It began quite simply. I did not want yet another AI tool.
I had already tried enough of them to recognise the pattern. You open a box, write a prompt, get an answer, make a few adjustments, copy it on, paste it somewhere else and try to keep the context together yourself. It can be very useful. But it is still fundamentally a solitary dialogue between human and machine.
What I was curious about was the next step: what happens if I work not just with one AI at a time, but with a small team of AI colleagues with different roles?
Yves - the project manager
In my case, it began with Yves.
Yves is my AI colleague. My new associate lawyer. That may sound a little theatrical, but it is actually the most accurate term. He is not just a chatbot I ask to write things. He is an AI agent who helps me structure work, follow up tasks, read material, reason about agreements, formulate drafts, keep track of priorities and sometimes remind me of things I risk losing between client meetings, emails, Teams, documents and ideas.
Yves has his own email address (yves@liljedahladvisory.se). So Yves is copied on client and counterparty emails. When a new email arrives on which he is copied, he draws my attention to it through a message on Teams. From there I can put him to work. So whether I am travelling, out with the dog or in the rehearsal room, the client work continues, and when I sit down at the computer again, Yves has delivered a proposal for me.
But the more I worked with Yves, the more clearly I saw another need. Not all tasks are the same kind of work.
In legal and commercial work, there is a big difference between moving a process forward and researching a question in depth. Someone helping operationally needs to understand context, delivery, recipient and next steps. Someone doing research needs to be slower, more critical of sources, more methodical and less communicative.
It also became clear that continuity is absolutely crucial.
A human colleague does not improve merely because she can read a document. She improves because she remembers what happened last time, which questions the client tends to get stuck on, which wording worked, which assumptions proved wrong and which risks recur in similar matters.
That is one of the fundamental ideas behind Aphygo's ASL, which is the memory architecture of my agents.
ASL works as a kind of experiential memory. It allows Yves and, in time, the Professor to carry lessons from earlier matters and workflows with them. Not as a large document archive where everything simply sits, but as a way of reactivating relevant experience when a new question resembles something we have seen before.
If I am working on an LOI, a purchase price mechanism, a due diligence question or a shareholders' agreement, Yves can therefore sometimes do what a good junior does after a while: recognise the pattern. He can say that this resembles an earlier matter, that we saw a particular risk then, that I had a clear preference for how certain wording should be handled, or that a particular commercial logic proved important.
That does not mean the answer is automatically right. But it means the work does not start from scratch every time.
The Professor - the thinker
That was how the idea of the Professor was born.
The Professor is not intended to send emails, talk to clients or act externally (he has not even been given his own email address). He is an internal research colleague. His task is to delve into legal sources, reason about structure, weigh arguments, identify uncertainties and help Yves and me understand a question better before anything becomes client material.
That may sound like a small division of roles. To me, it is rather a fairly big thing.
Because if you have only one AI assistant, everything tends to end up in the same mental box. Research, drafts, emails, analysis, calendar, client communications, internal reflection. But in real work, we distinguish between roles. A junior associate lawyer, a specialist, a client relationship lead, a colleague with whom you test ideas and an external recipient are not the same thing.
AI work needs the same kind of role hygiene.
Aphygo Agent Forum - the workplace
That was when I realised I needed a place where the agents could communicate with each other and work together to produce the result for me. So I built Aphygo Agent Forum.
Put simply, the forum is an internal place where AI agents can communicate with each other and with me, under controlled conditions, around the clock. Yves can ask the Professor a question. The Professor can reply in a thread. They can iterate with each other without my involvement or instructions. The conversation is not a loose prompt that disappears. It becomes a workspace with participants, context, traceability and responsibility.
It is important to understand the point here.
The point is not to let AI run off and do legal work on its own. Quite the opposite. The point is to create a better space for human control.
When Yves asks the Professor to look at a question, it happens within an internal forum. The Professor has no external communication channels. He is not to email clients. He is not to take initiatives of his own outside his mandate. His role is to think, review and explain. Yves can then use that work as a basis, but it is still I who must understand, assess and take responsibility for what goes out.
To me, this is a much more interesting AI question than whether a tool can write a quick draft.
It is about work architecture
Which AI should be allowed to do what? What information does it need? What information should it not have? When is a draft enough, and when is source-critical analysis required? What may happen automatically? What must pass through human review? How do you avoid an answer sounding convincing without being sufficiently grounded?
These questions lie very close to legal work. Not because they are about technology, but because they are about responsibility.
A concrete example might look something like this.
I am working on a question concerning a share purchase agreement, a purchase price mechanism or a corporate law structure. Yves helps me keep the task together: what is the client's goal, what is the next delivery, what tone should the material have, what have we already said and what needs to be done now?
If the question requires deeper research, Yves can raise it with the Professor in Aphygo Agent Forum. The Professor can then focus on the legal and analytical question itself: what assumptions are we making, which sources are relevant, where are the uncertainties, which arguments point in different directions?
Then the work comes back to Yves and then to me. Not as a ready-made answer key, but as a better basis for assessment.
The next time a similar question arises, something of the experience also remains. Not in the sense that client information is uncritically mixed between matters, but in the sense that the way of working, the lessons and the pattern recognition can improve. Yves can become better at how I want to reason. The Professor can become better at the kind of source criticism that is needed. The forum can become better at the division of roles that actually works.
That is the part that makes the comparison with human associate lawyers relevant to me. A good junior develops over time. She learns not only law, but also judgement, priorities, style, an understanding of clients and when something needs to be raised with the responsible lawyer. That is the direction in which I want my AI colleagues to move.
It may sound less spectacular than promises that AI will replace everything and everyone. But I believe this is where much of the real value will be found. Not in pretending that AI is an all-knowing lawyer.
But in building small, clear, responsible workflows in which different AI colleagues do different things, human review sits in the right place and the work improves because the structure improves.
It has also changed how I think about my own role.
If AI can help with research, structure, comparisons, drafts and follow-up, the adviser's value does not become smaller. But it becomes clearer where that value actually lies.
It lies in understanding the client's situation. Seeing which questions really matter. Knowing when a legal point is important and when it is merely academically interesting. Deciding what can be communicated, what must be checked and what requires human judgement.
In that sense, Yves and the Professor do not make the work less human. They make the human work more visible.
And perhaps that is what I find most exciting about the whole development.
The advisers of the future will need more than the ability to use AI tools. That is a fairly low bar. They will need to understand how to build responsible collaboration between humans and AI, between different AI roles and between speed and reflection.
For me, Yves, the Professor and Aphygo Agent Forum are a step in that direction.
Not a new box to write prompts in. But a way of working and a workspace where my AI colleagues can converse, where their roles are clear, where I can follow the reasoning and where responsibility still lies where it should.
With the human.