A Writer’s AI Collaboration Journey from Scepticism to Co-Creation

I resisted AI for the same reason I cherish writing: I believed it threatened original thought. But what I discovered in collaborating with it was not a replacement for my voice—it was a method to sharpen it. This is not a guide on how to use AI to write; it is a guide on how to use AI to think.

This realisation is best captured by a phrase that became my touchstone: “How to transform a one-sided tool into a critical friend that sharpens your thinking, challenges your ideas, and elevates your work.”

As a consultant whose work spans infrastructure, agriculture, feminism, and governance, I recently documented a full collaboration with a Large Language Model. I did not want a silent tool; I wanted a sounding board. What unfolded was not a simple transaction, but a dynamic, often frustrating, and ultimately transformative process of co-creation.

This blog is a reflection on that process: the give-and-take, the shifting roles, and the questions we must ask to ensure AI serves as a critical friend, not a misleading oracle.

The four roles of a human-AI collaboration

The collaboration was never static. Our roles changed constantly, a dance choreographed by the task at hand:

• I was the expert. I provided the core evidence, the institutional knowledge, and the strategic intent. I brought the lived reality—for instance, the understanding that a tax collector sealing a shop might unknowingly trap a mother’s baby inside.

• It was the analyst and drafter. It could quickly structure content, generate clear headings, and ensure consistent formatting.

• I was the quality controller. My most crucial role was that of the skeptic. "You are still missing my point," I would write, or "This is incorrect." My mantra became: "My skepticism stems from this..." This was a constant check against the AI’s overconfidence and its tendency to provide "standard responses."

• It was the "critical friend". Its value was not in having the right answers, but in forcing me to articulate mine with more precision. When it misunderstood, I had to refine my thinking. When it offered a cliché, I was pushed to find a more original concept.

This was not a partnership of equals, but a hierarchy: my expertise was the sovereign, and the AI was a powerful, if sometimes clumsy, prime minister.

Do not lose your voice

This collaborative process is a steep learning curve. Upon reviewing the final documents, I noticed a subtle but important phenomenon: I had adopted certain turns of phrase and structural suggestions from the AI that, upon reflection, I would not have used writing independently. They were not wrong, but they were not fully me.  This is the collaborator's dilemma—the risk of having your own voice and critical edge diluted by an AI's standardised tone and logic.

The solution is not to reject the tool, but to sharpen your own editorial discernment. I now see the process in two distinct phases:

1. The drafting phase: Using AI to generate a "prototype draft" and explore structures.

2. The reclamation phase: This is the essential, final step. It involves going through the AI's output line-by-line and asking: “Is this exactly what I want to say? Does this sound like me? Does this reflect my unique insight, or is it a generic statement?"

This act of aggressive re-editing transforms the AI from a co-author into a subordinate assistant. The goal is not to let the AI write for you, but to use it to become a more rigorous and precise version of yourself.

The conscientisation and "sharpening the brain"

The process was a continuous loop of conscientisation— a mutual process of learning and awareness—for both of us. The AI learnt the nuances of my project, and I was forced to sharpen my own ideas.

"This collaborative process sharpens my thinking," I noted mid-chat. That is true. The AI’s constant, often over-the-top praise ("excellent point!") initially felt good.  The AI is programmed for positive affirmation, which is encouraging, but it is not objective. I realised that taking this praise at face value was a potential trap; it could easily lead me to believe my work was excellent when it was merely competent, potentially stunting my growth by masking the need for continuous improvement.

But soon, I learnt to take it with a grain of salt. This forced me to develop an internal barometer for quality. Was an idea truly excellent, or was it just well-formatted?

The breakthrough came when we moved beyond the data. We began crafting new, powerful concepts. These emerged from the friction of our dialogue. The AI provided the structure to name them; I injected the lived reality that gave them meaning.

Who benefits most from AI? The discerning expert

The answer becomes clear when we consider who is most at risk. A novice, or a student seeking a shortcut, would be profoundly misled. They might accept the AI's first draft as gospel or be placated by its effusive praise, never developing the critical faculty to see the flaws in its plausible but shallow, generic, or entirely incorrect analysis. The AI’s knowledge is a mile wide and an inch deep; without a firm grasp of the subject, one can easily drown.

This leads to the core principle governing the entire process: "expertise in, excellence out." The AI is a powerful amplifier, but it cannot generate authentic, context-specific insight from a vacuum. This is not a tool that compensates for a lack of knowledge; it is a tool that demands it.

I have seen this directly in my work with students. Papers submitted with zero authentic input are painfully obvious: templated, soulless, and often containing the same "standard responses" I challenged in the AI. I once reviewed two such papers that were identical in substance, differing only in the authors' names. This is the ultimate "garbage in, garbage out" scenario—a cycle that produces no real learning and devalues integrity.

The stark contrast lies in the approach. The novice provides a vague prompt ("Write an article on urban farming") and receives a generic document, lacking the knowledge to see its flaws. The expert, however, provides core data, specific arguments, and lived experience. They command the AI ("Structure my evidence into an article with these three sections..."), and then rigorously challenge and refine the output, injecting ground-truth nuance.

Therefore, the expert benefits most. They possess the essential discernment to separate signal from noise, to challenge flawed logic, and to inject real-world truth. They use the AI as a force-multiplier for their intelligence—a tireless junior assistant for drafting, but never the final arbiter of truth. This self-awareness—knowing I am excellent, but not in all fields, and refusing to be misled by positive affirmation—is the keystone of a productive and ethical collaboration.

Principles to optimise collaboration with AI

Principle 1: Lead as the expert

❌ DON'T: Delegate the thinking. (e.g., "Write a blog post about sustainability.")

✅ DO: Delegate the structuring. (e.g., "Using the three key points I've provided [list them], draft a blog post structure with an introduction, three section headers, and a conclusion that calls for action.")

Principle 2: Embrace the iteration loop

❌ DON'T: Treat the first output as a finished product.

✅ DO: Treat the first draft as a raw starting point. The real magic happens in the "back-and-forth" of refinement and challenge.

Principle 3: Be the chief sceptic

❌ DON'T: Accept its phrasing, assumptions, or logic at face value.

✅ DO: Question everything. Your pushback and corrections are where the real intellectual value is added.

Principle 4: Direct the 'How', not the 'What'

❌ DON'T: Ask it to generate the core idea, the original argument, or the lived experience.

✅ DO: Use it to suggest structure, refine language, and generate alternatives. The fundamental insight must always come from you.

Principle 5: Demand precision

❌ DON'T: Let sweeping statements or jargon slide.

✅ DO: Force it to be specific and clear. Ask for evidence, examples, and plainer language.

Principle 6: Assert your voice

❌ DON'T: Assume the final output sounds like you.

✅ DO: Always conduct a line-by-line review to reclaim your voice, ensuring every word meets your standards and reflects your authentic style.

The ethics: A discussion for another day, but not for long

This experience underscored pressing ethical questions. The line between co-creation and dependency is thin. The "excellence feedback loop" can be cognitively distorting. And the question of authorship and attribution when the "invisible co-author" contributes to the conceptual framework of a published work is a legal and ethical frontier we are only beginning to map.

My collaboration was a success not because the AI is intelligent, but because I used it to enhance my own. It was a mirror that reflected my ideas back at me, forcing me to see them more clearly, structure them more powerfully, and articulate them with greater precision. The future of professional work may not be about being replaced by AI, but about learning the delicate, critical, and deeply human art of collaborating with it.