Five More Obvious (But Not So Obvious) Truths About Service Design

The first five were about instruments pointed at the wrong object. Friction that sorts backward. A map of a call center that was never a room. Humans doing the integration work no budget can see. A workflow frozen around a constraint that has expired. A dashboard that scores the golden path as a failure. The bonus was the one that stings in an agentic shop: a person will wink, and a model will not.

Five Obvious (But Not So Obvious) Truths About Service Design stopped at the moment you agree to tell the truth. These five start there. You can describe the service honestly and still describe the wrong object, at the wrong resolution, with the friction sanded off, the bill hidden in another department, and the airplane already at altitude. Same test as the first list. Trivially true the moment you hear it. It invalidates a category of work, so you cannot nod and go back to the demo.

They come from three places that have been having the same argument in different dialects. Ben Reason and Lavrans Løvlie started Livework with Chris Downs in 2001, and they sat down with Josh Tyson and Robb Wilson on Invisible Machines to talk about what that practice is for now that anyone can generate software. Seth Godin, on the previous episode, had just insisted that AI will do what you said. And the second edition of Age of Invisible Machines has been describing the artifact those two warnings require: a plan the system can actually fly.

1. You are automating the product and calling it the service

The obvious part is that a banking app is not a relationship with a bank. People love a BMW. Almost nobody loves their bank, their hospital, or their mortgage. Reason's old shorthand for the field was to bring the quality you can feel in the car to the relationship you cannot point at.

The part that costs you something is the clock. A product is a promise: we will lend you this, you will pay it back like that. The service is how you find the offer, how you sign, how you survive the years you are in it, and how you leave. Banks like to call the mortgage the product. The service is the way in, the way through, and the way out. Break that life into journeys and each one has a need and an outcome you can count. The trouble is that the customer's "done" and the institution's "done" are often different noon. At one bank, the customer felt the account was open when the card arrived. The bank felt it was open when the money landed. Nothing was resolved, because the two definitions had never been forced into the same sentence.

Automate that journey and you industrialize the mismatch. You get a very fast path to a card that is not an account. Reason has watched a European telco do the organizational version. Most people walking into the store were already customers who needed the tariff changed. The sales director would not take the figure, on the order of 60 percent of footfall, because his compensation and his whole makeup were product and sales. The service was the majority of the visit. The company was instrumented to see a SKU.

Age of Invisible Machines keeps returning to the same instruction, which guests on this show keep trying to compress into one line: automate what you should be doing. The product you already ship is what you are doing. Godin, the episode before, put the question in front of it. What is the work for, and who is it for? You can answer honestly and still aim the agents at the thing you can point at. The category this retires is the AI roadmap that starts from the product catalog.

2. The right fidelity for both is the wrong fidelity for either

The obvious part is that a person and a machine do not read the same document. The part that costs you something is that service design spent twenty-five years learning this the hard way, and then AI made the lesson urgent in the other direction.

Løvlie and Reason made service blueprints that tried to hold everything: purpose, customers, staff, systems, the whole behavior on one map. The overview helped. Past a certain resolution, organizations could not use them. Do this corner now, that corner in two years, that corner in four, and the instruction died in the corridor. What survives is a compression a human can sign. What an agent needs is the other document, the hydrated one, full of the detail it would be irresponsible to leave implicit. Robb Wilson's line for the blend is the one to keep:

The right fidelity for both is the wrong fidelity for either.

One artifact fails both audiences. Finance cannot approve a novel. An agent should not improvise from a poster. Reason has been cutting the same map different ways for years, by hand. A blueprint for JPMorgan went nowhere until the IT costs were drawn on it. A director who needed to see what the technology organization was actually delivering finally had a picture. Someone else needs the regulatory view. The studio joke was "make a version for Jonathan," and then a team lost two days recutting. Models make those cuts cheap. They do not make the underlying agreement optional.

This is the architecture Age of Invisible Machines already described from the cockpit of a 787. The old airplanes made pilots carry manuals that barely fit in the flight deck, step-by-step sequences for every phase. The useful move was not a faster manual. It was a system that held the flight plan and showed the controls for the phase you were actually in. Canonical knowledge stays whole. The person sees the slice. A service blueprint in the agentic era is that split: the hydrated plan is what the agents run, and the compression is what a human can refuse.

