Molly Graham's Give Away Your Legos AI Advice

Imagine being handed a new coworker who is smarter than anyone on the team, told to pour everything you know into training them, and warned they might take your job within six months. That is the story a lot of people in tech are being told right now, and it is a bad one (00:24). Molly Graham, who has spent two decades helping companies survive rapid growth, from Google to Facebook to Quip to the Chan Zuckerberg Initiative, thinks the story is mostly wrong, and that it is making an already hard moment feel worse.

Thirteen years ago Graham wrote a career essay that became one of the most read posts on First Round Review, an outlet for startup advice (11:12). The idea, nicknamed give away your Legos, said that during fast growth the smartest move is to hand off the project, team, or product you built rather than cling to it. She had lived the problem herself: her department at Google grew from 25 people to 125 in nine months, inside a company of 10,000 employees in 2007, and at Facebook she watched headcount balloon from 500 to 5,500 employees during her five years there (06:08, 06:29). People she managed kept trying to hold on to what they had built, the way a toddler grips a Lego tower when someone reaches for it. Her advice was to let go on purpose, because holding on guarantees you fall behind.

Some of that advice, she says, still holds exactly as written. Job one in a fast-changing company is to make yourself irrelevant, because that is the only way to be ready for whatever comes next (13:42). And she still believes the future belongs to people who learn quickly, not people who already know a lot, a line that lands differently now that generative AI resets what counts as expertise every few months (15:56).

The grief of rowing versus steering

What has changed is the emotional weather. Graham describes conversations with engineers who say they miss what their job used to be: sitting down and writing code, rather than supervising a swarm of AI agents and checking their work (18:55). One framing she keeps returning to, borrowed from a guest on a previous episode, is that the job used to be rowing and is now steering, and a lot of people never wanted to steer (20:39). A product leader recently told her the new work feels lonely, because collaboration with humans has been replaced by collaboration with robots. Graham's response was simple: let the sadness be real before rushing to the upside.

"Change sucks. It also can be awesome, but we don't have to be fluffy bunnies about this. We can also just say, 'This is hard.'" — Molly Graham [21:44]

The data backs up the mood. A workforce survey run by this show found that 55% of tech workers now report feeling burned out, up from 44% a year earlier (30:03). Graham thinks part of that is simple exhaustion from narrative whiplash: a friend at OpenAI told her it took the company six months to go from urging everyone to use AI constantly to asking whether all that AI use was actually paying off (31:38). None of this means everyone is miserable. The same survey found roughly half of respondents report the happiest they have ever felt in their career, concentrated among people who feel AI has amplified what they are good at, while designers reported the lowest happiness because design still requires slow, human feedback loops that cannot be sped up by an agent (33:15, 34:00).

Treat it like an intern, not a genius

The bigger break from Graham's original advice is this: some Legos should never be given away, and the reason is that AI is not the coworker the hype describes. Her mental model is blunt.

"AI is like a lazy intern. It needs the same coaching and training that a human does. It needs context. It needs onboarding." — Molly Graham [35:48]

Most AI slop, in her view, comes from people skipping that step, treating an AI's first draft as if it came from the best hire they ever made and forwarding it unedited. She points to CEOs who ask AI for a strategy memo and paste it straight into a company-wide message, which quietly tells every employee that outsourcing your thinking is fine (37:32). The fix she recommends is what she and this show's host call an AI sandwich: a human sets the goal and the vision, AI produces a draft, and a human reviews and iterates before anything ships (63:04).

There is a second, more surprising twist. Giving a project to a person and giving it to a robot are not the same act, even though both look like delegation. Handing work to a colleague means truly letting go, what Graham describes as chucking the Lego at their face and running the other way so your mind is free to move on (43:55). Handing work to AI does not work that way, because the human still owns the oversight. The mental tax never leaves. Output goes up, but so does the invisible cost of checking, correcting, and being accountable for whatever the AI produced (44:47, 45:09). That is one reason burnout can rise even as productivity metrics look great: people are managing more, not less.

Because everyone with an AI tool is now effectively a manager, Graham's old rule of thumb for team size applies more broadly than it used to. She caps the effective limit at 10 to 12 direct reports, and worries about what happens to people's psychology once AI agents get counted toward that number too (47:51). She also names, for the first time in this advice, categories of work that should stay human no matter how good the tools get: tasks that require judgment, tasks that depend on deep trust between people, and tasks that require a vision of where something should go, not just a fast path to somewhere (61:55). An engineering report cited in the conversation found that while AI coding tools made engineers faster, the amount of code that later had to be rewritten rose roughly 8 times, a concrete example of speed without judgment (39:22).

None of this argues for retreating from AI, and Graham is explicit that she does not think jobs are vanishing as fast as the layoff headlines suggest; she calls most AI-branded layoffs a cover story for companies that simply over-hired (27:14). Her closing advice is to stop asking whether a job will disappear and start asking what it will look like in five years, following what she calls the Manoush Zomorodi test, named for the journalist and podcast host who has rebuilt her career every few years across three decades of disruption (28:28). Graham also pushes back on the current fashion for stripping out management layers at large companies, arguing that a manager is still the single strongest lever for how happy someone is at work, and that removing that layer to save cost now will cost more later (82:56, 83:17).

She ends with an image borrowed from executive coach Chip Conley: sometimes you need to hold a small funeral for the version of your job that is going away, grieve it honestly, and then ask what you could build next that you would be proud to describe years from now (67:49). The people living through this moment, she says, are going to have stories worth telling, about what they got right, what they got wrong, and what they helped build before anyone knew what to call it (86:36).