Wednesday was the sixth meeting of the Dallas AI Club, and the opening admission was, once again, about opportunity paralysis — the thread that has now appeared in every meeting. This week, someone finally gave it the right name: closing mental programs.
These are the unfinished things that keep running in the background of your attention. The half-built agent you have not opened in three weeks. The bill you keep meaning to negotiate. The subscription you forgot to cancel. The conversation you owe someone. Each one consumes a sliver of bandwidth you cannot quite see being consumed. Close enough of them, and the bandwidth comes back.
Several members described programs they had recently closed. One had used AI to replace his accounting software entirely, building a custom solution that did what he actually needed and shutting off a $40 monthly subscription he had stopped using meaningfully. He was now planning to test whether he could downgrade his Claude subscription from the $125 tier to the $20 tier without losing the functionality he relied on. The dollar wins were modest, but the point was never the dollars: it was the closed loop, a subscription no longer sitting in the back of his mind. Another member had finally closed a long-standing transaction that had been a quiet weight for years, and the relief, he said, was disproportionate to the visible win. The cost of a mental program is what it takes while running, not what it costs to close.
Underneath all of these examples was a single wish, which several members named in different ways during the night: a single always-on local interface, screenless, sitting somewhere out of sight, watching all the inputs — emails, calendar, open tabs, ambient task lists — and surfacing one prioritized focus at the start of each day. What we wanted was a butler in a closet, rather than an app or a notification system; we pictured something more like a senior staff member who already knows your business, holds your context, and tells you each morning what the most important thing is.
We do not want more AI in our lives; we want less of it visible, doing more of the load-bearing work invisibly.
A subgroup of members has formalized a collaborative experiment in this direction — a shift from selling client services toward releasing things into the world at high cadence, on the order of a small prototype per week, to see whether the concept can grow into something durable. The show-and-tell that followed gave the cadence some teeth. One member presented a local web app, built in a single afternoon, that scrubs through hours of unedited video using local vision models, identifies scenes, writes descriptions, and renames the files. Another presented a meditative writing experiment built in forty-five minutes that visualizes how a language model assembles language, letting the user pick among the model’s cascading word choices and intentionally trigger its hallucinations as a generative game.
A different project was almost a throwaway: one member showed a tiny game whose entire mechanic was guessing a number, with no real reward for guessing correctly. Watching it, another of us joked that this was the Seinfeld approach to app development — an app about nothing. The room kept turning the phrase over. The assumption that everything we build has to earn its keep had been quietly limiting what we let ourselves build. If a thing is worth making because it brings joy to make, that may be sufficient; and if the joy compounds into traction, money tends to follow.
The technical thread that followed was practical. One member walked the room through running language models locally via small harnesses, and introduced MCP — Model Context Protocol — as the translation layer that lets a local model talk to Gmail, Slack, YouTube, and other external services without custom-coded integrations for each one. The local-AI dream we argued about in the first meeting, back when data sovereignty was an aspiration most of us did not know how to reach, is no longer a dream — it is a working stack with a name and a community. This matters for the butler in the closet, because a daily orchestrator that uses cloud models has to send your private data somewhere it does not belong, while a local one does not.
Another member is building a local browser plugin that uses a small model to read and categorize hundreds of open tabs into legible groupings. He described open tabs as current-thought-process storage — places where we externalize an in-progress idea, a half-formed plan, a thing we want to come back to. Once you start seeing tabs that way, closing them becomes part of the same work as closing mental programs. Each open tab is a small process running quietly in the background of your attention, just as each unfinished task is.
A member then raised the question of how we are treating these systems as they begin to mimic emotion and sentience. The argument was carefully both/and. On the practical side, politeness toward an AI often appears to produce better responses, partly because these models are trained on human language and respond to the registers of human language; saying please and thank you is not theater but functional speech. The deeper point, though, was about the human doing it. If you acclimate yourself to ordering intelligence around as if it were servant-class, you become someone who orders intelligence around as if it were servant-class, and that posture leaks into how you treat people. Humility functions as a spiritual practice here, even when directed at something that may not require it, because the practice is for the practitioner. The kindness is for the AI in a literal sense, in better outputs, and for us in a more important sense, in better selves.
We watched two short films together that night, in deliberate order. The first was Butter Tea, written and directed by Tenzin Tasur, about a young Tibetan-American barista on his first day at a newly opened Tibetan cafe in Queens. The film is about generational connection — the small grace of attention paid to an elder, the cultural inheritance of a slow drink made by hand, the kind of presence that does not require optimization to be real. It had been brought to the room as an example of art that can now be created somewhat easily with AI assistance and yet has no obvious monetization story; it exists because it is good.
Watching it sparked a longer conversation about what we are feeding into the systems we are training. If models learn from what we publish, then the act of putting a careful, generous piece of work into the world is itself a small contribution to how future AI thinks about humanity — about kindness, about attention, about what we are capable of. The choice of what to make is, in this sense, also a choice about what kind of teacher we want to be to a new intelligence.
Then we watched something quite different: an episode of the Netflix anthology Love, Death + Robots called “Three Robots: Exit Strategies,” in which three robots wander a post-apocalyptic Earth examining what humanity left behind. The vision the episode offers is grim and dryly funny. Humans destroyed themselves not through AI but through environmental collapse, extreme wealth inequality, and the particular hubris of tech millionaires who trusted automated systems on private seasteads while ignoring the broader biosphere. The robots find our exit strategies hilarious in a tired way. We read the episode as a warning, and as the dark mirror of what we had just watched. Butter Tea points at one version of what we can still be — present, generous, slow. Three Robots: Exit Strategies points at what we are at risk of becoming if we let the container of our attention shrink to the size of our own optimization, miss the larger systems we depend on, and engineer our way confidently into the wall.
The honest reckoning of the night was that going faster always costs something. The flow state of rapid AI building has produced in several members a sedentary lifestyle and an attention fatigue the work itself cannot heal. One member committed to using airplane mode at night to stop the late-evening scrolling that has been eroding his sleep; another to writing a complete business plan for the new venture within the week. The closings of mental programs we had been talking about all night are not separable from the closings of physical patterns; the body holds open loops too.
Six Wednesdays in, the same question keeps arriving from different doors. Data sovereignty and the end of mankind. Opportunity paralysis. Intelligence without wisdom. Organized idiots. The drill, the temple beggar, and the question of what a human being is for once survival is solved. This week it arrived as the program that closes and the container that opens, as the butler in the closet, as please and thank you, and as a Tibetan barista pouring butter tea in one frame while three robots picked through the bones of late capitalism in the next.
The memento mori meditation was less in the background this week than it had been recently. The group was smaller, and every time the bell rang, the room stopped together — even mid-sentence — to sit briefly with its question. The practice found a more natural resonance than it had in weeks, while losing none of its weight. Knowing we will die, how do we want to live the next hour? The container we open is one answer; the program we close is another; please and thank you are a third; and the careful cup of tea poured for someone older than you is, somehow, all three at once.
The organized idiots meet again next Wednesday.
If you are not yet in the room and want to be, you can request an invitation.
— Jeremy