How to Be More Ambitious with AI

Speed, Altitude, Depth and Range. The four ways I push models past "make this email nicer", with the exact moves, prompts and guardrails I use for each.

Dr Sam Mokhtari

The model isn't the ceiling. Your ask is.

Here's the honest version: most people use AI to do the same small jobs slightly faster. Rewrite this email. Summarise this PDF. Fix this bug. That's useful, but it's like owning a jet and only taxiing it around the runway.

Model capability isn't even. It's astonishing in some directions and weak in others. The people getting extraordinary results aren't using a secret model. They ask bigger questions, and they set the work up so the model can actually answer them.

I break ambition into four types: Speed, Altitude, Depth and Range. Each one has its own moves, and each one can be practised. This playbook gives you those moves, the prompts I actually use, and the guardrails from my books so you don't fly into a wall.

Four ways to be more ambitious

Speed is the same work in less time. It's where most people start, and stop.

Altitude goes up. Own the outcome, not the task: the goal behind the goal.

Depth goes down. Work the way the best expert in the field would, from raw material and real instruments.

Range goes wide. Bring in fields the task doesn't obviously need.

Speed: four moves that buy back hours

Speed is the bread and butter of AI's value, and most people still leave half of it on the table. They're quick at prompting and slow at everything around it.

1. Turn repeats into skills

If I explain something to the model twice, the third time it becomes a skill or template: instructions, examples and a definition of done. One line runs it.

2. Batch, don't drip

Ten small asks in ten chats means ten cold starts. I hand over the whole batch with one standard, then review the set together.

3. Run work in parallel

Research, drafting and checking don't need to queue. I give independent pieces to separate agents or threads and merge the results.

4. Keep a living brief

Who I am, the project, the standards, the no-go list. The model reads it first, every time. Context is the slowest thing you'll ever type.

Measure the finished outcome, not the fast prompt

Here's the trap. A quick first draft feels fast. Three rounds of rework later, it wasn't. In The Agentic Operating Model I argue the only honest metric for agents is cost per completed task: every attempt, retry and human review, divided by the outcomes that were actually accepted. Time works the same way. Time to first answer is vanity. Time to an accepted result is the real number.

Fast prompt

• Fire off a one-line ask
• Get a draft in 20 seconds
• Spend 40 minutes fixing tone, facts and format
• Start from scratch again next week

Fast outcome

• Spend 3 minutes on goal, audience and "done"
• Attach two examples of good
• Ask it to check itself against the criteria
• Save the brief as a reusable skill

Altitude: climb from the task to the goal behind it

The model executes your best guess at how to reach an outcome. So give it the outcome. Ask yourself: what do I actually want to happen next? Here's a routine job, the quarterly board review, climbed one rung at a time.

Build the loop that takes you out of it

The real prize at altitude is a durable workflow, one that keeps running without you feeding it task after task. But you can't hand an outcome to something that can't see, act or check its work. And the moment you step out of the loop, governance stops being optional. That's the core argument of The Five Gates.

When to climb, and when to stay low

Altitude isn't always the right move. It shines when there's a real outcome behind the work. It struggles with creative jobs where the "goal" is your own taste.

Signal	What it tells you	My move
You'll hold a meeting about this work	There's a decision behind the deliverable	Climb. Ask for the decisions, not the document.
You keep sending the same kind of task	A standing outcome is hiding in the repeats	Climb. Hand over the outcome and build the loop.
The work is creative and taste-led	The goal lives in your judgement	Stay low. Iterate together and keep the pen.
The model can't reach the data or verify results	It's guessing at the outcome	Fix access first, then climb.
An error would be costly or irreversible	Autonomy multiplies the risk	Climb only behind the five gates.

Depth: do it the way the best in the field would

Sometimes you don't want a bigger task. You want the same task done brilliantly. Depth means pushing the model to use the methods, tools and raw material of a field-leading expert, until even that expert might learn something. Say the board review keeps hitting one problem: customers are leaving.

Insist on expert instruments

Told to "act like an expert", models change vocabulary, not method. When it reaches for a markdown table, insist on a database and a statistical model.

