The best AI Creative Process partnership with AI is not constant collaboration: use it to widen the field, protect the human middle, then bring it back to strengthen what you have made.

Most people use AI as if it should remain beside them throughout the whole creative process.

They open a chatbot when the task begins. They ask it for ideas, request an outline, generate a first draft, improve the language, challenge the argument and write the final summary. AI is present from the blank page to the finished work.

That feels efficient. It may even produce something competent.

But competence is not the same as creativity.

The more interesting model is a circular process with five phases:

1. Expanding Option Space
2. Generating with Independence
3. Evaluating with Perspective
4. Embodying for Depth
5. Integrating for Impact

The sequence matters more than the individual labels.

AI is used heavily at the beginning of the AI Creative Process Step to expose possibilities. It then steps back while the human generates, judges and experiences the idea. It returns near the end to test, connect and strengthen what the human has created.

That separation is not an inconvenience. It is the mechanism that protects creative ownership.

I think of it as the Five W Creative Cycle: AI Creative Process Step

Widen. Withdraw. Work. Walk. Weave.

It gives AI two clear entrances into the process, with a protected human space between them.

AI Creative Process Step 1. Widen the field

The first phase is about increasing the number of ways you can see the problem.

This is where AI is genuinely useful in the AI Creative Process Step.

A language model can scan across industries, disciplines, historical examples and competing viewpoints faster than any individual can. It can expose assumptions, find comparisons and generate questions you had not considered.

The purpose is not to ask AI for the answer. It is to make the question larger.

Suppose a programme team is trying to improve attendance at a weekly delivery meeting.

The obvious prompt would be:

How can we improve attendance at our weekly delivery meeting?

That will probably return familiar suggestions: shorten the meeting, improve the agenda, clarify attendance expectations and send reminders.

Nothing is necessarily wrong with those ideas. They are simply trapped inside the original framing.

A widening prompt asks something different:

What might poor attendance tell us about the design of the meeting, the information people need and where decisions are actually being made?

The problem is no longer “people are not attending”. It may be that the meeting duplicates another forum, provides no decision rights, arrives too late to influence delivery or exists mainly to produce reassurance for management.

AI can help expose those possibilities.

Research into AI-supported knowledge work shows why this phase can be valuable. In a field experiment involving hundreds of consultants, people using generative AI completed suitable tasks faster and produced higher-quality work. The researchers also found that performance depended heavily on whether the task sat inside AI’s uneven area of competence, which they called the “jagged technological frontier”. (Harvard Business School AI Institute⁠)

That is a useful warning. AI can widen the field, but it does not reliably know where the field ends.

During this phase, I would ask AI to:

• identify hidden assumptions;
• find analogous problems in other sectors;
• describe the issue from different stakeholder positions;
• separate symptoms from possible causes;
• generate questions rather than finished solutions;
• show where evidence is missing or contradictory.

After 30 years across mainframes, client-server systems, cloud platforms, agile delivery and now AI, I have seen the same adoption mistake repeatedly. We take a new tool and point it at the old question.

The better use of technology is often to help us discover that we have been asking the wrong question.

Useful question: What possibilities am I currently unable to see because I have accepted the original framing?

AI Creative Process Step 2. Withdraw the machine

Once the option space has been widened, close the AI tool.

This is the most important move in the cycle, and probably the one people will resist most.

AI makes the blank page easier. It also makes it dangerously easy never to face one.

When suggested sentences, plans and images are always available, we no longer have to tolerate the uncomfortable period in which an idea has not yet formed. We can move directly from a vague intention to a plausible output.

But that uncomfortable period is often where authorship begins.

Withdrawal creates a boundary between what the system has surfaced and what you choose to make of it.

This does not require pretending that AI was never involved. The research, questions and provocations from the first phase remain available. What changes is who now has responsibility for creating the direction.

Write down what surprised you.

Remove the generic material.

Select the tensions that matter.

Then generate your own response without asking the machine to complete it.

In organisational work, this could mean asking AI to explore possible causes of recurring production incidents, then switching it off before the team designs its intervention. The people who understand the service, operational pressures, customer impact and history should decide which problem deserves attention.

