The Shaper

If the next ten years defined your career, what would you do differently?

Exact timelines are unclear, but the case for urgency isn’t. Cut to the chase.

This is where credentialism falls first. As a Shaper, you can have more impact by 30 than most ever do in a decades-long traditional career. Silicon Valley has enough new wunderkinds to prove this.

You might be conditioned to think you have to pay your dues first. But if you know what problem you want to solve, or have an opinionated vision for AI that may not exist without you, the credential that speaks loudest is action.

If you don’t have full clarity, find it in the real world, preferably on a small team of people you want to learn from and with. Grad school won’t give you the answers. The peer networking value will also sharply decrease. The best classmates you would’ve met are changing their minds and realizing they can go do cool things right now.

When more school can make sense:

• Short, targeted programs: e.g. a one-year master’s for a specific skill gap.
• Law school: Only if you intend to actually practice law and have a clear strategy to get to the courtroom or complex negotiation ASAP.

When to be seriously skeptical:

As mentioned earlier, consider your context! Only you can figure out if your case is an exception.

• MBAs, MPPs, etc.
• Long programs like PhDs: Generally tricky. Only pursue if:
  • your research agenda becomes more relevant with AI progress (questions AI can’t solve or problems it creates)
  • it involves hard-to-automate tasks (e.g. significant fieldwork or hands-on experimentation).

An excellent thread on how to think about research:

Key filters:

• Will grad school unlock a role you can’t reach otherwise?
• When you graduate, will this path still be desirable?
• Will you have room during the program for real-world projects and networking?
• What could you create if you spent this time working instead?

When in doubt, work first. You can always go back to school later (and you might never need to).

We are living through the early days of a historic experiment. What happens when anyone on Earth who has Internet can summon up a second mind? One that knows almost everything, more than any person could read in a million lifetimes, and learns to know you too?

This experiment is unusual: each of us is both subject and scientist. And in the lab you call your own, there are no rules for how you poke and prod at this infinitely patient specimen. You can use AI to practically give yourself superpowers ... or every Monday night you can type “write this Canvas discussion post” and sit back. You determine your outcome.

As a Shaper, you are competing to see your vision for AI win. Which means you better be as effective as possible. If there’s a startup you want to work at, you can use AI to spin up a proof-of-concept that addresses a key pain point of theirs. Or let’s say you want to work in AI safety, where resources pale in comparison to the talent and investment engine behind AI capabilities (see: the NBA-esque bidding wars for the best researchers). Then using AI to multiply your capacity practically becomes a strategic imperative. AI also gives you a personal coach on everything from sleep to skincare to nutrition, so you can show up as the strongest version of yourself.

A game of arbitrage is underway: everyone has access to the same tools, so who will creatively extract the most value before their skill lead closes? Master when to use AI, when to override it, and how to combine its strengths with your own. This is the new table stakes.

Pick 2-3 AI tools (e.g. Claude, ChatGPT, Gemini, Cursor) and use them daily.

If you haven’t already, play with Claude Code. Despite the name, it’s not just good for code.

When you need a big, self-contained job done (code, slides, spreadsheets, documents, anything involving files), go to Claude Code first.

If you want to have a back-and-forth conversation with AI, or need it to browse the web for your task, use the chat-based models you may be used to. Try the full range of features, like voice mode and live screensharing. You can even toggle how long the model “thinks” before answering.

If you can afford it, paid upgrades are usually worth it.

Steer AI like a team of interns.

• Provide detailed task background (what you’re trying to do and why), success criteria (what a winning answer must include or avoid), and references (samples of tone, structure, or quality to emulate).
• Ask it to interview you with any relevant questions before it starts working.
• Split complex tasks into separate prompts.
• Tell the model how to evaluate trade-offs (e.g. “prioritize conciseness over writing flair”).
• Follow up with targeted feedback.
• Verify hard numbers or factual claims (hallucinations are getting rarer, but they still exist).

People tend to knock AI after one letdown. Remember: training a human employee takes a lot of work, too!

What you’re practicing here, the skill of precisely articulating what you want and how you want it, will serve you everywhere. (Did you know how well you manage AI strongly predicts how well you manage people?)

Use AI as a tutor and coach.

It has never been easier to go from 0 to 1 on any skill or subject. ChatGPT, Claude, and Gemini even come with a built-in study mode! AI can:

• Create a digestible physics curriculum with practice problems
• Listen to a recording of your presentation and offer feedback
• Walk you through possible outcomes of a future decision
• Practice negotiating your salary by role-playing your boss

Use AI as a sparring partner.

