FRAMEWORK

Mental model 3.0:
A new way of thinking

Most of us have spent years using the internet in a navigation mindset – which app, website or feature do I need for this?

With AI agents (ChatGPT, Claude, Gemini, Copilot, etc.), that’s changing.

You’re no longer just clicking around interfaces – you’re working with a system that can think with you.

Mental Model 3.0 is a simple way to understand this shift, and breaks it down into three questions:

  • how you think about the work (not just tools),

  • the role you play when an agent is involved,

  • the workflow you use so it doesn’t turn into chaos.

Mental model 3.0 at a glance

A new way of thinking, a new role, and a simple shared system for working with agents.

🧠

Agent mindset
How do I brief my agent?

NEW WAY OF THINKING

✍️

Editor and coordinator
How do I shape the work?

NEW ROLE

🔁

Shared system
How do we work together?

NEW WORKFLOW

Agent mindset 🧠

NEW WAY OF THINKING: How do I brief my agent?

For the past 25 years or so, using the internet has mostly meant having a navigation mindset. We ask “Which app / menu / feature do I need?” – then open a browser or app, search, click around an interface and fill in forms.

With agents (ChatGPT, Claude, Gemini, Copilot etc.) a new agent mindset is emerging. Instead of starting with “Which app?”, we start with:

“How do I brief my agent to help me do this?”

A good brief usually includes four things:

  • Goal – what you’re trying to get done

  • Audience – who it’s for

  • Context – what’s already happened / what you know so far

  • Constraints – format, length, tone, deadlines, must haves / must avoids

We can then ask the agent to help us find and organise information, structure tasks, draft, review and refine – with that shared understanding as a starting point. 

Right now most AI use is still task-based – one-off prompts in fresh chats, with no long-running context. Over time, as agents become more capable, and we get used to working in this way, this new mindset will extend further into how we plan, coordinate and review work.

The core shift is:

From “Which app do I open?”
to “What does my agent need to know to help?”

Editor and coordinator ✍️

NEW ROLE: How do I shape the work?

In an agent mindset, you’re not just “using a tool” – you’re the editor and coordinator of the work.

The agent can generate ideas, drafts and options at speed – but it has no lived context, reputation or responsibility. You do.

Your role is to:

  • Decide what stays human
    The agent-era mindset is not about giving up control. Judgement, values, ethics and identity remain firmly human-owned.

  • Decide what to delegate
    Where can the agent genuinely help – research, structure, drafting, options, checking – and where do you want to keep full control?

  • Shape and refine the output
    Iterate with the agent: review, edit and rewrite so it sounds like you, meets your standards and fits your context.

  • Hold the line on quality and impact
    You decide what is acceptable to send, publish or act on – and when something needs more checking or a different approach.

Over time, this editor / coordinator role will probably become a habit – you’ll start with an agent earlier in a task, give it clearer briefs, and refine how you work together.

The agent mindset is:

“The agent is my thinking partner, not my replacement.”

Shared system 🔁

NEW WORKFLOW: How do we work together?

To work effectively with agents, you need a simple shared system that both you and the agent recognise.

In practice, this means three things:

  • Set some rails
    Write some simple, explicit rules that cover what is out of bounds to the agent (e.g. legal decisions, confidential data, anything that must stay human), and how you want it to behave (e.g. level of detail, how to handle uncertainty).
    👉 We go into more detail in the Rails framework

  • Choose the right level of structure
    Not every task needs the same setup. For quick, simple tasks you can keep it light. For longer-running work with specific outputs, you’ll need more structure – work in steps, be clearer about what “done” looks like, and keep track of decisions.
    👉 Our Four Modes framework gives you a simple language for these different kinds of work.

  • Add some routines to your workflow
    A few simple habits can really help to make the process of working with an agent smoother and calmer. For example, ask your agent to write a summary with next steps at the end of every session, or do a weekly what worked / what didn’t recap.  

The agent mindset is:

“I have a way of working with an agent, not just random prompts”

Cognitive shifts in working with an agent

Working with an agent isn’t just about learning new tools or changing how you work – it also changes how you think about work. You may start to notice some of the following shifts.

  • Externalised thoughts
    You get used to externalising early thoughts, half-formed plans and complex decisions, which can help with clarity and reduce mental clutter.

  • From one-off prompts to a thinking partner
    Instead of starting a new chat and firing off a quick prompt, you have an ongoing conversation. You stop expecting a perfect response first time and instead refine and question together with the agent.

  • From “Can I trust AI?” to “How can I work safely?”
    You think less in terms of blanket trust/distrust, and more about guardrails, structure and simple routines that keep things safe and human-led.

  • Less novelty, more normality
    Over time working with an agent stops feeling like a novel experiment and becomes simply part of how you work – and you wonder how you managed without it.

What this means for organisations

At a team or organisation level, Mental Model 3.0 is less about “rolling out a new tool” and more about how people think about and work with AI – and giving them enough structure to do that safely and confidently.

A few shifts to look at:

  • From deploy tools → design workflows
    Instead of “we’ve given everyone Copilot”, the focus shifts to:
    “What are the 3–5 workflows where human+agent teams can meaningfully change outcomes?”

  • From prompt tips → shared collaboration language
    Rather than circulating lists of “best prompts”, give people simple shared concepts and a shared language – Modes, Rails, routines – so they can describe how they’re working with agents.

  • From “shadow AI” → visible collaboration
    People are encouraged to say how they’re using agents, what they’re delegating and what stays human. This makes it easier to spot healthy patterns, risks and missing Rails.

  • From AI anxiety → practical AI literacy
    It’s natural to feel anxious when things change. Teams and organisations can lower that anxiety by offering coaching and guidance on how to work with AI safely and effectively – how to brief an agent, what to keep human, and where Rails are needed.

How to use this framework

You don’t need a big reset to use this. Just experiment with one real piece of work and see what changes.

1. Pick one project to run with an agent mindset
Choose something where an extra thinking partner would genuinely help – a report, research, workshop, strategy piece etc.

2. Write a short brief using the four ingredients: goal, audience, context, constraints

3. Decide what you’ll delegate to the agent and agree a light shared system: a couple of rails and a routines such as always end by asking for a recap + next steps.

4. Reflect and tweak
After you’ve worked with the agent on that project, think about what went well / what you would change next time - whether the brief, rails, structure or routines. Then carry on that over to the next project - improving your work one step at a time. 

Next steps

Four Modes – understand how you’re working with agents today

Boundaries and safety – set simple guardrails so AI use feels safe and aligned with your work

First messages with a new agent – a template for setting up your first 3.0-style collaboration.