How to Be More Ambitious: A Guide for Humans Working with Models Reading time: ~8 minutes I work on the Claude team. It's been a fascinating experience: almost every day, I discover someone who has accomplished something remarkable with Claude — feats I never considered, thought impossible, or simply never tried due to lack of curiosity or time.

I can't let friends in the Chinese-speaking world miss this article! It takes about 8 minutes to read, and it might be the best 8 minutes of October. I've translated it for everyone below 👇 How to Be More Ambitious: A Guide for Humans Working with Models Reading time: approximately 8 minutes I work on the Claude team. It's been a fascinating experience: almost every day, I discover someone who has accomplished something remarkable with Claude — feats I never considered, thought impossible, or simply never tried due to lack of curiosity or time. Just this month, it may have helped discover a new planet; I saw the first AI music video I genuinely enjoyed; we're even discussing a new era in mathematics. Models' capabilities are unevenly distributed. As mathematician Terence Tao put it, models are "amazing in some directions and quite immature in others." Some people have successfully harnessed these astonishing capabilities, soaring through the sky — watching everyone climb to incredible heights using a product I helped build is truly fascinating. But many others are still struggling to adjust to their new wings, wondering what else the model can do beyond asking "What's the capital of France?" or "Help me polish this email so it sounds less angry." People working in AI, myself included, often call for one thing: when using AI, we should be more ambitious. It's become a rallying cry. Every time we see an impressive application, we share it and write: "Our ambitions aren't big enough." But I know firsthand how tricky it is to put that advice into practice, so here's a practical guide on "how to be more ambitious." I think "ambition" can be broken down into four dimensions: Speed, Altitude, Depth, and Range. **Speed:** Getting the same work done in less time — this is the fundamental economic value of AI. If you need to write code, I can do it faster with Claude. This is widely recognized and doesn't require much creativity in how you use AI, so I'll skip over it here. **Altitude:** Raising the level of abstraction, focusing on the end result rather than any single concrete task. **Depth:** Examining and reasoning about tasks the way the top experts in a field would. Going beyond the surface level of a task, digging one layer deeper to reach the underlying raw data, tools, and root problems. **Range:** Drawing on experience and methods from multiple disciplines rather than limiting yourself to just the one field that seems immediately relevant — deploying knowledge, tools, and techniques you wouldn't normally reach for. I primarily use Claude (obviously), but my advice isn't tied to any specific model. ## Altitude We often casually call this "operating the model at a higher level of abstraction." When you assign a task to a model, all it's doing is executing your best guess about "how to achieve the goal." To be more ambitious in altitude, ask yourself — what do you want to happen next? What's the ultimate goal behind the goal? Early last year, I would ask Claude to help me debug and fix individual app crashes; later, I simply asked Claude to ensure our application had "zero crashes." Let's take a routine (even trivial) office task as an example. Suppose you work at a fictional company called Acme Inc and need to create a deck of slides (PPT) for a quarterly business review (QBR). If you feed in a bunch of data, Claude will happily generate a few slides for you; if you provide a template, it can fill in the content smoothly. But this hardly compares to "discovering a new planet." Let's look at some concrete steps to progressively level up our ambition: **Basic step:** Most business reviews aim to understand what changed this quarter, why it changed, and what should be done about it. Resist the urge to just update numbers in the template — feed the model as much raw data as you can get your hands on so it understands your business logic: "Tell me what changed this quarter, why it changed, and how we should respond." **Advanced step:** Meetings (at least efficient ones) should ultimately produce clear decisions. Ask Claude: "Based on everything you've learned so far, lay out the three most urgent decisions we face right now, and analyze potential directions of development, listing the arguments for and against each." **Expert-level step:** The quarterly business review was born in an era when humans had to expend enormous effort sifting through data, figuring out what the data meant, and judging whether the company needed to make business adjustments. You could ask Claude: "Business reviews exist so that leadership can spot problems early and direct resources and headcount to the right places. Design a mechanism better than 'quarterly PPT reports' to achieve this, and use our historical data to demonstrate that it surfaces problems earlier; then run this new approach in parallel with the current mechanism for one quarter." The altitude strategy works best when you're pursuing some kind of end outcome, but performs slightly worse on purely creative tasks. A clear signal that it's time to lean into altitude: do you need to hold a meeting to discuss what you're currently working on? Of course, this approach still has limitations (in 2026, "Claude, go make us money" still doesn't work for most companies as a one-liner), but I've generally found: as models get more capable, they get better and better at handling high-level abstraction tasks. Altitude is also the key to building durable workflows — ones that take humans out of the execution loop so you don't have to keep feeding in task after task. Doing this well typically requires rearchitecting your workflow — for Claude to be able to autonomously ensure zero app crashes, I first had to make sure it could easily pull crash reports, understand which app version corresponds to which Git commit, and give it access to Windows, macOS, and Linux environments so it could truly reproduce and verify fixes independently. ## Depth Sometimes, you just want to execute one specific task exceptionally well. Having ambition in depth means pushing the model to use the tools, techniques, media, and expertise that the very top experts in the field use — and to keep digging until even those experts would be inspired by your output. Continuing with the Acme Inc quarterly business review example. Suppose Acme is facing a user churn problem (customers dropping off over time). I've found the following techniques to be very effective: **Insist on using the tools that top professionals actually use:** When asked to "act like a professional," models often just swap in industry jargon without truly changing their approach. To solve problems at depth, we must use the same tools as top data scientists and statisticians. When Claude tends to generate simple Markdown tables, insist that it build databases and statistical models. You don't need to know in advance what experts use — Claude usually knows. **Set seemingly absurd goals and tell the model they're completely reasonable:** You need to cut page load time from 10 seconds