The distinction is subtle but important.

AI may help you discover ten possible directions. It should not automatically choose the direction that becomes yours.

Researchers studying human–AI co-creation increasingly describe this as a question of agency: who controls the direction, how responsibility shifts during the interaction and whether the human remains able to understand and shape the emerging work. Recent co-creation research argues that systems designed merely to execute a sequence of prompts do not match the ambiguous, iterative character of real creative work. (MDPI⁠)

I have learned to recognise the warning sign in my own work. It appears when I am editing AI output before I have decided what I think.

The sentences may improve, but my position becomes weaker.

Withdrawal reverses that order. Decide first. Draft second. Re-engage the tool later.

Useful question: What do I think about this before I ask AI to express it for me?

AI Creative Process Step 3. Work the idea yourself

This is the generative core of the AI Creative Process Step.

You take the expanded field and make something that did not exist before: a proposal, diagram, argument, design, operating model, workshop or decision.

The aim is not to prove that humans are superior to machines. It is to preserve the parts of creative work that depend on context, intention and commitment.

AI can produce combinations. It cannot take responsibility for why one combination should matter to your organisation.

Consider the difference between generating a programme recovery plan and creating one.

A model can list familiar recovery actions:

• confirm scope;
• rebaseline the plan;
• address resource gaps;
• tighten governance;
• escalate dependencies;
• improve risk management.

A programme manager has to decide what is really happening.

Perhaps the milestones are unrealistic because nobody challenged the original business case. Perhaps the team is carrying too much work in progress. Perhaps governance is not weak but excessive. Perhaps the delivery problem is actually an unresolved commercial decision.

Those judgements come from accumulated experience, conversations, observed behaviour and knowledge of the environment.

Over four technology shifts, I have found that the hardest part of delivery is rarely producing more options. It is seeing which constraint is real.

That requires independent work.

Start roughly. Use paper, a whiteboard or an empty document. Build the argument in your own sequence. Sketch the process before making it presentable. Explain it aloud without reading from AI-produced text.

The first version may be worse than something a model could produce in seconds. That is acceptable.

A rough draft created by you contains information about your thinking. A polished draft generated for you may conceal the fact that the thinking never happened.

There is also evidence that AI assistance can raise the quality of individual creative outputs while making the collection of outputs more similar. In other words, more people can produce stronger work, but they may begin to converge on comparable ideas. That is a useful trade-off when consistency matters. It is a problem when difference is the point.

The risk is not that AI always produces bad ideas. It is that plausible suggestions arrive before your own associations have had time to form.

Useful question: What part of this work must carry my judgement rather than merely my approval?

AI Creative Process Step 4. Walk it into the real world

An idea can appear convincing on a screen and collapse as soon as it meets a real person, place or constraint.

That is why the fourth phase is embodied.

To embody an idea is to encounter it through more than words. You draw it, say it, simulate it, test it, move through it or place it in front of someone who will experience the consequences.

For a designer, that might mean building a physical prototype.

For a facilitator, it means running the exercise with a real group rather than admiring the workshop plan.

For a manager, it may mean walking through a proposed operating process with the people expected to perform it at 2 a.m. during an incident.

For a writer, it means reading the article aloud and noticing where the language no longer sounds like them.

This phase matters because human understanding is not confined to abstract reasoning. Research on embodied creativity examines how movement, physical context, tools and interaction with the environment can contribute to creative thought. The evidence is still developing, but it challenges the simplistic idea that creativity happens entirely inside the head before being transferred into the world. (Frontiers⁠)

I often understand a programme problem differently when I draw the dependency chain by hand.

The act of placing items on a page forces decisions about sequence, distance and connection. A dependency that seemed minor in a spreadsheet may turn out to sit beneath three critical milestones. Two apparently separate risks may reveal the same root cause.

The body detects problems that polished language can hide.

You can also embody an idea socially.

Explain your proposal to someone who was not involved in creating it. Ask them to show you how they believe it would work. Watch where they hesitate. Notice what they misunderstand without blaming them for misunderstanding it.

AI cannot fully simulate this because it lacks the specific organisational environment in which the idea must survive. It can role-play a sceptical stakeholder, but it does not carry that stakeholder’s history, incentives, fatigue or professional risk.

