An AI agent for every area of the workshop: why the AI that works is not one thing that does everything

That morning is not fixed with more effort or one more piece of software. It is fixed when each of those decisions arrives already prepared, with its reasoning beside it, made by someone who understands *that particular thing*.

An AI that does everything is good at nothing in particular

For a couple of years the promise was always the same: a general assistant you can ask anything. Fine for writing an email. But a workshop does not run on general answers, it runs on trade judgement: whoever orders the queue looks at bay occupancy and delivery commitments; whoever builds a quote looks at the catalogue, labour times and warranty; whoever decides a purchase looks first at what is already in the stockroom.

The software industry has reached the same conclusion on its own. Gartner expects 40% of enterprise applications to include task-specific AI agents during 2026, up from less than 5% in 2025. This is not a naming fashion: an agent with a clear scope is faster, cheaper and, above all, can be held to account for what it does.

“A workshop is not one task. It is the schedule, the front desk, the stockroom and the month-end close, and none of them looks like the last.”

What changes when every area has its own agent

The difference shows up where everything shows up: on the screen you already have open. In Cafler AI there is no separate AI panel to go to. Every module carries the agent for its own area, and it does its job right there, with your workshop’s data and in your language.

The ones that step in on your screen

These five do not wait to be asked: when you open the module, the suggestion is already there.

• In the schedule. Which vehicle to start with and why, in one line per car: committed delivery, duration, parts on hand and workshop capacity. The ordering is calculated by the platform.
• In your customers. Which customers are worth going back to and for what specific reason — an inspection coming up, maintenance due, tyres near the limit — ordered by priority, and it also shows you who it ruled out and why. Only those who have given their consent make the list: no grey areas there.
• In quotes. You take what you wrote by hand or in a message and it turns into line items matched against your real catalogue. The amounts come from your own rates, and whatever it could not match comes back flagged for you to look at.
• In purchasing. It proposes which parts to order and from which supplier, checking your own stockroom first, with the reasoning next to each recommendation. Comparing stops meaning six open tabs.
• In work orders. When you close the checklist, it gathers what was found on the vehicle and drafts the customer update: how their car is and what it needs. Ready to send, it goes out when you confirm it, and it is logged on the order.

The ones that wait until you ask

The rest live in the platform chat, and here is the comfortable part: it is one conversation for everything. No picking an agent, no hunting through a menu; you write what you need and the right specialty answers.

• Scheduling and appointments — “what do I have on Thursday afternoon?”, “move the 10 o’clock to Monday”.
• Customers and vehicles — the full history of a number plate without opening four screens.
• Quotes — dictate one while the car is right in front of you.
• Purchasing and stock — what you have, what is missing and what moved.
• Workshop setup — hours, services, bays, employees, warehouses, pricing rules and your branding, step by step and without a manual.
• Finance and month-end — margin, revenue trend, top customers and where the lever for improvement is.

Everything it answers is looked up in your data at that moment. What it cannot look up, it does not assert.

The real difference: it knows cars, not just conversation

Here is what genuinely separates these agents from an ordinary chatbot. A generic chatbot can hold a conversation and read a list of FAQs; the moment the conversation moves into the trade, it stops at the door. Our agents have an automotive knowledge library behind them, written for this and reviewed by workshop people, built on the standards of the sector: the ETRTO load and speed index tables, the European labelling regulation, roadworthiness-test rules.

We started where a workshop plays most of its day-to-day, the tyre, and there the level of detail is that of someone who has spent years at the counter:

• It decodes any sidewall marking. The customer photographs their tyre and asks what it means: 205/55 R16 91V, XL, RunFlat, the manufacturer homologation (MO, AO, ★), the DOT date. It reads all of it and explains it.
• It gives the exact figure, not an approximation. What speed a code allows, how many kilos a load index carries.
• It knows the rules that are not up for debate. Minimum legal tread, what cannot be mixed between axles, why new tyres go on the rear, when a puncture can be repaired and when it cannot, what the roadworthiness test looks at.
• It reads symptoms. “My tyres are worn on the inside”, “there is a vibration at 120”, “it makes a noise when I brake” — and turns that into what is worth checking.
• It advises, it does not push. Summer, winter or all-season depending on use; what changes on an EV; which range makes sense for each need and why, with the European label on the table.

And there is one boundary it never crosses, which is exactly what makes it trustworthy: what it knows about the sector it explains; what belongs to that car or that workshop — which size it takes, what is in stock, what it costs, which slot is free — it only says if a query to the system returns it. It never makes it up. And it never closes a fault as a diagnosis: it talks about a possible cause and suggests bringing the car in.

And one that serves your customers, not your team

The last one does not work inwards. It handles the conversations coming in through the channel your customers already use every day — WhatsApp in most markets, LINE in Taiwan — and moves them forward while the workshop is at full stretch: whoever asks about opening hours gets the opening hours, whoever wants an appointment books it, whoever asks about their car finds out where it is. And whoever asks something technical gets a technical answer, because the same library sits behind it.

It always introduces itself as what it is — an agent, not a person — and the workshop decides how much it is allowed to do: book and move appointments, answer prices for what is in the catalogue, explain a quote, flag maintenance that is due. Each one is switched on separately, and each one can stay at “it proposes, you confirm”.

And this is no small matter: in the Cox Automotive service study of 2025, run with almost 2,000 drivers, 45% were dissatisfied with their experience, and the two main reasons were unexpected costs and poor communication. Not the repair: the communication.

When a person is needed, a person steps in

No AI knows everything, and this is precisely where many tools leave the customer hanging. Here the handover is planned, and the workshop chooses when it happens. A person takes over the conversation when:

• The customer asks to speak to someone. First time of asking, no pushing back.
• The agent does not know the answer. It escalates rather than invent.
• There is anger or a complaint. It reads the tone and hands over without arguing.
• The fault touches safety. Brakes, steering, airbags, smoke or an odd noise are not settled over messages.
• The amount goes beyond what was agreed. Above the limit you set, it prepares the quote and waits for your approval.

Whoever steps in does so with the whole conversation in front of them, so the customer repeats nothing. And once it is resolved, control goes back to the agent and the conversation carries on as normal. You decide who picks up escalations and where we notify you; if nobody picks one up in time, the manager is alerted and the customer is told when they will have an answer.

Why you can trust what they propose

An agent that decides on its own and cannot explain itself is no use in a workshop. So they all work the same way, and it is worth knowing how:

• Two layers. The numbers — prices, priorities, quantities — are calculated by the platform, not the model. The agent puts the reason on top, in a line you can understand.
• Anchored in your data. What a query to your workshop does not return is not asserted.
• Every workshop, its own world. Scope comes from the session: an agent only sees the data of the workshop it is being asked from.
• Limits you set. Maximum discount, the amount it can close on its own, subjects it will not touch, hours when nobody gets messaged.
• You have the wheel. You choose which modules it steps into and whether each one only proposes. Anything going out to your customer, you confirm.

The following Monday

Back to 8:05. The order of the day is proposed with its reasoning beside it. Thursday’s quote has a concrete reason to pick it up again. The order has been compared against your own stockroom. The update on the car delivered yesterday went out when you confirmed it. And of the three people who wrote on Sunday night, two have had their answer since Sunday night — one was asking what the marking on their tyre meant — and the third is waiting for you, because the agent saw that this one needed someone from the workshop.

None of that was done by an AI that knows everything. It was done by several agents that each know their own trade, with the knowledge of the sector behind them, in the place where you already work. You are still the one deciding. You are just deciding on things that arrive ready.