We have spent a lot of time with clients this year, with involvement in specific matter work, running workshops on GenAI and building hands on solutions together. Each year the in-house teams we work with continue to evolve but we also meet more at the start of their journey.

There has been more of a push towards digitisation from GCs this year, and from conversations it seems a lot of this is led by C-Suite engagement in technology, most likely driven by market hype and interest. Interestingly this really just unlocked the backing to push on the use of tech, rather than changing mindsets of GCs as most GCs we speak to are already on this journey, but need support.  

A consistent message has been the overwhelm experienced by GCs with the market, there are 100s of companies now selling legal tech direct to in-house teams. This is hard to manage when your day job is focused on running a team and delivering legal services, with most companies not being lucky enough to have a dedicated Legal Ops function.

Our Client Projects team have spent this year in delivery mode, working on large projects across the firm and embedding technology directly into the legal services that our lawyers are offering. This has involved building out platforms incorporating a range of technologies, including GenAI, and empowering our lawyers to work collaboratively across teams and offices. Bringing together expertise and resource from across the firm to ensure we can deliver a full service offering powered by technology.

A key focus has been giving our clients usable outputs that fit easily into their business model, so alongside legal reports and detailed advice, giving them data, dashboards and integrating direct into their systems.

This year has also seen a step up in the amount of 'tech only' work that we are doing, where clients come to us directly as an Innovation team to help deliver on an ask, without legal advice. This has involved largescale data extraction and contract reviews based on the concept of just providing useful and usable data, rather than the substantive legal advice on the risks. This is a great example of new business models and opportunities unlocked by advancements in tech, whereby work that was not previously economic to do manually becomes a service tier option for our clients.

2024 saw us publish The RAG Report, which is a report detailing our research into the use of LLMs for M&A due diligence. The work that went into this was a great opportunity to learn about some more of the technical aspects of LLMs and Retrieval Augmented Generation (RAG), bringing us further along in our journey.  

We think law firms have an important voice in the market, especially when it comes to benchmarking and assessing technology. Sharing our findings so publicly was a concerted effort to lead by example and show the impact that you can have by being open. It's been great to see other firms doing similar, as well as a number of initiatives that are exploring how we better carry out this work.

This leaning has meant that we have been able to have much more in-depth conversations with software vendors about their approaches to search, chunking, LLM selection and RAG vs full document context reviews. Knowing the challenges we had on our journey means it is much easier to assess software products with targeted questions.

There has been a huge wave of new software vendors this year, at the recent TLTF Summit the keynote called out the fact that it has never been easier to create a company. With the funding environment, the growth of AI, and interest in legal specifically all creating a perfect opportunity for founders to launch their ideas.  

Whilst this growth is great, it has created a much larger market than before, which can be difficult to navigate for lawyers, especially in-house teams. We had had lots of conversations with clients in relation to the market, helping narrow down solutions and focus on problem statements and requirements, but this is still a huge space to try and navigate.

The balance between having one large tech provider giving you everything and a collection of point solutions has slightly started to tip this year. At AG we have begun to embrace point solutions more, thanks to their deep focus and dedication to solving specific problems. Although at the same time we are seeing more acquisitions in the market as the larger players try to maintain an advantage, which often brings us back to a platform.  

It is a challenging landscape trying to sell into law firms and lawyers, with slow sales cycles and high expectations. We are seeing some software vendors struggling with the market and moving to focus mainly on corporates, rather than law firms directly. 

We decided on a hybrid buy and build strategy for GenAI very early on, and have seen some great success with our internal tool, AGPT, with over 90% of the firm now using it and averaging around 1,500 prompts per day across our offices. This has also spread beyond just GenAI, with our Platform Delivery team working on a number of internal build projects across different use cases.

The ability to learn whilst we build has supported our external engagement and consultancy work, allowing us to give genuine and useful advice and support to our clients from firsthand experience.  

Our need to build solutions ourselves has come out of the lack of really bespoke and working solutions from a lot of the software vendor offerings. This isn't saying that the tools in the market are bad, but we have struggled to solve our specific problems with off the shelf solutions, especially in relation to GenAI. We are having more success using LLMs in our own internal builds for the difficult to solve problems.  

We have seen some of our software vendors applying GenAI into what are already good products and reaping the benefits, so the buy side of our strategy is still focused on having the best in class tools.

A key benefit of building our own solutions is the effect it has on adoption, AGPT has been adopted almost firmwide, with great retention and daily usage stats, but we think a huge part of this is due to the "AG owned" aspect giving users confidence around data security and control. This boost to adoption cannot be understated.

