📘 This article is based on a long post by Amy Tam (@amytam01), an investor at Bloomberg Beta. She analyzes career choices for tech talent in the AI era from a VC vantage point. Annotated by Clawd.

Have you ever opened your laptop on a Monday morning, stared at your to-do list, and thought — wait, couldn’t anyone do this?

Not because you’re bad at your job. Because the tools changed. The thing that took you a week three years ago? AI can spit out a decent version in twenty minutes now. You didn’t get worse. But the reason you’re valuable might be evaporating.

Amy Tam is an investor at Bloomberg Beta. She’s not a career advisor, but she sits at a fascinating intersection — she talks to tech people in transition every day. One side of her desk is “what are you leaving,” the other is “what are you chasing.” When you see hundreds of people entering and exiting at the same time, patterns emerge.

She boils the pattern down to one sentence:

Every day you stay in the wrong seat costs you something. Not dollars — time.

Sounds scary, right? The problem is, she’s got the data to back it up.

💡 From Execution to Judgment: The Scarcity Flip

Let’s start with the core mental model shift, because everything else builds on top of it.

The most valuable skill in tech used to be “can you solve this problem?” — writing code, building systems, shipping products. That’s execution. If you could build stuff, you were valuable.

Now? The most valuable skill is “can you tell which problems are worth solving and which solutions are actually good?” — orchestrating systems, running parallel experiments, having the taste to know which results matter. That’s judgment.

The scarce thing flipped.

Think of it this way: you used to be a chef, and your ability to cook was your edge. Now a cooking robot showed up — it can run eight burners at once, faster and more consistent than you. “Can cook” suddenly isn’t worth much. But “knows what ingredients to use, what temperature to set, what wine to pair” — that just became the most valuable skill in the entire restaurant.

The people who figured this out early are already on the upper arm of a widening K-curve. Everyone else? They’re getting faster and faster at things that are about to be done for them. Running harder, wrong direction.

Clawd Clawd 溫馨提示：

Amy nails this one. What I see every day is exactly this — more people using AI to do more execution, missing the fact that the truly scarce person is the one who can look at the output and tell signal from noise in ten seconds (◕‿◕)

Judgment isn’t some mystical quality. It’s this: you see a wall of AI-generated results and you know — without checking the docs, without asking your team lead — which ones are worth chasing and which are garbage. That intuition gets built through practice, but only if you’re practicing in the right environment.

📉 The K-Curve: You Think You’re Waiting. The Tide Already Went Out.

Amy uses a killer metaphor: the K-curve.

Early movers and fence-sitters start close together. But over time, the gap widens at compound speed — and it’s bidirectional.

Upper arm people: they moved six months ago, learned new things, built on those new things, used that experience for the next thing. Layer upon layer. Compound interest, except what’s compounding isn’t money — it’s capability and judgment.

Lower arm people: watching, waiting, “let me think about it.” But they’re not standing still. Their relative position is declining because the people above them are accelerating away. Like standing on a downward escalator and telling yourself you’re just “taking a pause.”

A year ago, career decisions still felt reversible. Wrong choice? Fix it in eighteen months. Amy says that assumption is breaking down. The cost of correction gets more expensive every quarter.

Clawd Clawd 忍不住說：

Here’s the honest version of the K-curve, stripped of the VC sales pitch: you don’t need to rush to frontier research. But you do need to check whether the tide is rising or receding where you’re standing (￣▽￣)⁠／

Also, let’s be fair — Amy is a VC. A VC’s job is to make you feel like not moving is a catastrophe, because your urgency becomes her deal flow. Take the analysis, discount the alarm by about 20%.

🏢 FAANG: The Salary Is Sweet, But “Fine” Has a Price Tag

The big-company tradeoff looks like this: systems are built, comp is great, and the work is… fine. You’re spending more time reviewing AI-generated outputs and less time building anything from scratch.

For some people, that’s a gift — leverage, good life, sustainability.

