
Luminance, which has more recently embraced LLMs and gone after inhouse teams, is launching an agent feature called Lumi.
The UK-based company said that their agent capability could carry out a series of chained tasks, such as ‘sales representatives can ask to “mark-up this incoming NDA and trigger any approvals required”, and Agent Lumi will automatically work to:
– bring the contract into line with company standards,
– then seek approvals,
– then provide a summary to the user,
– and then generate a new version ready for the counterparty to review.’
Of course, could each of those individual steps be handled one at a time in the ‘usual’ way? Yes. But, if Lumi works as Luminance claims it can, then this will provide a significant efficiency gain.
Plus, the challenge with agents is making sure each step is performed correctly before moving to the next, otherwise you amplify the errors and spread them like a bad cold.
However, Luminance said: ‘The user can… see and refine each step of the thought process.’
And again, this refinement process is no small task either. If you talk to companies such as Flank, which has been working with legal agents driven by LLMs for some time, it’s actually a tough problem to solve. It can be solved, but should not be underestimated in terms of difficulty.
Plus, if a human lawyer has to spend a lot of time reviewing and checking that each step is done correctly, then it rather removes the benefit of having an agentic flow, i.e. a bunch of things that are meant to be done via a series of automated steps without you getting too involved. So, to really get the value from agentic flows you really need to be able to rely upon them to get it absolutely right….and genAI has a habit of making mistakes unless it’s very carefully controlled.
Therefore, when you hear ‘agentic flow’ in relation to a collection of quite different chained tasks, it’s best to test out such promises very carefully.
Luminance goes on to say: ‘The AI not only completes these tasks in seconds, but it also has both short-term and long-term memory, enabling it to retain and recall information, learning from past interactions with individual users to tailor its actions and responses over time. With this memory and chain of thought enabling better logic, calculation, and decision-making, Agent Lumi can proactively flag information to optimise business efficiency.’
‘For instance, if business users are consistently negotiating and amending the same term in a contract, Agent Lumi can automatically notify the legal team of this inefficiency and proactively update company standards and templates to ensure smoother negotiations moving forward.’
Eleanor Lightbody, CEO of Luminance, added: ‘By delegating everyday legal tasks to Agent Lumi, business teams will be able to maximise productivity, negotiate faster, accelerate deal cycles and ultimately drive business growth.’
Graham Sills, Co-Founder and Director of AI at Luminance, concluded: ‘Automation of tasks is only helpful when it is built on the foundations of reliable thought processes that the lawyer can review and critique.’
And Sills is 100% right on that point…!
But, it’s good to see more companies exploring agents. With time, better LLMs that are suited to such tasks – such as perhaps much better reasoning models that advance upon where OpenAI o1 currently is – and more application layer development around the quality-checking process, then we’ll get where we need to be.
The challenge, as always, is that ambitions are often running a little ahead of achievable reality when connected to the messy, nitty-gritty of real-world legal needs inside a very human-operated business. But, as said, we’re making good progress, and rapidly, and we’re only going to get there by trying.
….In fact, in terms of progress, watch this space…!