Meet Leeloo.
Scroll to explore
An ongoing experiment

Exploring what AI can become
in the hands of people
who work with law.

Human judgement.Machine intelligence.

A little curiosity. A lot to discover.

Beyond
the first answer.

A platform for experimenting with legal AI.
Building, testing and asking better questions.

Question.
Build.
Try again.

New possibilities. Real legal work.
Human judgement at the centre.

Choose a question to follow

A convincing answer is only the beginning.

We experiment with the steps between a legal question and a conclusion: finding sources, interpreting them, applying the facts and testing the argument. The interesting part is making those steps visible — including the gaps.

How does the rule meet the facts?

Legal method connects a question, the relevant sources, their interpretation and the facts of a particular case. Subsumption is the step where those facts are tested against a rule’s conditions. A fluent summary can skip the difficult part entirely.

We experiment with workflows that separate finding a rule from applying it. Each conclusion should have a source, a factual premise and an explanation of the connection. Missing information should remain visible, rather than quietly becoming an assumption.

In practice

A contract allows termination on notice. An amendment changes the condition, and the file does not establish when notice was received. Copying the original clause accurately is useful extraction. Deciding what follows requires the amendment and the missing fact too.

What we’re exploring

Can the workflow identify the missing premise before it produces a neat conclusion?

Go deeper: Legal method
  1. Start with the legal question

    Define the jurisdiction, relevant date, decision to be made and facts already established. A request to summarise a contract is a different task from assessing whether a particular action is available under it.

  2. Make the application inspectable

    For each condition, record the supporting source passage, the relevant fact and any uncertainty. Ask for a concise, checkable justification. A long narrative about how the model supposedly thought is not evidence that the application is sound.

  3. Test a fact that could change the answer

    Change one material fact or introduce an exception. Check whether the analysis changes for an identifiable reason. This helps distinguish a workflow that follows the legal structure from one that repeats a familiar conclusion.

Watch for

A source can be quoted correctly while its scope, an exception or its application is wrong.

Selected experiments

Tools and experiments
in practice.

Behind probably right. LinkedIn

Anton Kraev Christensen

Lawyer. Mediator. Founder of probably right.

Anton Kraev Christensen, wearing clear glasses and a navy suit.
Anton Kraev Christensen · Copenhagen

Experience

Head of EU & Employment Law

Dansk Erhverv · 2024–present

Previously: legal and leadership roles at Teknisk Landsforbund, Dansk Erhverv, TDC, HORESTA and Accura, across employer and employee organisations.

Why probably right.

I build practical legal AI tools and share what the experiments teach us: better sources, clearer workflows and work that can be checked, with human judgement at the centre.

I also want to bring lawyers, researchers and builders into closer conversation. By sharing methods and inviting challenge, probably right. can help others develop their own approaches as the technology changes.

Education

  • Master of Mediation & Conflict Resolution 2022
  • Danish Bar admission 2019
  • Master of Laws (cand.jur.) 2016
  • Cand.merc.(jur.), MSc 2012