Set unreasonable goals

Need a page to load in 5 seconds instead of 10? Ask for 200 milliseconds. When it says impossible, tell it the problem is solvable and to keep going.

Challenge what you "know"

"We believe churn is an onboarding problem. Try to reproduce that from the raw data." Most business beliefs are rough approximations.

Produce new knowledge

"Simulate last year's customers from our logs. Find the earliest churn signal we don't know about, and show it predicts on unseen data."

Never let the model mark its own homework

The depth move that matters most is an outside judge. Ideally it's a tool the model can run for instant feedback: a test suite, a holdout dataset, a compiler, a benchmark, a real customer.

In Agentwashed I call the hidden cost the Compound Error Tax. At 90% accuracy per step, a seven-step workflow succeeds less than half the time. In The Agentic Operating Model I go further: judge the trajectory, the whole route of plans and tool calls, not just the final answer.

The depth prompt I reuse

When a task feels shallow, I don't ask for more. I ask for deeper. This prompt makes the model tell me what a top professional would do that I'm not asking for, then rewrites my request to demand exactly that.

Before

"Write a competitor analysis."

After the depth prompt

"Pull 12 months of pricing pages, job ads and release notes for five competitors. Map their positioning, then test the map against our last ten lost-deal notes."

Range: swap the nouns for verbs

Range means looking at your work through more fields than it obviously needs. That used to be unaffordable. Now a crew of specialists costs minutes, not salaries. The hard part is seeing which fields matter, and my trick is to replace the noun with a verb. A noun hides the job. A verb shows who could help.

The noun

• A board deck
• A pricing page
• A training course
• An onboarding email

The verb

• Getting busy people to believe a story about cause and effect, and decide under uncertainty, in 30 minutes
• Helping a sceptical buyer feel safe spending money
• Changing what people do on Monday morning
• Getting a new customer to a first win before they forget why they signed up

Assemble the crew, not a costume party

Raise the stakes in your head: this deck must be brilliant or the world ends. Who would you send? For the board review I'd bring an analyst, a journalist and an editor. To present it, a motion designer. As a red team, a short seller and a forensic accountant.

Structure matters. In The Agentic Operating Model I recommend a central orchestrator with specialist workers for most real work, because every contribution stays traceable through one hub.

Belief is the real ceiling

Here's the part no prompt guide covers. We're held back less by tools than by what we believe is possible. Psychologists call it self-efficacy, and it can be trained. In The Skill Curve I show that most people quit just before the turn, mistaking a normal plateau for proof they're not cut out for it.

Believe it's solvable

In 1939 a late student copied two problems off a Berkeley blackboard and solved them as homework. They were famous unsolved problems. Tell the model, and yourself, it's doable.

Watch someone do it

Once the four-minute mile fell in 1954, others soon followed. Collect examples of ambitious AI work. Each one lifts your ceiling.

Give yourself permission to fail

Gripping too hard makes you play worse. Treat ambitious asks as cheap experiments, not bets.

Consistency beats intensity

One stretch ask a day beats a heroic weekend. Ambition is a habit, not a mood.

Your 7-day ambition plan

Don't try all four at once. Run one move a day on real work and note what changed.

Day	Ambition	The move	Done when
1	Speed	Turn your most repeated prompt into a reusable skill	It runs from one line
2	Speed	Write a living context brief and attach it to every session	You stop re-explaining yourself
3	Altitude	Take one deliverable and ask for the decisions behind it	Three decisions, each with for and against
4	Altitude	Design the loop for one recurring task	Owner, kill switch and audit trail named
5	Depth	Run the depth prompt on a task that matters	An outside judge is defined
6	Depth	Ask the model to test one belief against raw data	Confirmed or broken, with evidence
7	Range	Rewrite a noun as a verb and brief a crew of three fields	One idea you'd never reach alone

Stay in the work

YouTube — Subscribe · @sam_mk87

LinkedIn — Connect · where I publish the thinking

X — Follow · @sam_mk87

Instagram — Follow · @sam__mk87