"Inimica est multorum conversatio: nemo non aliquod nobis vitium aut commendat aut imprimit aut nescientibus adlinit."
— Seneca, Epistulae Morales, VII — The company of many is the enemy: no one fails to either recommend some vice to us, or to impress it, or to stain us with it without our knowing.
The AI debate is everywhere, and it is mostly not useful. Every week brings new conferences, new essays, new op-eds about what AI will or will not do to labor markets, to creativity, to society. Thought leaders think out loud. Ethicists raise objections. Futurists offer timelines. Everyone has an opinion. The crowd is very loud.
Seneca's seventh letter is about what happens when you spend too much time with the crowd. Not that the crowd is always wrong — sometimes it is not. But exposure to the crowd's opinions is corrupting in a specific way: it alters your judgment without your noticing. You absorb their anxieties, their frameworks, their bad analogies. You walk in thinking one thing and walk out thinking something slightly different and slightly worse.
I stopped attending most AI conversations around mid-2024. Not because I knew better, but because I noticed that participating in them was affecting my actual work. I would come away more uncertain, not better informed. The debate was generating heat without light, and I was letting it warm me when I should have been cold and focused.
"Recede in te ipse quantum potes... cum his versare qui te meliorem acturi sunt."
— Seneca, Epistulae Morales, VII — Withdraw into yourself as much as possible... keep company only with those who will improve you.
What I was actually doing during those years of debate: running three companies with AI agents as the operational core. Not as a political statement about jobs. Not as an experiment. As a practical choice driven by what I could afford and what actually worked.
Trillion Initiative. Fly Raising. Agent School. Each one has a pipeline of agents that handles work I would otherwise have to hire for. Drafting. Research. Outreach sequences. Administrative processing. The work gets done. The overhead stays low. I am one person. This is Tuesday.
The crowd's debate runs parallel to this reality without ever quite touching it. On one side: AI will replace all jobs, the apocalypse is coming. On the other: AI is just autocomplete, nothing is really changing. Both sides are looking at the technology abstractly. Neither is asking the only question that matters to a founder: what can I specify precisely enough that an agent produces the output I need?
That question has a practical answer. The debate does not.
Seneca's observation about the crowd is that its influence is mostly invisible. You do not feel yourself being corrupted. You feel yourself being educated, updated, included. The corruption comes in the form of absorbed assumptions that were never examined — the crowd's models of the world entering yours without passing through any scrutiny.
For the AI founder, the crowd's corrupting assumptions are things like: AI is not ready for production. You need a team to deploy AI seriously. The legal risks are too high. The hallucination problem makes agents unreliable. These are not completely false — they are partially true and routinely overstated by people who are not building with agents and do not know their actual failure modes.
The answer is not to declare these concerns invalid. It is to go build something, observe it failing in specific ways, fix those specific ways, and acquire your own model of what AI can and cannot do — not the crowd's model, not the pundit's model, not the ethicist's model.
Your model. Built from what you have actually seen.
Seneca would have found the AI discourse familiar. He lived in a city full of competing schools of philosophy, each with vocal adherents, each convinced of their own correctness. His advice was not to engage until you had developed enough of your own foundation that the crowd's noise would not dislodge it.
Build first. Form opinions second. Let the crowd have its debate.