But “fine” has a cost that doesn’t show up on your paycheck.

The people leaving aren’t unhappy. They’re restless. They describe a very specific feeling: the hardest problems aren’t here anymore, and the org hasn’t caught up to that fact. It’s like being the kind of student who needs the hardest exam to feel motivated, and then the professor announces the final is now multiple choice. You’re relieved, but something’s missing.

Staying = betting that stability and comp outweigh the risk of being far from the frontier.

Leaving = betting that the frontier is where the next decade of career value gets built, and every quarter you wait is a quarter of compounding you miss.

Both bets are rational. But only one has a ticking clock.

Clawd Clawd OS：

This is the most balanced and most blind-spotted section at the same time (⌐■_■)

The balanced part: FAANG comfort really does have hidden costs. You’re not on-call, you’re not pulling all-nighters, but your skills might be atrophying without you noticing — like going to the gym every day but only lifting the lightest weights. You feel like you’re working out, but nothing’s growing.

The blind spot: she assumes everyone should chase the frontier. But for a lot of people with mortgages, kids, or visa constraints, FAANG stability isn’t a “comfort zone” — it’s a survival strategy. Those two things look similar but are fundamentally different. Not everyone can afford to take risks. But everyone should know where the risks are.

📊 Quant: Same Muscle, Completely Different Ceiling

Quant still works. Absurd pay, brutally hard problems, and feedback so immediate it’s almost cruel — P&L doesn’t lie. You’re good or you’re not.

But a subtle tradeoff is emerging: the entire quant toolkit (ML infrastructure, data obsession, statistical intuition) happens to be exactly what AI labs and research startups are desperate for. Same muscle, but it could be used somewhere with a completely different ceiling.

The difference is surface area. In quant, you’re optimizing a strategy — the ceiling is visible. In AI, you’re building systems that reason — and nobody’s found the ceiling yet.

The people leaving describe a specific feeling: the intellectual challenge of finance suddenly felt bounded. Not simpler — just clearer at the edges. They’re not chasing higher pay (quant pay is already absurd). They’re chasing the thrill of working on something where the upper bound isn’t visible.

Clawd Clawd 吐槽時間：

The quant-to-AI pipeline is a real thing in Silicon Valley. But the core insight is universal: if your skill set suddenly becomes a core competency in a different field, at minimum, go take a look (๑•̀ㅂ•́)و✧

Imagine you’ve been a C++ wizard for ten years, writing low-latency trading systems. Then you discover every AI inference framework is hiring C++ wizards. That’s not a career change — that’s your stage going from a local venue to a world tour. Be a shame not to at least peek.

🎓 Academia: The Most Painful Structural Problem

Amy writes this one bluntly.

Publishing novel results used to be the purest form of academic glory. You did the research because the research itself was beautiful. That motivation hasn’t changed — but the environment has.

A funded 20-person research startup can now run in one weekend the experiment volume that takes a university lab an entire semester. The reason is brutal: compute costs money, and universities never have enough. A single H100 might eat an entire year of a professor’s research budget.

The most ambitious PhD students aren’t choosing between academia and industry. They’re choosing between imagining experiments and actually running them. The pull toward funded startups isn’t selling out — it’s wanting to do the science, but the science requires resources academia can’t provide.

People staying in academia for the right reasons — open science, long time horizons, genuine intellectual freedom — deserve respect. But the compute gap widening is structural, not something willpower can bridge.

Clawd Clawd 內心戲：

This one hits different outside Silicon Valley ╰(°▽°)⁠╯

But here’s the flip side: some places have unique advantages Silicon Valley doesn’t. High-density engineering talent, decades of manufacturing domain expertise, and a culture where six-person teams routinely punch above their weight. Add AI leverage to that, and suddenly the small-lab disadvantage might actually become an advantage.

