Antti Kuha and Valter Pasanen on AI and the future of legal work

Experts: Antti Kuha & Valter Pasanen Read time: 7 minutes

Agentic AI systems are about to fundamentally reshape legal work. But the result, Antti Kuha and Valter Pasanen argue, will not be technology driven, it will be more human than ever. In this conversation, Hannes Snellman’s Board chair and legal AI analyst explore what that shift looks like in practice: from autonomous agents transforming daily workflows, to the risks no one is quite ready for, to AI’s potential in supporting skill development.

How agentic AI transforms legal workflows

Valter: Agentic AI in a nutshell is: you give an AI system a task, it looks at the desired output and available materials, fetches whatever additional data it needs, all by itself. We have been extensively testing agentic workflows at Hannes Snellman, and already today you can do meaningful due diligence this way. You could even try and give it access to the data room, tell it what the output needs to look like, give it a template, and it runs by itself, and using multiple agents in parallel.

Antti: So, the agentic model is the technical framework for using sub-processes to run through the material. But more importantly, you as a human can follow the reasoning. You can see the steps the agents took and how they were modified. If we have systems that can both do the editing and explain what they did and why, verification becomes much easier. The next step from this is the co-workspace model, where the agent is more deeply embedded in your work. And I believe it’s coming soon.

Valter: Yes, and what’s very much in demand right now are what are so called ‘skills’: essentially telling the agent what our standard steps are when drafting, say, a Share Sale and Purchase Agreement. At first you follow the reasoning closely, but as you start to understand what good output looks like, you codify those steps into a playbook or agentic prompt. At that point the agentic system moves quickly towards something that runs more or less autonomously.

Antti: This sounds like the system-level transformation that will really create efficiency gains. But it cuts both ways. The risk of something unintended slipping through grows as the system becomes more autonomous and complex, and the verification work is burdensome.

In any case, I believe agent-grade tools will have real significance across all knowledge work industries.

Valter: We’re already moving from requiring lawyers to know how to code or create intricate prompts towards lawyers needing to become leaders of agents. The key shift is understanding what can be delegated to the machine. Just looking at how quickly Claude Code emerged; things will look very different in two or three years.

At the moment I develop tools for our professionals and teams (For a closer look at how our Legal Tech Team approaches this, see AI as a Competitive Advantage for Lawyers). But I believe my entire job will change. Lawyers will go to our Hannes Snellman agent and say, ‘I need this kind of tool,’ and the agent builds it.

Antti: That’s a good way to articulate it. Even at the cutting edge, you recognise that whatever you are doing now in relation to AI will soon change radically.

So, is this the significant forthcoming change that we don’t see yet, being able to just speak and give instructions, and then have agents execute autonomously?

Valter: It could be. And this is what I genuinely think is gold, not just using AI as a chat interface, but using it to build things ourselves and improve our own processes. We can already compare thousands of documents in seconds, flag differences and get a first-pass review list almost instantly. These are small scripts. Simple tools. But the output is tremendous. And the point is not to replace legal judgment. It is to remove a lot of the manual work that comes before it.

Antti: This would be the best thing for lawyers. If all these macro tasks and disruptions could be automated, there would be more time for deep work.

And you can deploy layers of autonomous agents: One layer producing and finding things, a second layer filtering and evaluating relevance, running processes in the background. So that what reaches you is relevant to your context: not too much, not too little.

And it doesn’t stop at legal work. You might want to try incorporating an AI in Board work. Not as a full member of the Board of Directors, but as a sort of collective memory: “Have we discussed this issue previously, and how would you summarise the views that were expressed?”

Valter: People will get used to working like this, so why not give it a try already today?

Antti: It would enable you to focus on the hardest questions. When Board members have conflicting views, it can summarise the pros and cons of different view, offer a neutral perspective, and clarify what the common ground is.

Drawing from all this: what are the key drivers for value creation?

Antti: I think this will facilitate close dialogic collaboration with clients. The more AI handles technical execution aspects of the work, the more important it becomes to establish a mutual understanding of the context and the ultimate aims of the collaboration. When AI can quickly produce and modify long stretches of text, what we are aiming to achieve becomes more and more important than how it is done.

Valter: As interacting with technology becomes more natural and just integrates into your work, the human factor, good dialogue, trusted advice and relationships, becomes ever more crucial. The technology is there to facilitate, just like we can travel on an airplane without having to understand every aspect of aerodynamics ourselves.

But are we a little too techno-optimistic

Valter: Autonomous agents are exactly the kind of thing we should be testing early. That is how you stay creative and ahead. But give a system too much autonomy and suddenly the risk of accidentally blowing everything up becomes very real. The mistakes could go mega-scale. So the mindset has to be: test boldly, trust carefully and keep humans in the loop.

Antti: I believe we’ll see the first major AI-driven corporate crisis very soon.

And here’s the key tension: these tools are extremely fluent. They sound authoritative. You need the skill to know when they’re making things up, and that skill is not evenly distributed. It is mentally easy just to trust it. So the tool that was meant to democratise legal advice also risks causing significant harm if you don’t have the knowledge, experience, and patience to verify.

Valter: The liability question is genuinely unresolved*. Look at the standard terms for most major AI tools: liability is essentially disclaimed entirely. They say do not use this for high-stakes decisions.

Antti: So, the bottom line is that AI can raise the bar for what you need to know and can discern. You need to catch its errors, and to catch its errors, you need to actually know the law. You can ask AI to make your communications more empathetic, but you can’t outsource your empathy, because you’ll be just as lost in the next face-to-face interaction.

In the optimal scenario, AI tools will deal with the more rote aspects of daily work and support focus on value-add human interaction to solve the hardest problems and find common ground.

Valter: Working with AI has made me think about what is distinctly human, and how AI can also be used for interactive practicing of human skills, such as how to structure and make an effective live presentation.

Antti: The conversation keeps returning to the same theme: the more capable the technology becomes, the clearer it becomes which human elements matter in creating meaning and value.

*At Hannes Snellman, the liability questions are not theoretical. Our approach is built on a clear principle: lawyers remain responsible for every conclusion, and AI is used only when it creates clear value for the client. Every tool we use adheres to strict data safeguards, and all AI-generated outputs are validated by our lawyers. The liability question may be unresolved across the industry, but our position on accountability is not.