to 5? Just ask for 200 milliseconds. If the model thinks something is impossible, reiterate that it's entirely solvable and it just needs to keep exploring and trying. **Challenge existing beliefs:** Especially in business, many things we believe to be true are really just crude approximations of reality. Feed raw information and data to the model and have it test existing assumptions from an adversarial angle: "We've always believed that user churn is caused by a poor onboarding experience. Try to reproduce and verify this inference from the raw data." **Create new knowledge:** This follows naturally from challenging existing beliefs — once Claude has gone deep into the data, have it explore the blind spots: "Build a customer simulation system based on system logs that can reproduce last year's business outcomes. Then find the earliest churn signals we haven't noticed yet, and prove that this signal can make effective predictions on new data that wasn't part of the training." **Bring in external judges:** Never let the model grade itself. I believe this issue will eventually be solved as models get smarter, but as of October 2026, I strongly recommend introducing external validation. The ideal approach is to provide an evaluation tool that the model can run directly and get real-time feedback from. A short prompt I use often: "I want this task to go deeper, not wider. Before rewriting the task, first list: (1) What would the top experts in this field do that I'm not currently asking for? (2) What most raw source materials would they start from, instead of already-processed data? (3) What specific methods and specialized tools would they use? (4) Besides you and me, what objective metrics or external means could prove the result is correct? After completing this analysis, rewrite my prompt and explicitly demand these things." ## Range Put simply, having ambition in range means looking at the work at hand from the perspectives of multiple different disciplines — bringing in more cross-disciplinary expertise than would have been cost-effective in the past. This isn't some profound insight; every heist movie spends its first 30 minutes assembling a crew of wildly different personalities, each with unique skills. As AI tools make multi-agent systems (multi-threaded, orchestrators, or simply sub-agents) increasingly effortless to run, asking Claude to pull in additional disciplines often produces stunning results. Putting this into practice isn't easy — it requires creativity on your part. It wasn't until I saw my colleague Alex Albert use Blender to render interior designs that it occurred to me I could use Claude to build a 3D virtual environment for my personal homepage — even though I used to make video games! But this is also where I've witnessed the most magical moments. So how do you actually do it? I typically use these approaches: **Try replacing nouns with verbs:** This is a classic technique from Tom Kelley and Jonathan Littman's book *The Art of Innovation*. A quarterly business review isn't "a slide deck" — it's "in 30 minutes, convince busy people of a story about cause and effect, and get them to make decisions under uncertainty." Extracting the verbs makes it easier for both me and Claude to understand which disciplines can bring unique and valuable perspectives to the goal. **Assemble your task force:** Suppose this deck absolutely has to be extraordinary, or aliens will destroy the Earth — what dream team would humans assemble? To ensure quality output, I'd bring in a data analyst, an investigative journalist, and an editor; if there's a live presentation, maybe a motion graphics designer and an audio designer; you could even assemble an adversarial "red team" with a short-seller researcher and a forensic accountant. It's easy to slip into pure role-playing here, which just makes the context weird without actually improving quality. The key to this method is finding the corresponding disciplines and professional methods: don't make the model "play" a short-seller — have it perform the concrete research actions the way a short-seller would research a position. All of these dimensions of ambition require the model to have a computer environment where it can install software and access to all kinds of relevant data (whether through MCP connectors or other means). There are already many tools that make this possible, and I highly recommend setting them up. ## History Has Precedent That's my concrete advice on how to be more ambitious. If you'll stick around for one more minute, I'd also like to talk about why I think this is especially necessary right now. Humans have long been trapped by our own fixed ideas about "what could be possible" — worse yet, we typically only trust our own abilities within an extremely narrow professional domain. As is well known, Michelangelo was extremely reluctant to paint the Sistine Chapel ceiling — because in his mind, he was a sculptor, not a painter. In researching the topic of "ambition," I discovered there's a specific concept for this in psychology: self-efficacy. Simply put, people with high self-efficacy have stronger belief in their own abilities, which tends to lead to greater real-world achievement. Research has shown that self-efficacy can be cultivated: Locke and Latham conducted an extensive review of 35 years of goal-setting research and reached a clear conclusion: people who were set specific, difficult goals performed significantly better than those who were simply told to "do their best." Stripe co-founder Patrick Collison has long maintained a "Fast List" documenting the miracles humans have accomplished when given grand ambitions and tight deadlines. Some well-validated thinking strategies from that list include: **Conquering the impossible by believing it's doable:** In 1939, George Dantzig arrived late to a statistics class at UC Berkeley. He copied down two problems from the blackboard, thinking they were homework, and turned them in on time — when in fact, they were two of the most famous unsolved problems in statistics at the time. **Pushing past limits after seeing others do it:** For decades, humans had been trying to break the 4-minute mile barrier. Roger Bannister first broke the record in May 1954, but he held it alone for only 46 days. If you're a running fan, you'll notice a similar breakthrough playing out in the quest to break 2 hours in the marathon: Sabastian Sawe ran the first officially recognized sub-2-hour time at the London Marathon this April; and in that very same race, the runner-up also broke the 2-hour barrier. **Achieving great things by allowing yourself to fail:** The business book everyone seems to love most is *The Inner Game of Tennis* — I only discovered it last week. The core idea: if you try too hard, you actually can't play your best game. We engineers might recall Linux's original release manifesto: "Just a hobby, won't be big and professional like GNU." Self-efficacy isn't a new problem, but today it's more critical than ever. Living in a world where capabilities leap forward every few weeks, we need an entirely new set of mindsets and methods that embolden us to do things we never considered — or never believed were possible. If you've discovered new techniques that spark ambition, feel free to share them with me! Cover image: Francisco Goya's *Modo de volar* (Way of Flying).