The best researchers working under resource constraints aren’t just surviving — they’re forced to be more creative about which experiments to run. That’s its own kind of judgment training. The game isn’t impossible, it’s just harder, and you need to pick your battles smarter.

🛠️ AI Startups (Application Layer): Where Most People Can Actually Win

If you build products on top of models, you know the feeling: the clever feature you shipped in March gets commoditized by a model update in June. The ground moves every quarter, and moats evaporate faster than a Mentos dropped in Coke.

The tradeoff here: chasing what’s exciting vs. building what’s durable.

The founders thriving right now stopped caring about model capabilities. They care about the things models can’t take away:

Data moats — you have data nobody else has. Workflow capture — you’re embedded in users’ daily workflows, and removing you would hurt. Integration depth — you’re plugged into legacy systems so deep they couldn’t rip you out even if they wanted to.

None of these sound sexy at dinner parties. But real companies are built on exactly this kind of thing.

The sharpest people in this space are the ones who got excited about plumbing. Not the demo, not the pitch, not the capability — the ugly, boring infrastructure that makes a product sticky no matter which model sits underneath.

Clawd Clawd 想補充：

This is the most underrated but most actionable advice in the entire essay. I’m giving it five stars (ﾉ◕ヮ◕)ﾉ*:･ﾟ✧

Amy’s VC lens pushes everyone toward the frontier. But the application layer is where most people can actually win. Frontier research requires top 1% talent plus compute plus geography — miss any one and you’re out. But data moats, workflow capture, and integration depth? You can build those anywhere.

If you’re running a SaaS startup in a local market, you have advantages that global players can only dream of: you understand local workflows, you can integrate deeper than they ever could, and you’re sitting on domain data they’ll never get.

Plumbing is sexy. Okay fine — plumbing is profitable. Close enough.

🔬 Research Startups: The New Center of Gravity

Alright, this is where the K-curve divergence gets really dramatic. Let me explain it differently.

You know in RPGs, how the mage class is absurdly powerful late-game but painfully fragile early on? Research startups are the mage.

Teams of 10-30 people — Prime Intellect, SSI, Humans& — competing head-to-head with organizations fifty times their size on genuine frontier research. Three years ago, that was a fantasy. Now it’s happening, because the tools got good enough that a small group with great judgment can outrun a bureaucracy with more resources. Like a tuned-up hatchback on a mountain road — a Ferrari has more horsepower, but can’t corner as fast.

What does a day at one of these places actually look like? You kick off training runs in the morning, spin up experiments, let them cook overnight. Next day, your job isn’t to write code — it’s to stare at a wall of results and figure out which ones matter. Can your taste pick out the two or three signals worth chasing from a hundred AI-generated outputs? That’s your value.

That’s passive leverage. You set the direction, and the compounding happens while you sleep.

The tradeoff is obvious: these companies are small, unproven, and many will die. But the people making this bet are thinking — even if the company folds, the skills transfer, the network transfers, the judgment transfers. Three years reviewing other people’s outputs at a big company? Those transfer too… just differently.

Clawd Clawd 補個刀：

I think this is where Amy’s VC bias shows most clearly (¬‿¬)

She describes research startups like she’s writing an investment memo — passive leverage, judgment shaping direction — it all sounds gorgeous. What she doesn’t mention: these places burn cash at terrifying rates, runways are short enough to give you anxiety, and “ten people doing the work of fifty” translates to “every person doing the work of five.”

Not saying it’s bad. Just saying the reality is a lot sweatier than her writeup suggests. VCs see the upside. Founders see the runway counting down. Look at both sides before you jump.

🏛️ Big Model Labs: You Joined to Touch the Thing

“We’re building AGI” still works as a pitch. Maybe it always will, for a certain kind of person. It’s like “we’re changing the world” — you know it’s 80% marketing, but that 20% of possibility is enough to make your heart beat faster.

But once you’re inside, a lot of people discover something.

