What ‘Agentic AI’ Really Means for In-House Legal Teams — and Why It Matters

European Business Magazine caught up with  Ruben Miessen, Co-Founder and CEO of legal tech startup, LegalFly to discuss company LEGALFLY positions itself as “agentic AI” rather than just a legal copilot.

In practical terms, what does that mean for an in-house legal team using the platform day-to-day — and why does that distinction matter commercially?

The distinction between legal co-pilot and agentic AI is important. The way we see it, a ‘legal co-pilot’ is something that has been trained to speak ‘legalese’ and is therefore limited to a very conversational interface. For example, these enable users to undertake simple Q&A-like use cases, which are relatively simple — therefore a legal co-pilot cannot be used to execute the legal work required by an in-house team.

It is our ambition to integrate LEGALFLY within the existing suite of tools used by internal legal teams, and to mimic their current workflows as precisely as we possibly can – something that a co-pilot cannot do.

We therefore position ourselves as ‘agentic AI’. We build workspaces within a setup of more than a dozen agents, which have each been developed in a unique way to mimic very specific legal tasks. We have an agent that is conversational and used for onboarding; agents that you can drag and drop a contract on and it will output a version with risks flagged on individual cards; a red lines / redraft class that you can then insert into a Word document to redraft any contract in a matter of seconds. This broad service is something that cannot be achieved with a co-pilot in a solely conversational interface. We’re doing this because we want to empower our in-house clients. LEGALFLY executes all the core legal work we can (about 80% of an in-house legal team’s work), enabling them to refocus their time on more strategic tasks.

In addition, we’re about to take this one step further. The Discovery agent (which is our conversational interface) acts as the intake of legal work, and whatever question you pose in Discovery, it will route it to all the other agents. Therefore, when you ask a question and insert a contract, Discovery will activate all agents and convert the contract into a red line document and a workflow which is no longer conversational – like Microsoft Co-pilot. We have also integrated with email: customers can send any email to LEGALFLY, for example forward a contract, and one minute later LEGALFLY will reply with a red line rewritten contract, based on your internal company policies.

Building features is handy, but ultimately this has a huge impact on company ROI by speeding up internal processes. A legal copilot is limited in its capabilities. For example, it might be able to offer first line legal support from an advisory perspective, but we didn’t want to build a solution which only advises. We wanted a product that gets things done, which is only possible through an agentic solution. We’re aiming for an ROI of at least 10x on the licence cost for the client, versus what they are saving in legal spend. More broadly, we’ve already witnessed a significant macro-shift in the market, where legal spend is being pulled away from law firms and ASLPs, and invested in legal tech solutions to make the internal team more efficient and enable them to handle a broader range of legal tasks, versus simply outsourcing it to a service provider.

In legal tech, data security and confidentiality are arguably more important than raw AI capability. How did you architect LegalFly to win trust from enterprise legal teams, and how does that differ from the way big US foundation-model vendors approach legal workflows?

There are two elements to my answer here: accuracy and legal depth, and data security.

First, on accuracy and legal depth: generalist solutions are good, but they have a high error rate – currently sat at 31% with solutions like Co-pilot or ChatGPT. So, whilst AI might be useful to carry out more creative work, you simply cannot take the risk in legal tasks.

The legal depth of the solution is extremely important. LEGALFLY is legally trained in 35 jurisdictions, with live access to over 250 government portals which feed the platform directly at the source, to track and incorporate any changes across the legal landscape which have been published by public and government agencies.

Whenever LEGALFLY provides advice, every answer is grounded in a reputable, authoritative source. This means that the legal professional using the tool can very easily verify sources within seconds by clicking on the link. On top of that we run a confidence check in the background, offering a secondary back-up from our internal knowledge base. If LEGALFLY is not confident, it will not offer any answer, or if unsure, will answer with a disclaimer which tells the user to seek further legal counsel. This is a crucial distinction from LLMs which currently may hallucinate answers to achieve a goal.

Second, on data security: one of our principal design cores is an anonymisation model that can even be deployed on premises. This model redacts all sensitive data from every single document or contract before an AI agent is connected – we’re the only provider globally that does this. This is one aspect which is highly sought after by sensitive industries and as a result, I’m very proud to say we’re working with several governments, including the government of Luxembourg. These are the type of client very concerned about security when deploying AI on their most sensitive documents. But thanks to our unique approach, we can help them feel comfortable around using AI in legal tasks.

We also care deeply about data sovereignty, so we offer a wide range of deployment options, including single tenancy which we host in every region, whether that’s mainland Europe, the UK, Middle East, or US.

You’ve built LegalFly in Belgium, with operations in London and Dubai, rather than San Francisco. What advantages — and constraints — come with building a legal-AI company in Europe, especially around regulation, data sovereignty, and enterprise sales?

Europe has a tendency toward hyper-regulation. This means it’s tougher to bring legal AI solutions to market in Europe, compared to the US. But the benefit is that as soon as we get it right here, then we can get it right anywhere, because Europe operates one of the strictest legal frameworks in the world.

