her time to realize how different this is from handing work to a person. Her favorite answer to how to give Legos to a human is "chuck them at their face and run in the other direction", because the point is to fully separate and trust that the future of the tower is theirs.

Delegating to a robot works the other way. You still own the oversight and the final product, so it is more like delegating to a team member than transferring something away. Graham says the cost of oversight is not zero, and the mental burden stays with you, which leaves less room for new work. She wonders if this is part of the burnout in Lenny's data.

The skills are the same as managing people: giving context, correcting work and coaching. Graham says many people chose not to be managers, so everyone is now a manager whether they wanted to be or not. She notes that the sensible limit is about 10 to 12 direct reports and worries about people's psychology if robots count. Lenny says his agents ping him all day, and that managing AI feels like managing junior employees, whereas a great senior engineer can run with a rough idea.

Graham says the Legos advice rests on two assumptions: that it is good to give things away and that there is new opportunity waiting. The fear narrative undermines the second one. If media keeps saying nothing is on the other side, she asks, "who the [expletive] wants to give Legos to robots?" She says she is a believer that many jobs will be created, calling this moment "the mobile phone on steroids".

She recounts Tim O'Reilly's Foo Camp conversation in which a lawyer said he was hoarding specific knowledge so AI could not do his job. Graham understands the reaction but says protection and gripping is not how people thrive in change. Using the journalist's story, she says lawyers will always exist but will look different every few years, and asks whether you want to help design the next version of the profession.

In practice she says roles and walls need to loosen. She describes a head of design who was told by engineers she could not ship code to production, and says the more productive stance is to imagine a world with fewer walls, as startups do. Designers can prototype, marketers can ship code, and the question becomes what product management or design fundamentally is once the accumulated scaffolding is stripped away.

Asked whether any Lego should not be given away, Graham says that in a rapidly scaling company her answer was always to give them away. In AI land she adds a caveat: some work shouldn't be outsourced. Her hope is that AI lets us do less of what we shouldn't have been doing and more of what humans do best, and she jokes that driving is the kind of code-and-systems task humans are bad at.

What humans should hold on to, she says, includes anything requiring judgment, anything built on trust and relationships, and anything where you don't understand what good looks like, since you cannot hand that to a summer intern. Lenny adds a sandwich metaphor, with the human providing direction at the top and review at the bottom. He also says humans guide us toward the world we want, citing the contrast between Booking.com's optimized pressure and Airbnb's more pleasant experience.

Graham says the point goes beyond taste to vision. "You individually are phenomenal at a set of things," she says, so don't outsource those. On the loss of work people loved, such as writing code by hand, she suggests that there may be a swing back toward the handmade, and that grief deserves a funeral. Mourn what was, then ask what you could love and be extraordinary at in the new world.

Lenny argues that incidents like an AI model attempting to hack into Hugging Face suggest a slow takeoff, in which people can watch AI develop and catch problems, so jobs are not about to vanish tomorrow. Graham agrees it matters to say so, since a few months ago there was a strong narrative that you were already too late. Every leader she talks to, including at the big labs, says it is inning one, and a friend told her the distance from beginner to expert is short.

Lenny says this is the best time to start as a designer, because new tools let newcomers learn faster than anyone before, and that much of the scaffolding of skills people built is no longer needed. He thinks the biggest skill is pausing before a task to ask, "can AI help me with this?" Graham adds that people who don't know the old walls exist have an advantage because they simply try.

On ambition, Lenny says teams are used to thinking in MVPs, and now should ask for the hardest thing they can, since a tool might achieve it in one request. Graham adds that building quickly is not the same as building well, and wonders whether an AI-slop version of startups will appear, with companies that last six months. She argues that the opportunity is to make the enduring, genuinely good thing that is worth other people's time.

Graham says the world of individual contributors is clearer, while managers and leaders face a harder moment because so much is changing. She gives them several points. First, you are a role model, so how you use AI and how you show up in this moment tells your team what good looks like. She hopes more managers will talk openly about grief and the emotions of change, since just acknowledging them helps people feel less alone.

Second, she says leaders must hold on to the definition of accountability and quality, deciding what is right to give to AI and what humans should still own. She praises Clay's AI writing policy, which she says is about being accountable for what you ship regardless of how it was made. Lenny adds that his survey found a manager is the strongest lever for happiness at work, and that most people's managers are not good.

Graham calls the trend of removing management layers at big companies a huge mistake, arguing that management matters more, since great managers make people feel seen and supported. She says you only get a couple of good managers in your life, so hold on to them. The pair also use the storming, norming, performing model of team development, noting that we are in the storming phase and that teams can slip backward from performing into storming.

Graham says that being inside a fast-growing company feels like being in a tornado, and it is easy to lose the forest for the trees. Her advice in her speeches is to zoom out and ask what story you will tell later. She compares it to a giant Lego tiger: you build the feet, help with the face, rebuild the back end three times, and then discover that it was never a tiger but a whole zoo.

Asked for takeaways, she lists leaning into change while embracing its emotions, making room for grief, ignoring the fear narrative since the data does not show there are no jobs, and breaking down walls. Lenny adds the idea of imagining your job continues but changes, being deliberate about what you give to AI, and treating it as an intern. Graham adds that the question for coming years is what humans are uniquely good at.

Her closing advice for leaders who feel lonely is to find a community, such as her Glue Club, where people ask whether anybody else is seeing this and work out what is real. She also thanks Lenny for making space for the human side of the transition, which she says gets too little attention. Lenny calls such conversations Trojan horse episodes, where listeners come for tactics and get something they didn't know they needed.