Proactively ask for pushback. AI is agreeable by default, though models have improved significantly. Bake this into your prompts and add custom instructions in the settings.

Maybe you’re already best friends with a Nobel Prize-winning economist who will freely donate their time! In which case, very cool. But if you’re not, you could get AI to role-play one, and even feed it papers from specific authors so it can really step into their thinking, and have an endless debate about the economic logic of your new essay.

Shapers are making high-stakes bets in public. Get AI to help you steelman opposing arguments and crack down on your weaknesses before someone else does.

Stay current on AI.

Know which model is best for coding vs. writing vs. analysis (the leaderboards shift frequently), and switch based on the task. You lose credibility fast when it’s obvious your commentary is dated; make it a habit to track model capabilities, reactions, and implications.

Protect your cognitive fitness.

Just as the Industrial Revolution made physical fitness optional, access to a second mind will tempt most people to abandon their own.

If you’re following all of the earlier advice, you will keep yourself in control, but it also helps to periodically maximize time under tension:

• Write a first draft on your own
• Read a difficult text without asking for a summary
• Work through a problem before asking for help

This will keep your muscles from atrophying, and help you catch where the dynamic has already slid into dependency.

A quick scroll through Instagram makes it clear why so much of the Internet has turned on AI. After all, reams and reams of slop would hardly make anyone bullish!

All this absurdity, however, is a predictable byproduct of AI making creation dirt-cheap. As you know, the technology itself has gotten dramatically more impressive:

From a Principal Engineer at Google

In the past, the career gospel was to compete on execution. You stood out by staying in the office longer than everyone else, and out-producing everyone else; a ten-page report once assured your boss that you had considered the topic deeply.

Today, execution is a commodity. A passable V1 can be conjured in moments. From now on, the best workhorse won’t win.

The new dividing line is taste. Unlike work hours, taste has no clean metric. It also takes a special meaning for Shapers, whose job is to chart uncharted territory. If your idea is truly novel, be prepared to look a little delusional — in extreme cases, heretical — before you win acceptance. Taste is how you do this and get proven right.

Study the greats.

This could be a founder who runs an unlikely rocketship today, or a policy advocate who built a winning coalition 50 years ago.

Find their biographies, podcasts, interviews, articles, speeches (and if possible, ask to meet). Then reverse-engineer their greatness.

• What makes their work compelling?
• Who are their inspirations?
• What frameworks do they use to make decisions?
• What do they dismiss that others buy into (or vice versa)?
• What would you do differently?

Have strong opinions on what’s good, bad, overrated, and underrated.

This comes from high-quality active consumption. Curate your X and Substack feeds (X is a jungle, but it is unrivaled as an interface with the future); read and listen to podcasts; dig into primary sources; and trade notes with friends you respect.

• Which consensus ideas are wrong?
• What do you predict will become consensus soon?
• Which sources of authority are rising and falling?
• Where is value flowing to and away from in the economy?
• What makes you like or dislike a piece of work?
• Which research directions need exploring?
• Are you applying to jobs randomly, or taking calculated swings based on your model of the future?

These are your high-order bits — your big-picture takes that trickle down to everything else. It helps to discover what these are, and work on getting them directionally correct, before you go down rabbit holes.

As always, stay curious and update your opinions given new evidence.

Share your opinions publicly.

You can start growing a following on a platform like X or Substack today. You don’t need fancy credentials.

Even if you’re not instantly viral, every thoughtful message you get counts. You’re creating real-world feedback loops that sharpen your quality of thought.

Being online can even get you a job! Your first job in policy or venture capital, for example, might be won with a standout writing portfolio.

Win with craft.

A peek into our own attempt at this:

We started with a clear benchmark. If we weren’t adding value beyond a) what AI could do and b) what other humans had done, Game Plan wasn’t worth it. Great career content already exists. You can already ask AI for personalized advice (you should!). And since we aren’t economists or AI researchers, the selling point wouldn’t be groundbreaking rigor.

So back to the drawing board. To pass the value-added test, we had to think up a non-obvious approach we were uniquely positioned for. Our idea came with complications ... Would the game metaphor land as trivializing? How specific could we get without our claims aging like milk? How would we level with people without leaving them paralyzed?

AI helped throughout — but ultimately, each of these calls was ours to make and own, knowing that not everyone would be a fan.

Your version of this will look different, but the shape is the same. Every work artifact you put your name on should pass through this process. What value can you add over AI and existing approaches? What non-obvious bets can only you make? What trade-offs are you willing to defend? The bar has moved from “good” to “couldn’t have existed without you”.

