News You Can Use

Clifford Chance shared details about their full roll out of Copilot. They have not been working with other vendors for the time being, focusing on what they could get out of Microsoft. CC are in a strong position through previous development work and close relationships with Microsoft, with their data science team working on AI for a number of years.

CC have rolled out Copilot with a strategy to address the risks, spending time on setting ground rules for the wider IT and Innovation team, led by the CIO and Chief Risk Officer. It still shows that not all use cases can be done with Copilot, and they are working on a case by case basis, with some difficult conversations with clients around data and use of AI. Key use cases for CC seem to be summarisation and search, so working within O365 for internal support rather than client facing work. It's a good case study to look at, as we have similar challenges, but we have looked to other solutions and focused on internal development, rather than Copilot.

PwC's annual report highlights financials, tech, and strategy trends, with a new focus on GenAI adoption. Key findings show a divide: the Top 10 law firms are optimistic about GenAI, with 83% seeing it as a way to increase client work, while firms ranked 11-100 are more cautious, fearing it may lower fees or reduce demand. Notably, 90% of the Top 100 firms have trialled or implemented GenAI (up from 55% in 2023), though only 19% report measurable gains in productivity or revenue.

These insights can guide client conversations, showing GenAI's uncertain impact across the sector. Only 1/3rd of firms expect cost savings from GenAI, with many concerned it could shrink work volumes. Internally, this highlights areas of resistance and external trends we'll need to address with our team and clients.

Two updates on AI Agents from the market, Microsoft launched Agents at their AI Tour recently, with some additions to Copilot Studio and some case studies with Clifford Chance and McKinsey. Claude also launched the ability to run Claude on your computer, meaning you can give it control and assign it a task to achieve. This is much closer to the actual vision of an AI Agent.

Agents are getting closer and closer, with solutions in the market allowing you to build out more complex workflows that are AI driven. Claude's release is the closest so far to a fully autonomous agent that can achieve specific objectives, but the potential in other solutions is that we can build out a workflow that deals with triage or instruction management without human input. The Copilot Studio case study from McKinsey is similar to what we want to do with a future Legal Front Door, so work is underway to properly test this.

There are currently a number of benchmarking initiatives across the market trying to score different AI Vendors. We are part of the ongoing LITIG one, but there was recently an announcement from Vals.ai that it is working on a vendor led benchmark with The Legal Tech Hub.

Our work in The RAG Report has garnered interest in relation to benchmarking, with people asking whether we would like to support other initiatives. It is important to figure out how we grade AI tools and when we decide if they are good or not, at the moment this is subjective, so work to develop a standard will be well worth it. We will still be running our own tests, but will look to be involved in the market where relevant.