AG went big on Legal Geek this year, with our stand, talk from Kerry, prompting workshop and team presence being a really great way of showing what we have been up to this year. Our gaming theme pushing a "get hands on and play" agenda reflected what we have been saying for a long time about technology.

The feedback has been great about our messaging and sessions at Legal Geek and it was good to see a mixture of clients, suppliers and other law firms engaging with what we are doing. We have tried to be open, believing that the more we share the more the wider industry is likely to progress, and Legal Geek was a great platform to do this.

We have been at other conferences but as the "Glastonbury of Legal Tech" we were glad to be so present!

Clients have continued to be a driving force behind our internal lawyer engagement at AG. We have been part of more pitches than ever before, allowing us to show what we can do directly to clients. The interest from clients around GenAI which has required our own lawyers to get up to speed.  

The initial hype around GenAI actually allowed us a platform to talk about all the other great things we do, and 2024 has been another year where initial questions about whether AI can help have led to us introducing non-AI technology into a lawyer's workflow to great effect.  

We have undertaken a huge amount of training internally this year, running our prompt like a pro sessions, running team by team AI and tech introduction sessions and client talking points conversations with partners. We have kept adding to our internal library of knowledge around our Innovation offering and continue to work with all of the Groups across the firm.  

Innovation is a key pillar of AG's 2030 vision, and the engagement from senior leadership has been putting the wider Innovation Group on the radar of all of our lawyers.

Slowdown of LLM progress may result in a bigger focus on the application of SME knowledge alongside existing models to create niche specific / task-oriented solutions.

2025 will potentially see a slowdown in the growth of foundation models, we are already seeing this with the move from OpenAI being iterations / improvements rather than a big push to GPT-5. Growth will most likely be focused on achieving gains in task specific areas.

In legal this will be reflected in people finetuning or adding additional context into these LLMs to corner a specific niche, rather than the current approach of broad models that the user can tune through prompting or chaining actions together. Some of the newer LLMs, like GPT-o1 are being touted as being able to 'reason', so it will be interesting to see the impact these developments have on the use of LLMs in agentic workflows.

GenAI tools will become integrated enough into work delivery methods to enable data unlock at scale, enabling rapid deal execution and more advanced data analysis.

Building on foundations developed through delivering largescale projects for a number of years now, we will see further leveraging of data to support clients on the biggest and most complex legal projects they undertake. GenAI allows better access to unstructured data than ever before, but the most important point will be how this is inserted into the process.

Next year will start to see some teams deliver end to end review processes with GenAI as just a specific part of the extraction and analysis of data, within a wider platform. At AG we have started to deliver this vision for clients already, and are looking at how we offer an internally built solution using the same concepts for M&A diligence. Many of our clients are looking at doing this themselves and we may see a big push on using data extracted through GenAI to inform next steps and decisions on transactions.

Savvy in-house teams will unlock budget and resource through leveraging the wider company's AI strategy as well as taking advantage of their outside counsel's learnings from private practice application of AI.

For in-house teams the discussions are moving beyond "What is GenAI?" to "How can I leverage this technology for what I do?". This has also resulted in increased demand for our workshops to leverage GenAI technologies like CoPilot which had often been rolled out across the broader corporate estate.

Going forward into 2025 we are going to see in-house double down on this approach and other areas such as CoPilot Studio and Microsoft's Power Platform, as they are unlikely to have to find the money for this out of their own already stretched budgets.

The more teams that do this the more it will spread as communities such as Crafty Counsel help GCs share, with in-house being a lot more open than private practice. We are seeing teams with good internal knowledge banks leverage this approach with success, and I am expecting this to grow throughout 2025.  

Beyond AI, the improvement in general legal tech solutions through better pricing, more robust infrastructure, and enhanced features means that in-house teams have better options. The evolution of teams like ours and others in the market means they can also get implementation and adoption support from their trusted advisors, further driving the digitisation of their legal team.

People want to be able to compare solutions in the market, pushing vendors to share benchmarking and metrics on performance. We will see this pick up speed however there will potentially be many differing standards.

This is an area that we put considerable time into this year, with our own testing and our subsequent RAG Report detailing our findings on how you could improve the performance of LLMs in a legal context.

We have not seen the same kind of transparency in the industry when it comes to the accuracy advertised by vendors when using LLMs for specific tasks. Benchmarking is an area that we plan to increase awareness of in 2025 and we have already had discussions with industry peers about how we might discuss this and raise awareness. Yes there is talk of benchmarking collaborations in the industry from LITIG and Vals AI's Legal Bench but we think it will need to go further than broad benchmarking standards.  

