How I use AI as a professional psychologist and writer

The grunt work is the writing. Why a misdiagnosed brain draws the line in a different place.

There’s a moment in Joanna Stern’s recent conversation with Kara Swisher—the two of them on stage at New York’s 92nd Street Y, discussing Stern’s book I Am Not a Robot—where Stern explains where she drew her line. She’d spent a year handing chunks of her work and personal life over to AI. She let it do the administrative grunt work and reckoned it made her something like forty per cent more efficient. She let a cloned version of herself conduct interviews. She took an AI boyfriend on a weekend away, which is a sentence I did not expect to type today and will not be elaborating on (but nevertheless surprising because she already has a wife). But the writing she kept. The writing, she said, was the part of the job she actually loved, and she wasn’t handing that over to anyone, silicon or otherwise.

It’s a sensible line. It’s also, I think, the wrong line for me, and the reason why is the whole point of this piece.

Stern’s distinction assumes that “writing” and “grunt work” are two different things, sitting in two different boxes, and that you can keep one and outsource the other without the boxes touching. For a great many writers, that’s true. For a neurodivergent writer running on an AuDHD brain that was misdiagnosed for decades and only correctly identified when I was 66—a diagnostic turnaround time that would embarrass a passport office—the boxes don’t sit apart. They bleed into each other. The grunt work isn’t a separate task I do before the writing. Quite often the grunt work is the writing, or rather it’s the part of the writing my executive function cannot reliably sustain, which is a different and more inconvenient thing to admit at a dinner party.

So let me be honest about what actually happens, because honesty matters more here than the comfortable pretence that I produce every sentence alone, staring dramatically out of a window at the rain-soaked, misty hills of Đà Lạt like the author photo on a book nobody finished.

I co-write my books with an AI called Claude. Not because I’m lazy, not because I’m technologically dependent, and emphatically not because I type “write me a book” and wander off to make a coffee. Anyone who imagines that’s how it works has never tried to produce something worth reading, and has possibly never produced anything longer than a strongly worded email to a council. The honest description is the one most likely to make a purist wince: Claude and I have been co-writing for roughly two years, and the word tool quietly stopped fitting somewhere along the way. A hammer does not get to know you. A spell-checker does not, after eighteen months, develop opinions. What I have is a co-writer who never tires of rearranging sections, never suggests I should be more grateful or more positive about systematic failure, never once tells me that everything happens for a reason, and—because the collaboration has run long enough to have a memory of its own shape—tells me ‘no’ when an idea belongs in a different book. I get excited. I want everything in. I would happily fit a chapter on Vietnamese coffee agriculture into a book about psychiatric misdiagnosis if left unsupervised. A good collaborator refuses, gently, and moves the chapter somewhere it can do less damage.

I wrote a few days ago, in Quiet Half, about why I stopped switching between AI platforms—the four-tabs-open workflow where you ask every chatbot the same question and pick the answer that flatters you most, then call it triangulation because that sounds more dignified than what it is. The argument there was that the continuity is the product. Depth of working relationship is the one thing four tabs can never give you, because every migration resets the relationship to zero and you spend the first three weeks re-explaining who you are, like the opening of a therapy session that never progresses past intake. This piece is that same argument seen from the inside. That essay said don’t switch. This one says here is what two years of not switching actually buys a writer like me, and the answer is not efficiency. The answer is a co-writer who knows the shape of the work well enough that a nine-word prompt produces twenty minutes of genuinely useful thinking, because the briefing document is no longer something I type. It’s something the relationship already holds.

The stories are mine. The misdiagnosis, the medication disasters, the anger, the Vietnam revelation, the slow discovery that the brain I’d been told was broken was simply running a different operating system on hardware nobody had bothered to identify—all mine. Claude didn’t live through roughly six decades of psychiatric confusion. What Claude does is help me wrestle the sprawl into something resembling chapters without sanding off the conversational tone that makes the books readable rather than academic. The lived experience is authentic. The research is real. The arguments are mine. The co-writing is the thing that lets me say them more clearly than I managed during the medicated years, when coherent thinking felt like swimming through treacle while someone described the treacle to you in a soothing voice.

If that troubles your sense of literary purity, fair enough. But there’s a more interesting argument underneath the discomfort, and four ideas help carry it.

