Tag Archives: AI

8 Hard-Earned Truths from 8 Years at Meta

Featured

Posted on by
Reply

At Meta we celebrate Metaversaries. 8 years ago today, I joined Facebook. Keeping up with my annual tradition, I take time on this day to step back and reflect on what I’ve learned over the past year.

Year 8 was easily one of the most challenging years in recent memory. Beyond a war, the macroeconomic climate, industry layoffs, the rapid rise of AI, and global turmoil, there were also many quieter, internal shifts that tested my resilience. Navigation during turbulent times isn’t easy, but it is deeply instructive.

Here are 8 hard-earned truths from my 8th year at Meta:

  1. Don’t fall for vanity metrics. It is easy to celebrate flashy surface-level numbers. The temptation is to treat metrics that are easy to track as core goals—until those proxy metrics become the target, optimization turns into gaming, and the whole system strays off course (just ask Goodhart). Genuine progress isn’t measured by easy counting; it is measured by sustainable impact, real utility, and long-term value creation. Nothing less, nothing more.
  2. Building (with AI) is fun again. Building has moved past the honeymoon phase of novelty. The initial thrill of “magic” has been replaced by the far deeper satisfaction of utility and craft. For a long time, I stepped away from hands-on building for all the usual reasons—time constraints, setup friction, and overhead. But tools like Claude (and Muse since it was hatched) changed the equation: they stripped away the setup friction, handled the tedious implementation details, and drastically shrank the gap between an idea and a working prototype. It freed me up to focus on product thinking, creative problem-solving, and user experience rather than getting bogged down in boilerplate syntax. I was able to revive old side projects, build utility tools I’d been wishing for, and even push code to production for the first time since my onboarding week.
  3. Endless learning. Change has been a constant throughout my career, but the sheer velocity of shift this past year raised the baseline—learning new things felt like table stakes just to stay afloat. The real currency hasn’t just been knowledge, but adaptability. Stagnation is no longer just standing still; it is actively falling behind. Staying curious, investing time to understand how new tools work and what they can offer, and constantly questioning my own assumptions became a weekly ritual rather than an annual reflection. We are playing Alien Chess, after all.
  4. Having opinions is sexy again — don’t delegate your critical thinking. In an environment saturated with agentic data collection and automated analysis, human judgment is now at a premium. While AI can build whatever you ask for—and even generate the documentation to justify it—it takes conviction, intuition, and strategic clarity to define what actually matters. The temptation to let automated consensus or processes make the call is high, but the most crucial decisions still demand a point of view: one informed by data, but owned entirely by you.


  5. Leading at a time of turmoil. Leading when everything is calm is challenging; leading through layoffs, strategic pivots, and global tension redefines the entire job. It requires providing psychological safety while remaining uncompromising on standards. I learned that during periods of extreme turbulence, your team doesn’t need false certainty or rehearsed corporate speak—they need radical transparency about the challenges ahead, absolute clarity on what matters today, and a steady hand on the wheel when the map keeps changing. This past year reinforced that it is precisely in the most troubling times that true leaders surface—not by avoiding the storm, but by being the calm reference point that others can anchor to.
  6. From chatting to delegating. This year, my relationship with delegation evolved fundamentally, applied equally to how I lead people and how I leverage AI. In both interfaces, it shifted away from narrow directives, frequent check-ins, and ongoing alignment. Instead, I was delegating entire bodies of work and taking myself out of the loop. When you step back from managing how the work is done, establishing clear intent, boundaries, and measurable objectives becomes absolutely critical. This high-trust, outcome-focused model is the only way to scale—whether you are leading a talented team or directing agentic systems.
  7. Communication is the ultimate hard skill. We often label communication as a “soft skill”, but in times of rapid change, organizational restructuring, and high ambiguity, it is the hardest technical discipline there is. Technical expertise gets you in the room, but precise, empathetic, and transparent communication is what actually drives alignment and moves mountains. When everything is in flux, ambiguity is toxic—and the ability to frame complex problems, deliver hard truths with candor, and align diverse teams behind a single intent becomes the ultimate high-leverage skill.
  8. Culture is built over years and can be shattered in one decision — values are the most important thing that brings people together. Strategy, organization charts, and product roadmaps evolve constantly, but culture is often assumed to be an enduring constant. However, culture is fragile. It takes years of consistent choices, trust-building, and shared experiences to cultivate, yet a single decision can break it overnight. When structural shifts or rapid change disrupt the familiar culture, you can’t rely on office perks or old routines to keep people together. In those moments of transition, it’s core values—the foundational principles of how we act and treat each other—that are the anchor that keeps people grounded, aligned, and united. And if those core values are scarce, retention becomes an uphill battle.

Entering Year 9 means that Meta is officially becoming the longest I have ever stayed at a single company. Crossing that line says a lot about this place, the great people I work with who make it worth showing up for every day, and how much I’ve grown along the way. Onward!

Don’t Stop Using AI (less of an argument and more of a plea)

Featured

Posted on by
Reply

A writer recently published an op-ed. It’s less of an argument and more of a plea. He asks that you never use AI to help you write.

Never let AI write for you – whether it’s school assignments, emails to colleagues or friends, an opening hook. Avoid doing any of this because you might actually learn to write better than the AI can. Writing with AI is problematic because it weakens us mentally, he argues. And as it becomes more pervasive, AI damages not only our individual capacity to create, but also our collective ability to explain, to innovate.

Well, I also get frustrated when I read “AI Slop.” But there is a huge difference between that and refusing to use AI altogether.

