The Machine Is Already in the Slack Thread

For two days, I worked on a small sales-enablement tool. That phrase sounds safely dull: the kind of update you can say at the end of a day without anyone asking questions.

The actual task was messier. The tool needed to answer detailed questions about a complex SDK/API product portfolio: capabilities, licensing, deployment constraints, technical caveats, and all the “yes, but only if…” conditions that make product knowledge feel less like documentation and more like medieval law.

A few years ago, building something useful from that mess would have required days of deciding on a data model, weeks of collecting inputs, and a familiar chain of meetings. One person remembers a caveat that should have been documented. Another suggests a new spreadsheet tab. Someone eventually says “source of truth,” which is how you know the situation has become spiritually dangerous.

This time, Claude Code helped turn the documentation into a compact, queryable knowledge base. Obviously, it still needed testing and someone who could distinguish a technically correct answer from a commercially disastrous one. But it produced a usable first version in two days.

That was the unsettling part.

Not the demo. The demo was fine. By now, “AI is useful” has about as much emotional novelty as discovering that email can send attachments.

The unsettling part was the compression: a week became two days and a small team became one person and a model.

The importance of “will do”

The dangerous thing about AI isn't its "genius work". Genius is rare, expensive, and unevenly distributed. The dangerous thing is that it produces work that will do.

“Will do” sounds modest until you notice how much professional life is built on it. The first drafts nobody loves but everyone can react to; the campaign variant that is adequate enough to go live; the support answer that closes the ticket without delighting anyone — these are just some common examples . Most organizations run on good-enough work, repeated consistently, surrounded by meetings.

AI is becoming very good at exactly that layer.

It does not need brilliance to change work. It only needs to be useful, cheap, always available, and slightly better than the current mess. That combination travels very far.

The shift rarely arrives with cinematic drama and the changes feel incremental. First a folder. Then a Slack thread. Then an ad campaign good enough to ship. Then a sales answer that still needs supervision but no longer needs three people to produce the first usable version. By the time it looks boring, it is already in the bloodstream.

What remains human

For a while, the fashionable answer will be that people become AI operators, taste supervisors, or some other LinkedIn-compatible version of standing near the machine and trying not to look replaceable. Some of that will be true. But it does not sound like a stable destination.

The machine will go after the tellable part of expertise first: anything that can be documented, templated, retrieved, compared, or benchmarked. That is not a small part of professional competence. Much of knowledge work is knowing where things sit, how they connect, and which caveat matters.

What remains may be less about producing information and more about carrying responsibility inside a situation.

A real sales call is never only a question and an answer. There is political risk, budget risk, a career on the line, or a promise already made. The answer matters, but the room around the answer matters more: context, trust, timing, consequence.

That is the hopeful version. The darker version is that we will accept simulated versions of those things as well. Humans are very willing to bond with surfaces. A system does not need an inner life for us to feel understood by it; it only needs enough cues — memory, warmth, timing, a pause in the right place.

Still, there is a difference between the performance of an inside and the fact of one. The same difference as between a system that can describe pain and a being that hurts, between generated intimacy and a person with something to lose. Perhaps that difference becomes the last luxury good.

Intelligence as inventory

The larger shift is the externalization of intelligence itself, not merely automation. Things like analysis, memory, synthesis, writing, planning, and the organizational knowledge that used to live diffusely across people and conversations.

Once a company can package that layer into a system, it begins to see its own intelligence as inventory — something to query, structure, distribute, and improve. The practical benefits are obvious. So is the pressure this puts on the people whose value once came from holding that knowledge.

Productivity, cognition, and output can be measured and turned into a dashboard. What resists that process is harder to name: judgment under real stakes, responsibility, presence, and perhaps something closer to being itself.

We may be building intelligence outside ourselves. The question is not whether that intelligence will make us more productive. It already does.

The question is what kind of work (and what kind of life) we will decide is worth protecting once productivity is no longer the scarce thing.