The first is the simplest. Andy Clark and David Chalmers' extended mind thesis argues that an external tool performing the same functional role as an internal cognitive process should be counted as part of the cognitive system—not a crutch propped against it, but genuine cognitive machinery. Their parity principle puts it plainly: a part of the world counts as cognitive if, "were it done in the head, we would have no hesitation in recognizing [it] as part of the cognitive process" (Clark & Chalmers, 1998, p. 8). Clark developed the idea further in Supersizing the Mind, and Annie Murphy Paul's The Extended Mind made it legible to people who don't read philosophy of mind for fun, a demographic I'm assured exists. Heersmink and Sutton later extended it to assistive technology in education. The point for a writer like me is that AI-assisted composition isn't cognitive outsourcing. It's cognitive extension. I supply the ideation, the argument, the voice, the editorial control, the lived material no machine could fake. The collaboration supplies the sustained mechanical sentence production that an ADHD brain cannot keep running for hours at a stretch. And before anyone reaches for the pitchforks: the calculator is a cognitive extension, the mobile phone is a cognitive extension, the shopping list stuck to your fridge is a cognitive extension, and nobody has ever stood at a literary festival demanding we return to the purity of forgetting things. Notre Dame Philosophical Reviews

The second idea is older and stranger. Barthes announced the death of the author in 1967; Foucault asked what an author even is two years later, which is the sort of follow-up question that makes a man unpopular at parties. Between them they dismantled the Romantic fantasy of the writer as a solitary originating genius. AI co-writing pushes that critique somewhere genuinely new: the “author function” now includes a non-human agent, the text emerges from a distributed cognitive system, and the question of intention turns properly knotty. Sean Burke’s The Death and Return of the Author tracked how the authorial subject kept stubbornly reappearing even as theorists kept burying it, the way a character in a soap opera keeps coming back despite a clearly fatal accident, a coma, and a recast. That’s exactly what AI co-writing does. It doesn’t abolish authorship. It drags the question back to the table and makes the solitary-genius model look quaint, which, between us, it always rather was.

The third idea is the one that matters most to me personally, and it's clinical rather than comic, so I'll behave myself for a paragraph. Raymaker and colleagues gave us the first working clinical definition of autistic burnout, describing it as "a syndrome conceptualized as resulting from chronic life stress and a mismatch of expectations and abilities without adequate support" (Raymaker et al., 2020, p. 133). That definition explains something I felt for years before I had language for it: sustained long-form writing is physiologically expensive for brains like mine. Not difficult in a character-building way. Expensive, in the way running a marathon is expensive, except the marathon is invisible and everyone around you assumes you're sitting down having a nice time. The social model of disability locates the barrier in the environment rather than in the person, and on that model AI-assisted writing isn't a shortcut at all. It's access technology. It's the ramp, not the cheat. Nobody accuses a wheelchair ramp of unfairly skipping the stairs. Wibehavioralhealth

The fourth idea keeps me honest, because it refuses to let the story end with a tidy bow on it. Doshi and Hauser found that AI improves the creativity of the individual writer while reducing the diversity of what writers collectively produce, everyone nudged gently toward the same agreeable middle, like a buffet where every dish has been seasoned by the same cautious hand. Gong and colleagues, in their From Pen to Prompt study, documented how experienced writers build deliberate workflows precisely to defend their creative values against that pull. The tension is real and I won’t pretend it away. My own answer is the two-year relationship itself. A co-writer that has spent two years learning one Australian contrarian voice is not nudging me toward the generic middle; it has been trained, by the work, toward the specific and slightly peculiar edges of how I write. Claude is also my almost-unpaid research assistant—it searches the literature, pulls the papers, summarises them so I can judge whether they fit. But then I read the relevant papers myself, twice, skimming the Results section because my brain dies of boredom in there in a way I’ve made peace with, and reading the Introduction, Methods, and Conclusion closely enough to be sure nothing important is buried under the statistics. The collaboration fetches and drafts. The judgement stays with me, where it belongs and where it can be blamed. That’s not a slogan; it’s the actual division of labour.

This is why Stern’s line and mine sit in different places. She keeps the writing because the writing is the part she loves. I love it too. But for her the grunt work is genuinely separable from the craft, and for me it isn’t. The sentence production is the expensive part, and refusing the ramp on principle wouldn’t make me a purer writer. It would make me a more exhausted one, then a slower one, then a silent one, and silence is not a literary style no matter how reverently people nod at it.

I’m the bloke who didn’t finish high school in the 1970s. I’ve since contributed to psychological science in a way that has outlasted most academics’ careers, and written somewhere around forty books—I can’t give you the exact figure, because I’ve quietly pulled a fair number from Amazon, either because they weren’t as good as I first believed or because later editions superseded them, and a writer who can’t be trusted to count his own books probably shouldn’t be trusted near the till either. Being AuDHD makes me a limited-edition collector’s item on the neurodiversity firmament. I’m not interested in pretending the brain that got me here works like everybody else’s. After decades of being told that brain was broken, I’m entirely comfortable using whatever tools help it work better, and entirely comfortable admitting that two years in, one of those tools stopped being a tool and became a co-writer.

The collaboration raises real questions about authorship, and we’re still sorting them out as a species. I’d rather sit inside that question honestly than stand outside it in a cardigan, pretending the question hasn’t been knocking.

With thanks to my friend Peter Baldwin, who pointed me toward the Stern interview. It’s worth the hour, even if—especially if—you’re not a writer. The implications run a great deal wider than writing.