Stop Telling AI What to Do

"Stop giving AI tasks. Give it outcomes." — Jaime López Feo, CEO, The Agile Monkeys

You wouldn't hire a brilliant strategist and stand over their shoulder telling them what to do. You'd tell them what you need to achieve — and trust them to figure out how.

But that's exactly how most people work with AI. Describe the task, wait for the output, describe the next task. Somewhere in that loop, they miss the entire point.

At The Agile Monkeys, we call this the Spectrum of Agency — three levels of how people work with AI, and most individuals and organisations are stuck on the first one.

The Spectrum of Agency

1. Task orientation. You describe a task. You get output. You feel efficient — and you are, a little. But you're still in the driver's seat for every decision, every prompt, every next step. AI is your very fast assistant.

2. Workflow thinking. You stop asking AI to do individual tasks and start connecting them. If this happens, do that. Things start running on their own. It feels like magic — and it really is a step forward.

3. Outcome-driven. You stop describing the process entirely. You describe what winning looks like: I have a budget of $5,000 and 100 users. Help me reach them. You're not telling AI how to do the work. You're telling it what success looks like — and trusting it to get you there.

The shift sounds simpler than it is. You're not thinking about managing steps — you're thinking about defining success.

AI Has Execution. You Have Judgment.

"AI is really good at execution, really good at gathering knowledge — but it lacks judgment. That's what we bring." — James Tate, SVP of Engineering

James spoke at our recent AI Salon about the gap between what AI can do and how most teams are actually using it. Outcome-driven thinking closes that gap. When you define the outcome, you're supplying the judgment AI doesn't have: the taste, the context, the understanding of what actually matters.

His team has already made that shift structurally — AI agents handling execution, humans owning the decisions that require something no model can replicate.

"We don't see AI as a tool anymore. We see it as a team member."

That's a different way of organising work entirely.

What Outcome-Driven AI Actually Looks Like

The mission, not a task list.

Instead of assigning work step by step, we gave an agent a goal: find state-of-the-art solutions in the literature, combine the best approaches, build evaluation datasets, test them in a sandbox, and iterate until something works.

In one month: 30% improvement in baseline accuracy. Work that could have taken years.

The key wasn't the prompt. The question was: how do we structure this so it can actually succeed on its own? Not: what should I tell it to do next? But: what does it need in order to figure that out itself?

When perfect becomes the enemy of shipped.

Pieter van Noordennen, Principal Product Lead for AI at USAFacts, arrived at a similar place through a different door. His team had paralysed themselves with an accuracy standard they didn't actually need. So he reframed the question:

"How many problems do you really have that require 100% accuracy? We don't make medical devices. We don't run nuclear power plants."

By defining the actual outcome — trustworthy public information at scale — his team went from skeptics to advocates. Their pipelines now run at 99.7% accuracy. The outcome pulled them forward. Perfection was the thing slowing them down.

This Is Bigger Than Productivity

Brian Haley, co-founder and CEO of Massive Blue, uses AI to disrupt human trafficking networks, counter disinformation campaigns, and support national security operations. For him, the shift from reactive to proactive AI is not an optimisation question — it's a moral one.

"Technology is morally agnostic. It's about the people who champion it and build with it to make a positive impact."

When the mission is defined at that level of seriousness, how to use AI stops being abstract very quickly. The intelligence is in the intention, not just the execution.

What Do You Want?

Outcome-driven AI is not a feature or a workflow. It's a fundamentally different relationship with the most powerful thinking technology ever built.

Most people are still at Level 1. Some are making it to Level 2. The ones who figure out Level 3 first aren't just more productive — they're operating in a different category.

The shift begins with one question: what do you want?

Not: what should I tell it to do next.

"Stop giving AI tasks. Give it outcomes." — Jaime López Feo, The Agile Monkeys

---

The Agile Monkeys is an AI engineering and strategy studio working with enterprise companies on the shift from AI exploration to AI execution.