As a Belgian company, we benefit from understanding complexity. We have six jurisdictions in one small country, and three official languages, which all need incorporating into training. An additional plus is that the European Commission is headquartered in Belgium, which gives us greater access to the legal system and legal policymakers in the region too.

From a commercial perspective, it can be tough. There are 27 member states of the European Union, all with different cultures, languages, jurisdictions in which we need to train LEGALFLY. Development comes with its own challenges, which in the short term slows us down versus a US company. But in the longer term, it puts us in a unique position. If you look at the clients we’re already selling to, they’re working across all those jurisdictions we already know by heart, so we’re building with that knowledge and those requirements taken into consideration. In terms of data sovereignty, we host data in the UK, Germany, UAE, and US, with the capability to do further deployment should that be required.

However, one key constraint is on fundraising, which is certainly tougher when building in the EU than the US. Thankfully, this hasn’t affected LEGALFLY. For us, VC was inbound for both our Seed and Series A rounds and continues to arrive.

When companies like Agristo or Duvel Moortgat deploy LegalFly, where do they see the fastest and biggest return on investment — cost reduction, risk management, deal velocity, or something else?

This depends on which team you are looking at: to expand, Legafly is indeed solely focused on working with in-house teams (96% of clients are in-house or public sector), but we’re not just selling to their legal teams. LEGALFLY works with Legal, Procurement, Compliance, and depending on the client, the Claims team. In most cases, at present, we’re working with Legal and Procurement teams, and in most cases, we work with both. But each team has its own requirements.

In Legal, the incentives are a little different: it’s usually purely about cost prediction, with significant reduction of reliance on the costly third-party legal counsel. If you’re asking the Procurement team, it’s mostly about deliverables; with other important considerations being deal velocity, cost reduction and reducing risk.

When we talk about cost reductions, it’s important to distinguish that it’s not about reducing the size of the legal team, because even in large organisations, internal legal teams are already rather small, operating with an intense amount of pressure. Instead, it’s about giving the in-house team independence from their legal counsel and shifting that legal budget spend from the law-firm to only spending a fraction on the AI. That’s not to say the law firm will be cut out entirely, as there will remain some specialised tasks, like litigation, but many other functions can be brought in-house.

With Microsoft, Google, and OpenAI all moving aggressively into legal and compliance workflows, what is LegalFly’s long-term moat — and how do you avoid being commoditised as “just another AI layer” inside Word and Outlook?

We have a strategic partnership with Microsoft, where clients can buy a LEGALFLY licence through Microsoft.

But why did Microsoft become interested in this partnership? We have all seen multinationals, such as Microsoft, add AI into every existing solution and product. They have achieved their success by being world-leading generalists. However, the one area where it would be difficult to sell a generalist solution is in legal. Microsoft’s existing product suite is not legally trained; plus, a legal co-pilot is not an agentic legal operating system, so it has zero capability to actually ‘do’ any legal work, besides advice — and even then, it may even hallucinate.

LEGALFLY’s ability to anonymise documents — especially sensitive documents — is also key to our success. Despite current behaviour, it remains unsafe to upload documents to ChatGPT or Co-pilot, particularly in a legal environment.

That is exactly why Microsoft has decided to partner with us, specifically for those Legal and Procurement teams. We’re definitely not just another AI layer. Inside Outlook we’re a legally verified solution to undertake legitimate legal tasks.

Do you see LegalFly remaining a best-in-class legal automation platform, or evolving into something closer to an AI operating system for corporate legal and compliance functions across Europe?

Last week, we launched V3 of LEGALFLY as an operating system. The more time we spent working with our clients’ amazing Procurement teams, the more we learned specifically about the processes and tasks we want to build an agent for.

We’ve now reached a point whereby we have built an agent for any task an in-house legal team is undertaking. As a result, our clients can spend an entire day within the LEGALFLY platform, so it essentially became a de facto legal operating system.

We have also ensured LEGALFLY is easy to use across platforms: it can be used as an agent, as a web platform, in Microsoft Word, to email, to Slack, through Teams. Therefore, we’re deeply integrated through any system on which our clients prefer to work.

Looking ahead, how do you think AI will change the structure of legal teams in large European companies — fewer lawyers, different skill sets, or simply much higher leverage per lawyer? And where does LegalFly fit into that future?

I don’t foresee a huge structural change for in-house, corporate legal teams. This is largely because they are already rather small, so there’s not a huge amount to change! To speak from experience — we meet with the world’s largest public institutions, airlines, construction companies, banks, insurance firms, and so on — the amount of legal work, and the number of risks that these small-and-mighty internal teams are defending the company against, is surprising!

So, whilst I don’t foresee a change for internal teams, I am certain that there will be a big change in the structure of law firms or ASLPs, where we expect junior hiring freezes, and slimmer law firms overall. This is because in-house teams will be powered by AI, putting a lot more work in-house, and therefore less money into the pockets of the law firms, which will significantly increase over time. Secondly, those law firms are becoming much more efficient, so you won’t need a full army of lawyers, even in the prestigious magic circle firms. It won’t make sense anymore.

The post What ‘Agentic AI’ Really Means for In-House Legal Teams — and Why It Matters appeared first on European Business & Finance Magazine.