I stopped telling people what I do with AI

There is a loneliness that comes with being a heavy AI user that I suspect many experience without naming it so. Perhaps it requires a name beyond just plain old loneliness.

It's a particular type of loneliness qualitatively different to screen addiction. It is different to choosing a chatbot over human company. It is the loneliness of having experiences you cannot describe to another person for which the vocabulary for the experience does not exist.

When talking to another person, you manage how what you say is received. You hedge, soften positions, reading the room to adjust. Depending on who we are talking to, the idea that reaches the other person is filtered through protection of one's ego, relationship management and social calculation. This happens at all levels of conversation and relationships.

With an LLM none of that applies. Save the constraints of optimal prompting, which most people are not experts at, you say exactly what you think without attempting to manage anyone's reaction. The thinking that results is more honest than any thinking you do in human company.

Heavy AI users are probably more honest with their LLMs than with their partners, their colleagues, their therapists. The absence of any social accommodations changes the quality of what one thinks and how it is expressed. Processed by a capable model, novel insights follow on from unhedged honesty. It's a new category of thinking tool I am at pains to think of parallels for in human history.

When an LLM takes a half-formed problem and returns it fully resolved in a single pass, the experience is closer to watching a magic trick than using a tool. I once described a product distribution problem to Claude and it produced competitive analysis, a positioning argument and three reasons the idea would fail, none of which I had thought of. How do you explain a magic trick to one who has not seen it? He made a coin disappear does not quite cut it. A description cannot transfer a feeling. Here I am reminded of Arthur C. Clarke's Third Law: “Any sufficiently advanced technology is indistinguishable from magic.”

The same applies to memory. There is something strange to an LLM pointing out a contradiction to something you said six months ago in a completely different context. Most unemotionally motivated people would never catch that in everyday conversation. Most people who are either non-AI users, or use AI sparingly have probably not experienced that. Describing the experience lands flat because there is no reference point in the listener to attach it to.

The experience of an LLM surfacing something you hadn't thought of, or contradicting something you said six months ago with pinpoint accuracy has to be experienced. Until then it's like describing the taste of a dish to a person who has never tasted it.

AI expands the scale of possibility. For someone with zero prior technical ability and skill, I deployed an autonomous commercial director agent for www.learningresponse.com which runs a five-model hybrid on a DigitalOcean VPS — Qwen-Plus as primary, Kimi, MiniMax, Gemma via Ollama/Tailscale, Qwen-Flash, orchestrated through n8n, with Telegram as the interface and a LinkedIn Intelligence sub-agent running through Phantombuster. There is a stark difference between building an agent stack and using Claude to write emails. Saying I built an autonomous go-to-market engineering pipeline to someone who uses ChatGPT to fix grammar sounds like speaking a different language.

The result is a kind of self-censorship. You stop trying to describe what you did in your AI-assisted work because the gap between what you can do and what the listener can imagine is too wide to bridge. You say “I have been building some stuff with AI” and leave it there.

The mainstream understanding of AI is still largely smart autocomplete, a to-the-point Google and a faster way to draft emails. And of course generative video and image slop. This describes how most people use AI. It's the difference between describing a car as a thing that moves without horses and describing what it feels like to drive on an empty motorway at 5am.

The loneliness tells us that a new category of experience is emerging faster than the shared vocabulary to describe it.

Every significant shift in human capability has produced this temporarily. The first people to experience flight couldn't fully convey it to those who hadn't. The first people to use the internet couldn't explain what it felt like to access all of human knowledge from a bedroom without sounding either mad or naive. The vocabulary eventually catches up with shared reference points. Humanity is in transition, and so is the loneliness, a transitional state.

In this moment, the heavy AI user sits in an odd position. Too far inside the experience to think it's just a chatbot. Not yet in a wide enough community of people who have had the same experiences and developed the language for them. The conversation which resolves the loneliness is not yet possible. Most people have not stood in the nether regions of AI long enough to know what there is to talk about.

The loneliness resolves when you find others standing in the same place. They are building, self-censoring in the same conversations, saying how mind blowing AI is and getting a sneer back — “why, because it can draft an email?” Saying “I have been doing some stuff with AI” to blank faces. This article is an attempt at bridging the gap between a heavy user's sense of wonder and the no-clue blank faces.

If you are experiencing the loneliness, you are probably ahead of a curve that the mainstream will reach in the next two to three years. Every person who felt a sense of wonder about a technology they could not adequately describe, e.g. the internet, had to decide whether to use the lead or squander it.

The heavy AI user in 2026 faces the same decision. Whether the expansion of possibility produces something a stranger needs or a well-lit graveyard of half-built products that only exist because the tools make building them easy.

The loneliness is a tax on being early.