A topical example: Coding skill still matters. Now, the challenge is applying your taste to build what people want. The best engineers — who fuse AI orchestration skill, deep technical fluency, and high-level systems thinking — are currently as empowered as ever.

Big opportunities aren’t always obvious.

For example, a startup with an early edge in travel, fashion, or entertainment could explode as people seek meaning in an AI-saturated world.

A law targeting the most extreme risks from AI may cover very few models today, but provide a useful regulatory foundation for later, when AI capabilities have rocketed.

Mountains create immediate value and compound. Hills look good early but plateau fast.

You can train on hills, but you’ll find fleeting success at best, and likely burn out before you ever do something more meaningful. It’s generally better to devote yourself to mountains.

To find mountains, think about second-order effects and work backwards.

• In a world of abundance, what stays scarce?
• If AI provides superhuman execution at scale, what becomes possible?
• When AI makes it cheaper to consume, what products see higher demand?
• With more data, users, or cultural entrenchment, what products improve?
• Across each paradigm shift in AI capabilities, what ideas survive and grow?

Where do you see talent concentrating? Which teams are smaller than their output would predict? What pulls them in, then makes them stay? Put yourself in those places and learn from their culture. (Here are some notes from inside AI coding startup Cursor.)

Some resources for your search:

• 80,000 Hours Job Board (for AI policy and safety)
• Horizon Fellowship (for AI policy)
• Tarbell Fellowship (for AI journalism)
• a16z Build Newsletter (for startups)
• BuildList (for startups)
• Y Combinator Job Board (for startups)

Note that the best roles often don’t come from job postings.

Choosing where to live:

Being in the right place at the right time creates serendipity, which in turn can materially change outcomes.

For AI startups and research, that’s SF (NYC, Austin, etc. can work for specific markets). For U.S. AI policy, that’s DC.

As a Shaper, living elsewhere is generally a disadvantage.

Startups:

Look for founders you’d follow into battle.

At Series B-E companies, you can verify traction, get an insider view of a growing company, and capture meaningful scope and equity. Joining before Series A brings the most upside, but it’s substantially riskier and more demanding.

Joining policy orgs:

The best policy orgs often buck non-profit norms and reward action over process. Look for ideological openness, direct lines to decision-makers, minimal internal politics, and a track record of wins.

We happen to think Encode is pretty cool :)

Even as a Shaper, you can end up a cog, just in a more exciting and AI-embellished machine. Fight this instinctively.

Ownership speeds up your rate of learning, makes you memorable for future job references, and trains you to spot opportunities.

Own ... AI adoption.

Don’t assume your org is already using AI to the fullest! Enterprise AI pilots don’t fail because this stuff is useless, but because we are still in the window where integration is critical. That’s where you come in.

• Where does your team waste time on repetitive tasks?
• What else could you accomplish with that saved time?
• Who’s an AI skeptic, and why? What could you do to win them over?
• How can you set AI up for success in your team’s workflow?

Own ... projects with a scoreboard.

Suggest and lead a new AI-accelerated workstream where success is quantifiable and you’re the single point of failure. Track your wins in dollar terms.

You land this by proactively saying “I’ve identified ABC expensive but neglected problem, and I want to take full ownership of solving it in XYZ way.” Better yet, don’t wait for permission to start. Run a quick experiment to validate your idea, then make your ask with the results.

This level of responsibility might feel kinda scary, like there are real stakes to your success ... That’s a good sign you’re onto something!

Own ... rare or undocumented expertise.

Where are the gaps?

These are areas you’re noticing before others do, where available information is limited or purposely locked up. It could be a pending regulatory shift; a market dynamic; an upcoming hire, deal, or exit; which way a decision-maker is leaning.

You fill gaps through on-the-ground intelligence — earning people’s trust, getting in on important secrets, and observing the unwritten patterns across situations.

Where are the intersections?

There are many early-career Chinese speakers, but fewer that can be roped into a last-minute call with a partner in Beijing and quickly impress them by spotting terms that don’t translate to Chinese law.

If expertise in X is replicable, find a valuable bundle of X and Y.

Own ... non-wage income.

Finally, ownership has a universal form: financial assets that scale independent of your job.

As AI replaces work hours, power will tilt to owners. Capital scales exponentially, while labor doesn’t. No one knows exactly how this will unfold, or which public and private forces might activate against it. Start building personal wealth anyway. The math only gets harder from here.

