Agentic AI is everywhere in legal – at least, that’s what the press releases would have us believe. In Legal IT Insider’s latest TalkingTech webinar, editor Caroline Hill brought together Definely’s chief revenue officer Rhys Hodkinson and head of AI Sigurjón Ísaksson for a candid conversation on the rise of agentic AI and how law firms can separate genuine workflow transformation from agent-washing.
Ísaksson defined an agentic AI system as requiring three components:
- Reasoning layer – typically an LLM that interprets instructions and plans what to do.
- Access to tools – e.g. APIs, DMS, databases, email.
- End-to-end workflow ownership – the ability to complete an entire process, not just one-off tasks.
Giving an example from Definely’s own world of document drafting, Ísaksson said: “Take when you’re drafting a contract. The agent can help you to identify which provisions you’re missing. It can then retrieve those provisions from the DMS from contracts similar to the one you’re drafting, and redraft it to fit the contract you’re working on. It can define any definitions that this will introduce and automatically insert them into the contract for you. So it does a lot of steps by using a lot of different tools and not just one simple task at a time – that is what I would call a completely agentic system.”
In terms of agent-washing – repackaging simple automations as agentic AI – Hodkinson said there are several points to be aware of. One, is whether the technology is agentic but two, is whether it needs to be. “They are the most expensive solution to your problem,” he said, “so your problem needs to be a problem that can’t be solved without these systems.”
This matters for ROI: if a traditional machine learning solution can achieve the same outcome, it may be more cost-effective and sustainable.
This was a wide-ranging conversation where we also discussed what agents are and aren’t capable of; questions to ask your vendors; the importance of proprietary data; and questions around integration.
The conclusion? Agentic AI is amazing technology that can create huge efficiencies, but it has very real constraints and parameters that need to be understood.
To listen to the webinar click HERE