A presentation may say that a process is simple.

A user attempting to follow it may tell you otherwise within three minutes.

Useful question: What will I only learn when this idea meets a real body, conversation or environment?

AI Creative Process Step 5. Weave it into something useful

AI returns in the final phase of the AI Creative Process Step.

Its job is no longer to originate the work. Its job is to help connect, test and strengthen it.

You now have something with human ownership: a chosen direction, a rough solution and evidence from the real world. AI can help examine that work from several angles without taking control of its purpose.

Ask it to find gaps.

Ask it to identify affected stakeholders.

Ask it to compare the proposal with relevant standards or research.

Ask it to find contradictions between the stated intention and the operating detail.

Ask it to generate scenarios, objections or failure modes.

Ask it to adapt the communication for different audiences while preserving the central argument.

Suppose a team has independently designed a new release-readiness process and walked it through with engineering, operations and business users.

AI could then help:

• compare the process against previous incident themes;
• detect controls that have no clear owner;
• turn workshop notes into a traceability table;
• test whether each identified risk has a corresponding mitigation;
• produce different explanations for executives and delivery teams;
• identify where two stages request the same evidence;
• draft questions for a final assurance review.

This is integration rather than substitution.

The human contribution gives the work direction and meaning. AI helps make the resulting network of evidence, dependencies and communication easier to inspect.

That does not remove the need for verification. AI remains capable of confidently inventing references, overlooking local constraints and smoothing genuine disagreement into harmless prose.

The human remains accountable for what crosses the line into use.

This is why the five-stage model forms a loop rather than a straight production line.

Integration reveals new questions. Testing exposes assumptions. Stakeholder feedback changes the problem. The cycle returns to widening, but it does not return to the same starting point.

You now know more.

The rule is temporal separation

AI Creative Process Step – Five W Creative Cycle works because the activities happen in a deliberate order:

Widen — use AI to expose more of the landscape.
Withdraw — remove AI before it supplies the direction.
Work — generate and shape the central idea yourself.
Walk — test the idea through action, conversation and physical reality.
Weave — bring AI back to challenge, connect and strengthen the result.

The order protects against two opposite mistakes.

The first is rejecting AI from work where its range and speed could be useful.

The second is inviting it so deeply into the process that we can no longer identify what we contributed beyond prompt selection and approval.

Those are not the only options.

In earlier technology shifts, the tools changed but the management question remained familiar: what should the system do, and what must remain a human responsibility?

With AI, that boundary can move several times within one task.

The right boundary at the research stage may be wrong during creation. The right boundary during evaluation may be wrong when responsibility for the final decision is assigned.

So do not ask whether AI belongs in your creative process.

Ask when it belongs.

Open it to widen the field.

Close it while you decide what is yours.

Reopen it when there is something worth challenging.

Sources used

• Dell’Acqua, F. et al., Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality. The research examined how generative AI affected the speed and quality of professional knowledge work and introduced the idea of an uneven capability frontier. (Harvard Business School AI Institute)
• Salma, Z., Hijón-Neira, R. and Pizarro, C., Designing Co-Creative Systems: Five Paradoxes in Human–AI Collaboration. The paper examines tensions including speed versus reflection, control versus serendipity and originality versus remix. (MDPI⁠)
• Malinin, L., How Radical Is Embodied Creativity? Implications of 4E Approaches for Creativity Research and Teaching. A review of approaches that treat creativity as situated in bodies, tools, environments and action rather than only internal thought. (Frontiers⁠)
• Frith, E. et al., Creativity in Motion: Examining the Impact of Meaningful Movement on Creative Cognition. An experimental study examining relationships between movement and creative performance. (Frontiers⁠)
• ACM Creativity & Cognition research on human-centred AI communication and co-creative systems, including agency, interaction dynamics and communication between the human and the system. (ACM Digital Library⁠)
• McKinney, P., Human–AI Creative Partnership: How to Harness AI While Preserving Your Innovative Edge. A practitioner framework arguing for deliberate division between human creative direction and AI-supported exploration and refinement. (philmckinney.com⁠)