We Gave It Our Reflexes

Design for Batch / Part two

We gave it our reflexes.

AI agents do think before they act. They think inside the room they are standing in, act, walk into the next room and do it again. Nobody walks the whole house first.

“Operations of thought are like cavalry charges in a battle.”

Alfred North Whitehead, 1911

For people, thought is scarce, so we act on reflex and save thinking for decisive moments. For AI, thought is the cheap part. We copied our shortcut into a machine that never had our problem.

Eight rooms of one lifeOne hallway in the middle

Eight glowing gold rooms arranged in a ring around a central hallway. Lines run from every room to the center, a five-line musical staff glows above the center, and the hands beside each room pulse together.
Mode

Think first. The reasoner walks into every room and takes notes, comes back to the central hallway and writes one score, and only then do the hands play, each from its part of the same score.

The rooms: inbox, calendar, texts, calls, transcripts, clients, money and home. In Reflex, an event lands in a room and that room’s hands react at once. In Think first, the reasoner walks every room, returns to the hallway, writes one score, then every pair of hands plays from it. Illustration, not measurement.

A text arrives while you pour coffee, and your thumb answers it before you have decided anything. By noon you have handled thirty messages, each one sensibly, and thought about none of them together.

Why we act without thinking

In 1890, in the chapter on habit in The Principles of Psychology, William James argued that habit is not a weakness. It is how a mind with limited attention survives a day. The more of daily life we hand to what he called the “effortless custody of automatism,” the more our higher powers are set free for their proper work.

Twenty years later, Alfred North Whitehead put the same idea in two lines in An Introduction to Mathematics. Civilization advances, he wrote, by extending the number of important operations we can perform without thinking about them. And thought itself is like a cavalry charge: limited in number, in need of fresh horses, saved for decisive moments.

That is the first principle. Reflex is not a flaw in people. It is an economy. Our thinking is slow, tiring and able to hold only a few things at once, so we spend it carefully and let habit handle the rest.

Why people act on reflexTrue for AI?
Thinking is tiringNo
Attention holds a few things at onceNo. It can read every room in one pass.
Deliberating everything would stall the dayNo. A half-hour cycle has time to spare.
The price: reflex is local and uncoordinatedYes. This cost carries over.
Three of the four reasons for reflex do not apply to AI. The one that does apply is the cost, and that is the part we copied.

We built agents in our image

When we gave AI hands, we gave it the way we use ours. An event arrives, and the agent acts. Agents today do think before they act, often carefully. But they think inside the room they are standing in. The email agent reasons about the email. The calendar agent reasons about the calendar. Each room acts without anyone walking around first.

That is a reflex, even when it is a thoughtful one. A reflex is not defined by speed. It is defined by scope: acting on what is in front of you instead of on the whole situation.

Before AI had hands, it did the opposite. You gave it a situation, it surveyed all of it, and it handed back a plan. That was the thing it did best, and in the rush toward agency we set it aside. We wanted it to carry out the plan, so we let each room carry out its own part, one event at a time.

We gave it our reflexes without our reason for needing them.

There was a reason it happened, and the reason is gone. For an AI to act without someone prompting it, something had to happen first. An event was the only alarm clock. So we built agents that wake on events, and an agent that wakes on an event naturally works in that event’s room. The scheduled task changed that. A clock can wake the AI now, with no event at all, and a clock does not belong to any room.

The score comes first

So let it think before it acts, the way only it can. On a schedule, it walks into every room: the inbox, the calendar, the texts, the transcripts, the clients, the money. It looks around and takes notes. Then it comes back to the central hallway and writes one plan in which every part answers to every other part. A score.

Only then does it hire the musicians. Each one gets a part written from the score, with room to play it well, and reports back what actually happened. Those reports are the rehearsal notes for the next score. Drums, guitar, bass and vocals recorded in separate sessions make a song only when someone already wrote the arrangement. Most agents today are recording sessions with no arrangement.

Nothing that needs to be fast gets slower. Conversation stays live, so a call or a text is answered in the moment, and a real deadline can wake the composer early. What changes is that responsibility, everything that is work rather than talk, is composed before it is performed.