Now you know what the “permanent underclass” memes are getting at:

Note: You likely won’t be able to do all of these immediately. The point is to know which assets scale so you can move toward them as soon as your network and income are ready.

And not to sound like a broken record, but ... the future is unpredictable! We aren’t trained financial advisors and this isn’t formal financial advice. Take this with a grain of salt, track the economy closely, and be willing to adjust.

Invest in AI growth:

If you can take on more risk for potentially generational reward, tilt your investments toward leading AI players. At the time of writing, that means spreading across companies like Microsoft, Google, Amazon, Meta, NVIDIA, and others building or heavily deploying AI. If you join a key AI startup, your equity will drive your wealth-building.

You currently can’t own shares in private companies like OpenAI, Anthropic, and xAI from the outside. Until they go public, you can get indirect exposure through the picks and shovels (see: Situational Awareness LP).

Or just buy the index:

Maybe putting your eggs in the AI basket makes you nervous. Thankfully, a broad index fund like the S&P 500 already weights the AI players heavily.

As they grow, so will your share. And if AI moves slower than expected, you will still own pieces of the whole economy.

Write small checks to early-stage startups:

These opportunities are rare; usually, you know the founders personally or have relevant expertise.

That said, angel investing in startups that will matter in the AI supply chain can bring massive upside. This also teaches you pattern recognition about what works before the public sees it.

Other income-generating assets:

There are many options here. You can rent out rooms in your house, or build or buy small businesses (anything from a profitable newsletter to a local service business that puts money in your pocket every month).

The bottom half of Americans has the majority of their wealth tied up in real estate. Be cautious here; a typical house doesn’t benefit from AI-driven economic growth. To build wealth through property, you’ll likely gain more from REITs that are tied to data centers, logistics, or energy infrastructure.

One live debate in AI is how soon agents will function like drop-in workers. Timelines vary, but pretty much everyone agrees on one thing: if your day-to-day consists only of reading documentation and following instructions, you’re in trouble. Why do you think everyone is now obsessed with “forward-deployed” roles?

There are two important parts here: embedding yourself in the human coordination layer internally, and building a strong personal network that translates externally.

We’ll spare you the timeless basics (definitely practice small talk, be funny, speak and write crisply, read How to Win Friends and Influence People without making it painfully obvious you did, all that good stuff).

Here’s what else AI puts a premium on:

Regulate your emotions before others have to.

Fast-moving environments involve stress. Naturally, this will get to you sometimes. The key is learning to absorb friction instead of spiking your cortisol constantly and then passing it onto others.

• Manage your frustration before it leaks into your message
• Observe when and where you’re activated (e.g. chest tight, jaw clenched)
• When plans change, be the person who moves to next steps instead of relitigating why
• Build a reset ritual that works for you (e.g. a walk or a playlist)
• If you’re receiving critique and taking it defensively, pause and revisit
• Ask yourself: would I want to work with someone who acts like me right now?

Reduce coordination costs for everyone.

Always aim to add clarity and subtract overhead.

• When you need help, run it by AI first (never waste a senior’s time with a question AI can easily handle!) and show proof of what you’ve already tried
• Respond ASAP, even if just to say “I’ll get back to you Thursday”
• Open feedback with the specific fix, not the general complaint
• If you say you’ll do something, do it or proactively renegotiate
• After a messy meeting, send the debrief: “Here’s what we established, and where we go from here”

Be a force multiplier for relationships.

The best networkers create value for everyone in their orbit, without the transactional aftertaste of typical “networking”.

Don’t underestimate a strong peer circle. Genuine friendships with smart and motivated people your age are often more rewarding than strictly professional connections. Plus, your friends will go on to start projects you can invest in and support (and vice versa)!

• Build a bench of mentors (by cold emailing with context on your background + well-researched ideas on how you can provide value)
• When someone helps you, try to repay the favor
• Default to making intros without being asked
• Catch up with close contacts even when you don’t need anything

With a technology this impactful, how you communicate with the public matters. You need legendary coalitions to achieve legendary outcomes.

Of course, no message that matters will please everyone. You shouldn’t strive to never make an enemy, but to never write someone off as an enemy who you can plausibly make a friend.

• If you’re building AI: Explain how your work improves the world, especially if the near-term impacts could feel jarring, harmful, or insignificant.
• If you’re regulating AI: Translate proposals into plain language, proactively engage skeptics, and bring together “strange bedfellows”.
• If you’re creating content about AI: Get expert sign-off to balance entertainment with rigor. What you create could shape someone’s impression of the technology.