News You Can Use

Advancements in GenAI models now enable the creation of task-specific solutions tailored to deliver on specific use cases. These solutions are expected to become more embedded in workflows and increasingly integrated. This paves the way for building complex workflows that chain multiple AI-enabled agents to complete intricate processes. While tools like Bryter, Microsoft Copilot Studio, Autologyx, and Power Automate already make this possible, the challenge lies in model accuracy - small errors can compound as tasks are handed off. With improvements in reasoning of LLMs the prediction is that this will become easier to do.

2025 will be the year that Agentic AI shifts from vendor demos to real workflow deployments. For AG, this means prioritising accuracy, testing, and orchestration as we build agent-based solutions, and being realistic that the value is in the chain working reliably, not in any single impressive demo.

The age of training on huge un-curated datasets is probably over. More attention is now being paid to specific training data that can help fine tune and target models to deal with specific tasks better. Some of this will be in the form of grounding models in curated knowledge but some will also be going into the foundation models to improve their abilities. 2025 should see a bit of a focus on clear data to combat inaccuracies for specific subject matters (like legal).

Knowledge and data curation becomes a real competitive edge. Firms that invest in clean, structured, well-governed legal knowledge bases will get disproportionately more value out of the same underlying models than those who do not.

Small Language Models may emerge as a way of combatting the drawbacks of both the above. Training models specifically on more niche data for niche tasks could be a way of delivering bridging AI capabilities between LLMs to create more robust agent workflows for example. This could also be a way for smaller enterprises to get involved in training models, potentially law firms looking at training a specific SLM for their internal use.

SLMs lower the barrier to building purpose-built tools for narrow tasks and could open up a middle ground between "use the frontier lab" and "build it yourself". For AG, this is worth watching as a future option for specialist use cases where a focused SLM could outperform a general model at lower cost.

In 2025 we have to figure out how to make it as easy as possible for humans to verify AI output. The challenges at the moment are around people not being able to verify easily and either not engaging or missing errors. Training will help with this but potentially combining narrow AI with GenAI may allow us to deliver more deterministic outputs that give structured answers, rather than text that must be checked.

Verification is the gating factor on adoption. Products and workflows that make it obvious where an AI output came from, and easy to confirm it is right, will win over ones that just dump text at the user. This is a design problem as much as a model problem.

Legal Tech as a market is getting more attention from investors and even some of the big tech companies, with Amazon, Google and Microsoft all dipping their toe in 2024. With hype and potential comes money, and many people expect the space to continue to grow with lots of new start-ups, funding and acquisitions in 2025.

The influx of capital will drive consolidation, more competitive pricing in some categories, and a much wider set of tools to choose from. For AG, this is both an opportunity (better tools, more options) and a risk (vendor volatility, noisier evaluation process). Vendor due diligence becomes more important, not less.

Recent regulation has been putting pressure on those using AI to fully understand it and have clear training. The further prevalence of it in day to day work is also meaning that people need to upskill and learn more about the fundamentals. 2025 could see a concerted effort to educate the majority of people on the benefits but also risks of AI. Hopefully this is a trend we can capitalise on with our GenAI Workshop offering.

Training and education is where genuine adoption gets unlocked, and it is also a commercial opportunity. Our workshop offering is directly aligned with this trend, and we should be prepared for increased client demand as regulation and internal pressure tightens around AI literacy.