Do you use a calculator to do math? Personally, I rarely do. I have strong math skills and can calculate quickly in my head, so I’ve never really needed one. Does that mean you shouldn’t use a calculator because it “mentally weakens” you? Of course not. If a calculator complements your abilities and helps you move forward to complete your task, then go ahead and use it with confidence.


I’ve always envied people who can paint with a brush and bring their imagination to paper. The same goes for gifted musicians. People are born with a specific set of skills, and they can only develop them up to a certain point. That’s just reality – not everyone will become a talented writer, even if they take writing workshops and read constantly.

Some people struggle to draft an email. Some write messages that aren’t clear enough. Some don’t know how to finish a thought, while others thrive if you just give them a rough draft to polish. All these people can, and should, use AI. Just like the calculator, it helps them bridge a gap where they are less skilled.


Will relying on AI prevent them from ever developing that skill? Maybe. But let’s be honest: we all have weaknesses, and most of them don’t magically improve over time. Historically, we just avoided them. Can’t write? I just won’t write. Can’t paint or play an instrument? I’ll express myself another way.

But today, AI can help you – help you write, assuming you have an idea and the ability to edit and refine the text until it truly represents you. Help you create images, visuals you could never draw myself, but could clearly see in your mind’s eye. And music? I enjoy using AI to bring to life the sounds I hear in my head but lack the technical skill to produce.


So here is my statement, which is also less of an argument and more of a plea:
Don’t let AI do your job for you. If you let it do the thinking, writing, and editing while all you do is copy-paste, then don’t. But absolutely use it as a tool—as a force multiplier, as an engine. Use it especially where it can help you overcome fear, push past your limitations, beat your weaknesses, and become better at what you do.

* This post was written with the help of AI

The 10x Trap

Featured

Posted on by
Reply

A post written by Meenakshi Tripathy reminded me of an old Meta motto: you shouldn’t confuse speed with progress.

Personally I’m less concerned about PMs writing code to production. Because PMs should, and will, focus on orchestration. I worry about Engineers writing 10x more code, because, well, they can. And are also, in 2026, expected to.

We’ve all seen the demos. An agent spins up a plan, touches 15 files, and hands you a completed feature in 45 seconds. It feels like 10x. But in a high-scale environment 10x speed without 10x rigor is just 10x gap.

“Speed is the form of ecstasy the technical revolution has bestowed on man. As opposed to a motorcyclist, the runner is always present in his body, forever required to think about his blisters, his exhaustion; when he runs he feels his weight, his age, more conscious than ever of himself and of his time of life. This all changes when man delegates the faculty of speed to a machine: from then on, his own body is outside the process, and he gives over to a speed that is noncorporeal, nonmaterial, pure speed, speed itself, ecstasy speed.”
– Milan Kundera, Slowness.

As we move from “Copilots” to “Agents” we are hitting a friction point where our traditional safeguards are buckling. Here is where the real challenge lives IMHO:

1. The “Diff Review” Fatigue
When a human writes a 20-file diff, they can explain every line. When an agent does it, the human reviewer (and the author!) often defaults to “vibe checking”. If the logic looks sound and the tests pass, we ship. But “looks sound” is how subtle architectural drift happens. We risk becoming “Code Janitors”, only cleaning up the obvious mess while missing the underlying rot.

2. Testing vs. Verification
Agents are great at writing tests that pass. They are less great at knowing if they are testing the right thing. At our scale, “it works” isn’t the bar; “it doesn’t break a downstream dependency three layers away” is. Agents often optimize for the local task but lack the global context of our massive infra.

3. The Experimentation Blindspot
In a culture driven by A/B testing and stats, agents often miss the “why”. They can implement a feature, but can they reason about the long-term impact on a specific metric or the edge cases of a complex rollout? We risk shipping features that are technically correct but strategically hollow.

4. The Standard is the Ceiling
A “high standard” isn’t a checklist; it’s a culture of craftsmanship. If we outsource the thinking to agents, we lose the “why” behind our best practices. The risk isn’t just bad code – it’s the atrophy of engineering intuition.

5. Speed Breeds Silos
When an engineer’s output becomes “agent-accelerated”, a dangerous velocity gap opens up between the code and the rest of the team:
i. Engineer-to-Engineer Silos: If I can prompt a complex refactor in an hour, my teammates – who are still managing their own workloads – no longer have the bandwidth to truly understand my changes. We stop being a cohesive team and start becoming a collection of “solo-operators” who happen to share a repository.
ii. The XFN Black Box: Product Managers, Designers and Data Scientists rely on the “friction” of the development process to provide input. When the time from Idea to Diff shrinks to near zero, we bypass the critical “Wait, should we even do this?” conversations.
iii. Technical Isolation: In the rush to ship, we stop documenting the why. If an agent generates the solution, the engineer might not fully internalize the tradeoffs, making it nearly impossible to explain the technical debt to a PM or a successor six months down the line.

We aren’t just moving faster; we’re moving away from each other. High velocity without high communication turns “collaboration” into “notification”.

So what is the path forward? I think we should use the time saved by agents to focus on these principles:
1. Review for Intent: Don’t just check if the code runs; check if the agent understood the system.
2. Deliberate Friction: Over-communicate with XFN. 10x coding requires 10x alignment.
3. The “Janitor” Rule: If you wouldn’t have written it that way yourself, don’t let the agent ship it.

Speed is a vanity metric. In the race to increase speed, rigor and impact, we should stay present and conscious. Because quality and collaboration are what keeps us going in this long-distance race.