The category this retires is the single master blueprint, the one source of truth that is somehow also the build spec and the executive summary. Godin had just warned that a demo hydrates with one dataset and looks perfect. A blended blueprint is that demo, printed.

3. Seamless is how you design vanilla

The first list included a truth from Jennifer Pahlka: friction you add can select against the people you built the service for. That one stands. This is the mirror. Some friction is the service.

An insurer Løvlie worked with made claims almost too clean. Suitcase gone while traveling: the rule says pay. They nearly deleted the box that asked what happened. Customers stopped trusting the payout. They needed the company to have heard the story. A team on UK student-loan applications caught itself about to make the form fast, and stopped. The applicants were seventeen and eighteen, borrowing thousands of pounds for the first time. The team built speed bumps: how interest works, how repayment works, a small check before you can continue. Fast would have been a decision the person had not actually had.

Reason has been saying a version of this since the start of the field. People do not want seamless. They want the friction that helps them feel something, finish something, or know what they just agreed to. Wilson's illustration, late in the same conversation, is that Reason was recording from a boat he had been living on for weeks. If good experience meant easy, you would not put your life in storage and go to sea. An agent pointed at easiest will strip the difficulty that made the thing worth trusting.

Løvlie calls the failure mode vanilla. Everybody likes vanilla. Seamless, similar, impossible to choose. He thinks that sameness can be a mercy in a public service, where a citizen should not have to learn a new personality at every counter. It is a crisis for a business whose advantage is being chosen. Check the box with a model and it will reference every other model. No quirk, no reason to pick this provider, nothing with soul. People bought vinyl back for a reason.

Godin, on his return visit for The Knot, was drawing the same edge.

You can't think outside the box. It's very dark out there. But at the edge of the box is where you can make a miracle happen.

The purveyors of AI, he said, would like you to believe there are no constraints. The people who use it well erect them. His test is a service test: if we adopt this constraint, is it going to help our customers get to where they want to go? The book has the matching instrument. A graphical interface cannot express a need it does not support. There is no click for the thing that is not there. A conversational service logs the request anyway, which is why the text field was never just a courtesy. Delete it and you delete the demand sensor along with the trust.

The category this retires is "remove friction" as a brief. The first list said some friction sorts against your intent. This one says sanding all of it off can delete the service you meant to run.

4. The hours you save are a bill you haven't opened

The obvious part is that a change in one office lands in another. The part that costs you something is that the attractive business case is often just the slice of the system your team is allowed to count.

Løvlie is running the proof in Norway, where the rules on alcohol look straightforward until an apple farmer wants to make cider, serve it to guests, and host a wedding. Three stacks of permits. The applicant and the permit office are both lost. His team is building a digital twin of that regulatory system: the rules in one place, and agents for the people inside it, the farmer, the bar, the clerk. Then they change the rules inside the twin. In one pass, a council's good idea was to delete the application. Thousands of hours back. Businesses delighted. The simulation agreed about the hours, then sent the bill somewhere else. More money for the police, because the streets get busier. More money for healthcare, because alcohol produces accidents and illness. A person can believe the first effect. The second and third are where the system was hiding.

This is a different wound from the one Jeff McMillan named on an earlier episode, and from the dashboard problem in the first list. "Hours saved times loaded rate" fails because you never ask where the freed capacity went. A local KPI can be true and still be a way of invoicing another department. The pin, in Godin's sense, is the permit queue. The entanglement is wanting fewer forms and less harm, held at the same time. Extra effort on the application will not pull it. You have to see both wants on one map.

Age of Invisible Machines calls a cousin of this a missing path: the need users are expressing that the system does not handle. A service simulation is that sensor pointed at consequences instead of clicks. The human planners are bad at it on purpose. The interactions do not fit in working memory. A systemic model is built for it, which is why the blueprint gets more useful as it gets thicker, and why you run the twin before you pour tokens into the live service. Not perfectly. Much better than discovering the hospital bill after you have celebrated the hours.

The category this retires is the business case that ends at your own budget line.

5. You are already at thirty thousand feet

The obvious part is that you should know where you are going before you take off. The part that costs you something is that, for twenty years, you did not have to. The backlog was the plan.

