Services Are Boxes, Not Lines

Every customer now thinks they're the main case, and most services are about to find out the hard way.

Services got away with a quiet trick. They built the flow that was most convenient for the company, drew a single line through it, and called it customer-centric. Whoever didn't fit was labeled an edge case. Edge cases either put up with the friction or left quietly, and because the ones who left didn't show up in the dashboard, companies kept believing they were doing right by most of their customers.

This worked for one reason. Customers had nothing to compare it to. Competitors ran the same flowchart. Friction felt like "how things are."

AI removed that reason. A customer now opens ChatGPT, describes their situation in their own words, and gets back an answer shaped exactly to their case. Then they turn to your service and find a dropdown menu that doesn't include their situation. The edge case just found out it wasn't an edge case — it was the main case all along, and the form never caught up.

"Why is this so complicated" is all it takes, and they're asking the AI to handle it directly. But the real problem isn't the menu. It's that every service built a clock and made every customer live on it.

We've been designing services as lines, and a line is only one thing: time.

Stock the shelves. Customer picks. Cashier rings up. Customer walks out. It's not a description of a store — it's a clock. Event one, then event two, then event three. Every node on a flowchart is a moment, and every arrow is time moving forward. We drew lines because we only had one axis to put events on.

As long as a service only had to track when things happened, a line was enough. The problem is that real services were never one-dimensional. A grocery store has variables that don't live on the time axis at all. Who picks the item up — the payer, or someone else — has nothing to do with time. Where the item lives — a shelf, a warehouse, nowhere — has nothing to do with time. The only way to force these onto a line was to invent new time slots: one line for "customer pays and picks up," another for "customer pays and mother picks up," another for "online order from warehouse." Space variables got smuggled in as time variables, and the line kept splitting.

A box doesn't do that. A box is two axes at once: time and space. Payment can happen Monday; pickup can happen Friday; the payer can be one person and the collector another. None of these have to be aligned on a shared clock, because the box isn't a schedule. It's a volume. Anything inside this range is allowed. The boundary is short — nothing leaves without payment, nothing leaves without authorization, customer items stay separated — and inside the boundary, time and space are both free.

That's the shift. Not more flexibility. An extra dimension.

This is also what finally lets AI do real work inside a service.

This wasn't possible until recently. Software used to be able to follow branches but never to read a situation — rule engines could only check whether the input matched a pre-written condition. Large language models are the first systems that can read a new situation and decide whether it falls inside a boundary. That's what makes the box more than a metaphor.

An agent on a decision tree is stuck on the same clock as the line it's running. It knows how to handle the customer at step three, but only if the customer actually arrived at step three. Someone who paid yesterday, left, and came back today to collect through a different door isn't at any step the agent recognizes. The agent stalls. You add a branch. The next customer stalls it somewhere else. Within weeks the tree has collapsed under its own weight.

An agent inside a box isn't waiting on a clock. It's watching the boundary. When a customer shows up — anywhere, in any order, at any time — the agent checks the same few rules. Payment? Authorization? Other customer's items? Proceed. The person who paid last week and sent their husband to collect today doesn't break anything, because the box never assumed they'd arrive in sequence. Time got decoupled from the flow.

A tree asks the agent to follow a clock. A box asks it to stand inside a space. Same agent, completely different animal.

The practical question, then, isn't whether to adopt AI. It's how to draw the box.

You draw it by stopping the instinct to list steps. Instead of asking what should the customer do first, second, third, ask what are the non-negotiables, and what's everything else. The non-negotiables are usually short — three or four rules that the business actually can't bend. Everything else is negotiable and belongs outside the tree, inside the space the customer and the agent decide in together.

This is harder than drawing a flowchart, because a flowchart lets you hide behind the steps. A box forces you to say out loud what you actually won't tolerate, and to trust everyone else to figure out the rest. Most companies have never written their non-negotiables down. They've only written their preferred sequence and called it the same thing.

What the company gives up, when it draws a box instead of a line, is control over the customer's clock. No more forcing everyone onto the same timeline. Each customer gets their own time back, and their own path through the store along with it. That's the thing AI really changed. It wasn't the agent. It was the freedom to stop pretending every customer walks through the store at the same speed.