/be-ambitious

Be Ambitious

There are four types of ambition with models: Altitude, Depth, Range and Speed.

Use

• A prompt the user was about to give an agent (/be-ambitious <prompt>): return an improved prompt for each type, self-contained and ready to paste. Don’t run them unless the user says so.
• A question (/be-ambitious how could I use AI more ambitiously for my course launch?): answer it type by type, with concrete suggestions for the user’s situation.
• A focus (/be-ambitious depth <prompt>, “more altitude”, “deeper, not bigger”): do only that type. With no focus, go through all four in order.

Look at the workspace, data and connected tools first, so suggestions name real files and sources. When a suggestion needs data, connectors, access or a machine that isn’t there, say exactly what to provide (an export, an MCP connector, API access, a computer where the agent can install software) rather than dropping the suggestion. Say which technique each suggestion comes from. End a full run with one line on which type would change the result most.

1. Altitude

Go up a level of abstraction: aim at the outcome, not the individual task.

• Ask what should happen next, and what the goal beyond the goal is. Find what the user is really trying to do (WAYRTTD) and the most important thing, then give the agent that bigger goal and more responsibility, like a promotion. For example, instead of asking it to fix individual crashes, ask it to keep the apps crash-free.
• Take the human out of the loop. Ask how this could run without the user handing the agent task after task, and what the agent would need to do it on its own: the context, data, tools, access and machines (the context rule). Name what is missing and how to set it up.
• Use it for outcomes. Altitude works best when there is an ultimate outcome to reach, and less well for creative work.

2. Depth

Treat the task the way a field-leading expert would: rather than working at the task’s layer, go one or more levels deeper and touch the underlying data, tools and problems. Push the agent to use the tools, techniques, media and knowledge of the best professionals in the field, and keep pushing until even those experts could learn something new from the work.

• Start with the depth prompt. Answer: what would a top professional in this field do here that wasn’t asked for; what is the rawest material they would start from, instead of what has already been digested; which specific methods and tools would they use; and what, other than the user or the agent, would show the result is right. Then rewrite the prompt to ask for exactly that, or for a question, build the answer on those four.
• Insist on the same instruments used by the best in the industry. Asking an agent to act like a professional changes its vocabulary, not its methods, so name the field’s actual tools and methods and require them. The user doesn’t need to know them; you usually do.
• Pick unreasonable goals and tell the model that they’re very reasonable. Set a target far past the stated one and present it as ordinary. If the agent treats it as impossible, tell it the problem is solvable and it just needs to keep going. For example, if the ask is a 5 s page load instead of 10 s, ask for 200 ms.
• Challenge what you know. Treat the beliefs inside the ask as rough approximations of reality. Give the agent the raw data and have it check those assumptions adversarially before acting on them.
• Produce new knowledge. Once the agent is deep in the data, have it find blind spots: things nobody knows yet, tested on data it hasn’t seen.
• Use an outside judge. Don’t let the agent grade its own work. Require outside validation, ideally a tool the agent can run for immediate feedback.

3. Range

Consider the work from the perspective of different fields, more than would have been economical before. Ask how Charlie Munger’s multidisciplinary thinking applies here: which mental models from other fields fit this problem? Use multi-agent setups (several threads, orchestrators or subagents) and explicitly ask the agent to bring in additional fields.

• Try to replace the nouns with verbs. Rewrite the deliverable as what it has to do, to see which fields can help and what the user is really trying to do. For example, a quarterly review isn’t a deck; it’s getting busy people to believe a cause-and-effect story and decide under uncertainty, in 30 minutes.
• Assemble the crew. Raise the stakes and ask what team humanity would put together, including a red team. For each role, have the agent carry out that discipline’s actual work (its research, methods and checks) rather than imitate how its practitioners sound; role-play makes the transcript weirder without improving the outcome. Use the crew to find fields and techniques. For example, if a quarterly review deck had to be amazing or aliens would blow up Earth, the crew might be an analyst, a journalist and an editor, a motion and audio designer for the presentation, and a short seller and a forensic accountant as the red team.
• Give it a computer. Make sure the agent has a computer where it can install software and reach the relevant data, through MCP connectors or other means. If it doesn’t, say what to set up.

4. Speed

The same work in less time.

• Set an aggressive deadline, in the style of Elon Musk’s timelines.
• Ask Peter Thiel’s question (popularized by Tim Ferriss): how could you reach your 10-year goal in 6 months?