Once a leadership team agreed the experience had to be better, the limiter was production. Legacy IT. Get to the back of the queue. Service design did the thinking, then a mountain of software work swallowed the decade. The bad experiences were often there because teams could not do the thing they already knew was right. AI breaks that constraint. Anyone can produce the software. The blank canvas is the new problem, and a UI at least had edges. Reason heard himself say it in the conversation: when the constraint comes off, the excuse comes off with it. "This is a cost center" stops being a strategy.

Wilson's picture for what replaces the excuse is a full airplane at cruising altitude, with someone in the cockpit finally asking where to go, and every passenger naming a different city. Then the airline commissions a report on why the model failed. He calls the skipping of the plan vanity token burning. The person at the keyboard is having a dopaminergic afternoon. The company is paying for a service nobody agreed to. Løvlie ties the waste to a premise that was in the field from the beginning, not bolted on for the AI era. When they were defining service design, Reason was studying sustainable business. A service that depends on burning tokens has to last economically, energetically, and as a relationship. Sustainability stopped being a mood. It became a line item you can know before you build: what are we spending to get this outcome, and what is the outcome worth?

The book already had the cockpit for this. On the 787, the flight plan was not a kickoff metaphor. It was the data the interface was allowed to obey. You do not load the airplane and negotiate the destination in the air. Godin, the week before, explained why so many programs do it anyway. No one is in charge of AI, and the early movers sold tools.

Here's all these cool screwdrivers and wrenches and saws, figure out what to do with it.

Features without a destination produce trauma. And the scoreboard people reach for is the one he had just called a false proxy: you should not hire a programmer by words per minute, and you should not manage a service by how many tokens someone is burning. Counting the burn lets a boss rank a team. It does not say whether the flight had a destination. That argument is the one in token maxing. Service design is the document the ranking cannot replace. Aim, then ready, then fire.

The category this retires is the postmortem that says the AI is broken.

BONUS: A speech will not cascade. The blueprint will.

The obvious part is that the CEO already sees the whole service. Løvlie says the top of the house usually gets it, because they own every interface, and then they organize everyone below them into product, IT, marketing, and retail. Those silos are the hard room. Wilson will say the part that does not fit on a quarterly slide. An inspirational talk will not cascade. A blueprint will, for the same reason a checklist does: have you answered this, and what did you say?

Reason has watched directors keep one on a single sheet in a briefcase, a way to see an organization that had become too fragmented to narrate. A friend at Airbnb got a call from Brian Chesky's office. They had built digital on digital until the CEO could no longer say how the service worked. She made the blueprint that hung outside his office, so a product, a policy, and an experience could be seen in one glance. His moral, which he arrived at during the conversation: dear CEO, you need to be the service designer. The teams that learn this stop hiring someone to do it for them and start asking to be taught.

And you start the drawing with the front line. Projects arrive as "improve the customer experience" and end as staff work. Reason designs healthcare mostly for nurses, because they are the experience the patient actually has. Happy employees are how you get happy customers. A system is experienced by everyone it touches. If the agents are supposed to help a person help another person, someone has to specify that human moment, including the job the person has to change into.

Godin had just said no one is in charge. The blueprint is how someone gets in charge without another keynote. The first list ended on the wink: the specification is no longer a slide, because the model will execute the granite. This list ends on who holds the pen.

What these five have in common

Read against the first five, they are the same argument one step later. Those said the service you designed and the service you operate are different objects, and most of your instruments point at the first one. These say the missing instrument is specific. It is a blueprint in two fidelities. A human can sign the compression. An agent can run the hydration. The right friction is still in it. The bill that lands in another department has been simulated. Somebody, usually the CEO, starting from the people who actually deliver the service, has agreed where the airplane is going.

None of that requires a new model. It requires being willing to lose the demo, the single diagram, the seamless brief, the hours-saved slide, and the speech. That is the entire cost of an obvious but not so obvious truth, and the reason the next five were sitting there the whole time.

The same split, from the years Josh Tyson and his wife ran a Denver salon, is the members essay They Aren’t Paying for the Haircut.

A companion to Five Obvious (But Not So Obvious) Truths About Service Design. Ben Reason and Lavrans Løvlie joined Josh Tyson and Robb Wilson on Invisible Machines (S7E19). Seth Godin’s return visit is S7E18, for The Knot: Problems Can Be Solved. Book references are to the second edition of Age of Invisible Machines (Wiley, 2025), including the Dreamliner account in You Can Automate a 787.