How to Actually Collaborate with LLM

Truly, two people can use the exact same AI tool and get completely different quality of results. The difference is not access. It is skill.

That skill has a name. It is called AI fluency, and it breaks down into these competencies: the 4D framework and prompting techniques.

The 4D Framework

1. Delegation

Knowing what to hand to AI, and what to keep for yourself.

Not every task belongs with AI. Understanding what you want to achieve, the model capabilities and and how to interact. This depends on your field expertise, and sorting the tasks for AI. First drafts (emails, reports, summaries, documentation), repetitive formatting or restructuring, brainstorming and generating options, summarising long documents and explaining a concept in simpler terms can be delegated to the AI.

Keep these for yourself:

• Decisions with complex stakes: strategy, hiring, pricing, anything where being wrong is very costly

• Anything requiring 'judgement' about your specific context, which AI doesn't have

• Final sign-off on anything that represents you or your organisation

A simple test: if you wouldn't hand this task to a smart intern/junior colleague with no context about your company, don't hand it to AI either.

2. Description

The how to communicate what you want.

This is where most people get the least value from AI, not because the AI is limited, but because the request is vague.

Here are the foundational techniques. None of these require technical skill. They require clarity. And you can use this reusable prompt formula RTCF: Role, Task, Context, and Format.

Also specify output constraints

Could be the length, tone, and structure you want, to avoid AI guess. Offer examples of what good looks like. Break complex tasks into steps

"Keep it under 150 words. Use a confident, conversational tone. End with a clear call to action."

The meta-prompting technique

let AI write the prompt. This is the most underused, and arguably most powerful technique. This works because the AI is genuinely good at understanding what makes a prompt effective.

Templates are a starting point. But a template doesn't know your specific situation, your audience, or what "good" looks like for your context. The extra work, like adding your context, your constraints, your examples, is what separates someone who uses AI from someone who collaborates with it. The better you get at Description, the less you'll need Discernment to catch problems.

Other Useful Prompting Techniques

A few more tools worth having ready:

Ask for variations

Request a different format

Check confidence

Reset when it goes off track

3. Discernment

Knowing how to evaluate what AI gives you back, before you use it.

This is the competency we touched on heavily in Part 2, because it connects directly to AI's limitations: hallucination, outdated information, and confidently-wrong answers.

Discernment means treating every AI output the way you'd treat a draft from a capable but unverified source, useful, often very good, but not automatically correct.

A few habits that build this:

• Read with your expertise on, not off. Your domain knowledge is the filter AI doesn't have.

• Verify anything specific, numbers, names, claims, citations, especially if it will be seen by others.

• Notice when something feels "off" even if you can't immediately say why. That instinct is often right.

Discernment isn't about distrust. It's about applying the same standard you would apply to any first draft, from anyone.

4. Diligence

Using AI Responsibly

Diligence is about the impact of how you use AI, on accuracy, on fairness, on the people affected by what you produce.

A few things this looks like in practice:

• Being transparent when AI was used in ways in your work

• Considering who's affected. If an output will influence a decision about a person extra scrutiny is warranted.

• Protecting sensitive information. Be thoughtful about what data, especially confidential or personal information, you share with AI tools.

• Owning the outcome. If you used AI and the output was wrong, the accountability is yours, not the tool's.

Bringing It All Together

Here's the thing about the 4D framework: the tools will keep changing. New models will arrive. Prompting techniques that work today may be replaced by simpler ones tomorrow.

But Delegation, Description, Discernment, and Diligence don't expire. They're not tool-specific, they're thinking skills. Someone who has internalised these four competencies will pick up the next AI tool and be effective with it in days, not months, because the underlying skill was never really about the tool.

Part 4 is about application, practical resources and directions for putting all of this to work in your field. For product managers specifically, I'll touch on what "AI in product" actually means in practice, the kinds of features, decisions, and considerations that come up when AI becomes part of what you're building, not just a tool you use to build it.

See you there.

This is Article 3 of the Collaborating with AI Systems series.

Catch up: [Part 1] · [Part 2]