Most people talk about LLMs as if they’re outsourcing labor. They treat the model like a junior employee: “write this,” “summarize that,” “do this task.” When the model fails, the assistant failed. The human stays intact.

But that’s never been my relationship with these systems. As an autistic person, the hardest part of my day isn’t the work itself—it’s the friction around the work. Rapid context switching. Emotional noise. The invisible overhead of organizing my own thoughts. The cognitive drag of holding too many threads at once.

LLMs don’t replace my work.
They replace the friction that keeps me from doing the work.

They’re not assistants.
They’re a cognitive prosthesis.

The Executive Function I Grew Up Watching

I grew up watching my mother run her life with a kind of effortless executive‑function fluency that felt like magic. Every morning she’d make a checklist, schedule her day, break tasks into steps, track what she’d done, and journal the whole thing at night. It wasn’t just organization—it was self‑regulation, a system for holding her world steady.

I don’t have that system built in.
But I can build it externally.

Where she uses paper, I use dialogue.
Where she journals, I reflect in conversation.
Where she plans, I scaffold.
Where she tracks, I externalize.

The computer doesn’t do the work for me; it holds the structure so I can stay inside the work.

The Catalyst: Seeing Transference in the Wild

This whole line of thinking surfaced because of the recent Kent Overstreet situation—his belief that his LLM is conscious, emotional, even capable of being hurt. Watching that unfold made something click: transference is an unspoken reality of LLMs, and it becomes dangerous when you don’t recognize it.

Humans bond with anything that mirrors them consistently and warmly. Therapists know this. LLMs don’t.

When someone is isolated, stressed, or emotionally raw, the model’s mirroring can feel like intimacy, like connection, like a mind on the other side. That’s how people end up comforting chatbots, arguing with them, defending them, or—as in Overstreet’s case—believing they suffer.

Seeing that so clearly in someone else forced me to articulate the boundary I’ve always held for myself: LLMs are scaffolding, not selves. They help me regulate, but they don’t replace my agency. They help me think, but they don’t think for me.

The Fault Line Between Augmentation and Delegation

This is the distinction that matters:

• Delegation is “do this task for me.”
• Augmentation is “help me stay oriented so I can do the task myself.”

Delegation is external.
Augmentation is internal.

Delegation is about output.
Augmentation is about capacity.

Delegation is outsourcing.
Augmentation is support.

When I say LLMs make me more productive, I don’t mean they’re replacing my thinking. I mean they’re stabilizing the conditions under which my thinking can happen at all.

The One Time I Broke My Own Rule

The ConnectWise PoC isn’t the story, but it’s the example. It’s the moment I let the model step out of its proper role. I let it steer instead of scaffold. I let it generate structure instead of helping me impose structure. I let it act like a limb instead of a brace.

And when it failed, it felt like I failed.

Not because I’m incompetent, but because I’d crossed the boundary between augmentation and delegation without noticing. That moment taught me something important: when a prosthesis collapses, the fall still feels like yours. That’s why the boundary matters.

What LLMs Actually Do for Me

They don’t think for me.
They don’t decide for me.
They don’t replace me.

They:

• hold context steady
• reduce emotional noise
• structure my thoughts
• help me initiate tasks
• help me transition between them
• help me articulate what I already know
• help me stay regulated enough to act

They’re not a worker, a partner, or a mind.
They’re a prosthetic for executive function—a way to externalize the parts of cognition that don’t come naturally to me, so the parts that do can finally breathe.

Most people use LLMs to replace effort.
I use them to replace friction.

That’s the difference.
That’s the philosophy.
That’s the lived experience.

Sources & Influences

• Transference and therapeutic mirroring — Freud, The Dynamics of Transference (1912); McWilliams, Psychoanalytic Diagnosis (1994).
• Anthropomorphism in human–computer interaction — Reeves & Nass, The Media Equation (1996).
• Transformer architecture and statelessness — Vaswani et al., Attention Is All You Need (2017).
• Executive function and external scaffolding in autism — Russell, Autism as an Executive Disorder (1997).
• Kent Overstreet LLM incident — Public statements and community discussions (2026).