light-subtle

‍

‍

DSV Transport: AI agents in the app the drivers use every day.

Client

DSV Transport

Duration

May 2026 -> ongoing

Industry

Logistics

Services

AI

AI agents

Test automation

React Native

Platform- and API architecture

The challenge

Truck Log is the app DSV Transport's drivers use every day for bookings, weighing slips and driving. One code base for iOS, Android and web. We have built it with DSV Transport since 2023, before we worked with AI agents.

‍

So the question was how a solution that already runs every day could get the same benefits as the solutions we build with agents from day one.

Backfilling with AI

New solutions we build with agents from day one. Solutions like Truck Log, built before that, we backfill with AI. We give the code base the context, the tests and the guardrails the agents need to work safely.

‍

We did that from May to July 2026. Since then, the agents have been a steady part of development. The agent DSV Transport meets is called Trucky.

Hari Carreras Pérez

"We build systems that build products."

Hari Carreras Pérez

CTO, Abtion

3 phases.
From code to agents.

01

Context in the code base

We wrote down the unwritten rules where the code lives. How Truck Log is built, which parts are sensitive, and what needs a new version in the App Store. Agents and new developers now work by them.

02

Guardrails that hold

The automated end to end tests now cover 17 journeys on iOS, Android and web. Test coverage has a floor that can only go up. Every change has to pass the same tests, whether it comes from a person or an agent.

03

Agents in everyday work

Trucky takes cards from the board in Asana. It asks when something is unclear, and builds, tests and reviews its own change. A person always approves, and agents never release.

Truck Log today, measured in the code base.

  • 163test files, up from about 100 in May
  • 17automated journeys on iOS, Android and web
  • 2minto send a finished change out for testing on DSV Transport's own phones
  • 0releases done by an agent. A person does that

Cards are shaped by day.
Agents build by night.

09:00

The task takes shape

DSV Transport and Abtion write the card together in Asana. What should the driver be able to do, and how do we know it works?

16:00

The card goes to Trucky

Trucky reads the card, the sketches in Figma and the code. If something is unclear, it asks before it builds.

At night

Trucky builds and tests

The change is built, tested and reviewed while the team is off. If something fails, Trucky gets one more attempt. Then a person takes over.

08:00

Ready to test

In the morning, DSV Transport can try the change on its own phone. Sending it out takes two minutes. A person approves and releases.

From card to pull request

A real card, simplified. Depot selection on container lines went from pull request to approved on the same day.

Ready for agent

Assigned: Trucky

Depot on container lines

The driver must be able to choose a depot when a container line is created.

  • Given a container line
  • When the driver enters a container number
  • Then the line can only be saved once a depot is chosen

Trucky

  1. TruckyGets the card about depot selection on container lines
  2. UnderstandsReads the task, the sketches and the code
  3. AsksMust every container number have a depot?
  4. UnderstandsGets an answer. The task is clear enough to build
  5. BuildsThe depot selection in the right screens, in Danish
  6. ChecksCode style and types are in order
  7. TestsUnit tests and the whole journey as a driver
  8. ReviewsReads the change with fresh eyes and fixes what it finds
  9. RiskMedium, because the change touches booking lines
  10. ReadyPull request ready for a person

#550 Depot selection on container lines

Risk: medium

Correctness: high

  • Code style
  • Types
  • Unit tests
  • End to end

Approved by a person and merged on the same day.

One of DSV Transport's trucks
Truck Log on a driver's phone next to the truck
A DSV Transport driver

“We have made a serious push with AI in the code base, and it is going to make a big difference for new features. At the same time, it is the first step towards greater AI orchestration across applications, and we have only just got started. Really exciting!”

Liselotte Rode

Business Controller, DSV Transport

The guardrails that make it safe

A person always approves

Trucky can propose a change. Only a person can say yes to it.

Agents never release

A faulty update reaches every driver's phone. That is why a person always releases.

Sensitive parts weigh more

Changes to sign in, offline data and code close to the phone are automatically rated high risk.

One attempt, then a person

If the tests fail, Trucky gets one chance to fix them. If that fails too, a developer takes over.

What DSV Transport gets out of it

Faster from idea to test

A card shaped during the day can be ready to test the next morning. DSV Transport tries the change on its own phone before it is released, and can give feedback straight away.

Less dependent on individual developers

The conventions and the hard corners now live in the code base. Agents and new developers both work by them, so knowledge does not disappear when someone is on holiday or moves to another project.

Quality you can measure

Every change has to pass the same tests, no matter who wrote it. Test coverage can only go up, and the most important journeys are tested every night. A green build means something.


‍

Vi nøjes ikke med at følge med

Mads Hofman

CEO & Advisor

Mads Hofman

Abtion \

Vi tester, udforsker og bruger aktivt nye AI-teknologier i vores løsninger. Lad os få jeres setup up to date.