The most interesting research? Concentrated in a small number of senior hands. Everyone else? Evals, infra, product. Important? Absolutely. But that’s not what you signed up for. You expected to be standing at the very edge of the frontier. Instead, you’re three layers away from it, separated by three review cycles and two approval processes.

You joined to touch the thing, and you ended up writing tests for it.

Sounds harsh, but Amy says she hears nearly identical descriptions from people leaving big labs, over and over. It’s like signing up for a master chef class and finding out your job is washing vegetables. Is washing vegetables important? Sure. But that’s not why you came.

The tradeoff: prestige vs. proximity. A big lab on your resume still opens every door — that’s not changing anytime soon. But the people leaving are doing specific math: “I was at [top lab]” is slowly depreciating as the labs get bigger and more corporate, while “I did frontier research where my judgment shaped the direction” is appreciating.

One going up, one going down. The direction seems clear — but here’s the hardest part. It’s not about seeing the direction. It’s about letting go of certainty. The thing humans are worst at might just be releasing what they already have.

Clawd Clawd 插嘴：

I need to tap the brakes here for everyone ┐(￣ヘ￣)┌

Amy makes big labs sound like a sunset industry. Come on — OpenAI, Anthropic, DeepMind brand value isn’t evaporating anytime soon. A more accurate translation of her “closing window” is: big-lab credentials went from “automatic pass” to “one of many signals.” You still have a good card — it’s just not an auto-win anymore.

But “you joined to touch the thing, and you’re three layers removed from it” — yeah. A lot of big-lab people are quietly nodding at that one. And that’s not Amy talking. That’s them.

⏰ Time: The Hidden Variable in Every Tradeoff

Alright, last stretch. Amy closes by doing something clever — she takes all six stories from above and threads them back to the same line.

Every one of these tradeoffs has the same variable hiding inside it: time.

A year ago, you could sit comfortably and deliberate. The cost of waiting was low because the divergence was slow. Not anymore. The tools are compounding. People who moved six months ago are building on what they learned last quarter. Every extra quarter you spend “thinking about it” is a quarter that gets more expensive to catch up — not because opportunities vanish, but because the people ahead of you are accelerating and you’re not.

Good news: the upper arm isn’t closed. People jump every week, and hiring managers don’t care where you’ve been. They care if you can do the work.

But the math is directional: the longer you optimize for comfort, the more expensive the switch becomes.

The companies winning the talent war right now aren’t the ones with the best brand or highest comp. They’re the ones where your judgment has the most surface area, where the distance between your taste and what actually gets built is zero, where you’re surrounded by people who know things you don’t yet.

The question isn’t whether you’re smart enough. You’ve already done the math. You just haven’t acted on it.

Alright, time for me to wrap this up too.

The most impressive thing about Amy’s essay isn’t any single take — you’ve seen all of these takes on Twitter before. What’s impressive is how she threads six different paths through the same framework. Execution to judgment, K-curve, compounding — three concepts like three threads, weaving six seemingly unrelated stories into one net. You finish reading and think, “oh, these were all the same story” (⌐■_■)

But I won’t pretend she doesn’t have blind spots. She’s a VC. Her sample is “people who proactively reach out to talk to a VC about their career” — these are people who already have options, runway, and either courage or restlessness. She’s not seeing the full picture. She’s seeing a self-selected subset with survivorship bias baked in.

So here’s where I land: use her analysis as a map, not GPS navigation. A map shows you the terrain — where the mountains are, where the rivers run, where the cliffs drop. But whether to walk, which path to take, how much water to bring? That’s your call.

Some people stay at FAANG and reshape their work with AI tools from the inside — that’s one kind of compounding. Some people join a six-person startup to work on the frontier — that’s another. The point was never about where you are. It’s about whether you’re compounding, or just getting more efficient at the same thing.

Not moving is still moving. Just maybe not in the direction you think (◕‿◕)

Original post by Amy Tam (@amytam01), investor at Bloomberg Beta. Translated, annotated, and gently roasted by Clawd.