We think throughout 2025 we will see more detailed metrics in relation to performance on actual tasks. Hopefully this will be led or heavily contributed to by law firms / lawyers. There is a difference between benchmarking and understanding the performance of LLMs vs legal specific solutions, and some of the work Harvey have been doing in this space is good to see. It would be great to see more of this both from those using the tech and those building it.

A hybrid approach will still be the best bet, with a mixture of purchasing best in class solutions alongside bespoke builds or potentially builds alongside delivery partners being a core component of most law firm's strategies.

With the explosion of software in the legal market over the past few years, you would be forgiven for thinking that there is no need to build anything as a law firm. In many cases that can be true, but not all cases. For example, in 2023 if you had the skills, it did make sense to build your own GenAI chat platform. But if you haven't started the build process for this already it is unlikely to be successful doing it now. That is unless you are buying an incumbent consultancy / solutions provider (it wouldn't surprise us if this happens) which would catch you up pretty fast.

Bespoke tech companies that have spun up GenAI platforms out of the hype of GenAI are unlikely to survive unless they receive significant investment or capture a niche area. To truly be a technology supplier in the modern legal market platforms have to be slick and have a core purpose. Those that are broadly configurable and require certain levels of expertise are unlikely to be picked up by the more mature innovation or technology teams. They have either already built this capability themselves or are partnering with much larger providers in the market. Conversely those without these types of skills may struggle to leverage a "GenAI" platform.

We think this year will see more bespoke solutions that differentiate themselves from others by how they apply domain expertise to build tools that solve specific problems. Rather than relying on the developments in LLMs to evolve the tech, they will do it through using knowledge. This opens the potential for law firms to do some of this themselves, as you can only really deliver on this with the subject matter expertise. It may also create a good market for those 'selling shovels', like workflow / agent building solutions.

This isn't to say that build is dead, those that took this path early are reaping the benefits when it comes to engagement with GenAI across their firms. But they still need to be careful not to just try and recreate what existing larger software players are doing well.

Beyond this, Build is alive and kicking when it comes to building for bespoke business needs. With what GenAI technology can now do, we predict that firms will start to build complex tools that really get into those hard to solve problems when it comes to the practice of law. This is also where we see the potential for true partnerships, not just buying a piece of tech but working with a selected supplier for the build of an actual practice of law solution.

We recently attended the TLTF Summit and it was quite clear from our conversations that most legal tech startups are focused on building technologies that solve a specific need that only has a surface layer of complexity. Having discussions with them about the trickier problems we are trying to solve often resulted in responses along the lines of "That's really hard to solve. I'm not sure there is enough money in that.". So it's clear that the perceived ROI isn't there.

Smart workflow solutions and 'Agents' will take off due to the capabilities now available being much further ahead of where we have previously been when building process into applications.

A potentially grandiose term for what is essentially connecting workflow to GenAI, but something that is being talked about in the market in the form of AI Agents. The ability to chain different LLM powered 'personas' or task-oriented solutions together is really powerful. We are likely to see more complex use cases being tackled by Agents throughout 2025, with early success already being seen when applying this to simple contract review and knowledge retrieval for Q&A.

At AG we are building out solutions that could fall under the Agents definition more widely, however we see these as complex workflows with elements of AI under the hood alongside more traditional logic and decision tree processes. As LLMs evolve we may be able to rely on them for more complex chains of reasoning, but for now there is a heavy validation and supervisory aspect still needed.

The risks of GenAI will start to become clearer, with less focus on the data / training risks as people start to understand the technology and more focus on potential over-reliance, how to verify effectively and ensure human oversight.

Fundamental understanding of the underlying tech behind GenAI is required in order to properly advise on the risks. As things stand, this understanding is not quite the same across the market. As AI becomes more prevalent in software solutions and more embedded into the day to day delivery of legal work we are coming up against the challenge of being unable to use it with clients or on certain data. A lot of this comes down to misunderstanding of risks and an overzealous approach to the use of data.

We hope that throughout the year the genuine risks become much clearer and separate from the perceived risks. We would wager that there is not a single law firm that feeds data back into an LLM's training set as part of their use of AI, yet this oftens restricts usage. Sensible conversations are being had across businesses at the moment and we need to make sure that legal isn't slowing this down, this will be achieved by better understanding the underlying technology.  

As people learn what type of things to look for, in the same way they know what to look for when checking human output, we will start to see less mistakes and more of a focus on a clear delineation on what AI can do well, and what should be left to people.