The delay is the point

A batch deliberately introduces a delay, and the delay is not a cost to engineer away. It is what lets the situation cohere. A client texts at 9:05 to move Friday, and at 9:20 her assistant emails a different time. A supplier says a part is late, and an hour later says which part. Act on each message the moment it lands and you lock in half of the story. Wait for the next walk through the house and the two halves arrive together.

People have always known this. We sleep on decisions, let a draft sit and wait for the dust to settle. We rarely get to, because the world keeps tapping us on the shoulder. A batch gives the AI what we never reliably had: the freedom to walk around before it acts, and enough time between walks for the pieces to become one picture.

In The Keyhole I argued that workflows block what the model can see. This is the other half. Even with nothing blocking the view, an agent built on our reflexes does not use it.

At six, before anyone at the shop is awake, the owner’s AI walks the whole business: overnight orders, a supplier’s delay, two quotes waiting, a crew member’s sick text, Saturday’s weather. By twenty past, it has a score. The delay moves a job, the moved job frees a crew for the quotes, the quotes go out with honest dates, and the customer whose job moved hears first, with a reason. Down the street, a competitor’s agents spent the same night answering each of those things in its own room, and three of the answers contradict each other.

What I expect to happen

  1. The best agents will act less and finish more.

    Fewer actions, each one fitting the others. Activity will stop being the sign of a good assistant. Coherence will be.

  2. The famous failures will be contradictions.

    Not one hand doing something wrong, but two hands doing opposite things in the same hour, each correct on its own: one room books the table while another cancels the evening.

  3. People will read the plan, not the activity log.

    The score becomes the thing you inspect, adjust once and then trust. A list of everything the hands did will matter about as much as a list of every note a violinist played.

  4. Triggers shrink to conversation and emergencies.

    A live voice and a real deadline will still wake the AI at once. Everything else moves onto the clock.

  5. We will stop asking AI to be more like us.

    The goal of the last few years was human-like agency. The next goal is agency no human could have: the whole situation, every time, before any hand moves.

Living it

I learned this the honest way. Tonight I asked an AI that had my calendar, my email, my meeting recordings and my notes to lay out what I should be doing. It thought carefully in each room it entered. It looked at my calendar going forward and never looked back, so it missed a client lunch I had two days earlier, and with it a deal that is very much alive. The door was open. It never walked the house. That is why I build Alli, my executive assistant, on Design for Batch, and why I want my AI to be less like me, not more. I act on reflex because I have to. It does not have to.

The claim ledger

Reflex is a human workaround that AI does not need.

Already true
Habit as an economy of scarce thought is more than a century old, from James in 1890 and Whitehead in 1911, and the cost of machine thinking keeps falling.
What has to happen
A full walk through every room has to stay cheap enough to run many times a day for an ordinary person’s life.
Where I am probably wrong
If thinking at that scale stays expensive, reflex will win on cost for small things, and the full survey will be saved for big ones, much as it is for us.

Fewer coordinated actions beat many separate ones.

Already true
In any team, a plan everyone works from produces fewer collisions than everyone doing their own part well.
What has to happen
The hands have to report what actually happened, so the next score is written from facts and not from intentions.
Where I am probably wrong
If the world changes faster than the cycle, the score is stale before it is played. Then the cycle has to shorten, or the deadline has to wake it.

The clock replaces the trigger.

Already true
Scheduled tasks now let an AI wake with no prompt and no event.
What has to happen
Builders have to choose the clock on purpose, instead of adding a timer to a system still organized around events.
Where I am probably wrong
Conversation will always need a live trigger, and some emergencies will too. That is fine as long as talk stays live and the work stays composed.

Back to the coffee. Your thumb answered the text before you decided anything, and that was fine. You are tired, the morning is long, and your attention is precious. Your AI is not tired.

Let it walk the house first.

Background

Author: John Rector

John Rector is a Charleston-based entrepreneur, author, and AI strategist. He co-founded E2open, the supply-chain software company acquired for $2.1 billion in 2025, and in 2026 opened Charleston AI, a 3,000-square-foot lab that helps people and organizations understand and use artificial intelligence. He is the creator of The Reality Equation — a lecture series, book, and curriculum exploring attention, prediction, and how reality is experienced — and the author of more than two dozen books. He writes and speaks widely on artificial intelligence, attention, and the future of human work.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.