How law firms have used AI in the last year

A rundown

Based on its findings, the AI adoption journey began with strong enthusiasm and was disrupted by critical product failures, which have since stabilised into a “more mature, evidence-based pattern”. A “baseline of curiosity and exploration” was established in September 2024, with a “future-focused interest” in AI workplace use. In March 2025, there was friction among users who adopted AI in their legal tech stacks, with respondents reporting “system instability, hallucinations, and inability to deliver basic tasks as promised”. The research found that by September 2025, a balance between positive and negative experiences surrounding AI formed – “where both advocates and detractors coexist”.

Early adoption

Its research revealed that legal practitioners overwhelmingly experimented with newly introduced AI features in September 2024. During this period, users were interested in how AI could be adopted to improve legal software, found AI helpful but expensive, noted that paid solutions generated some of the same answers as free AI software, and showed interest in further AI advancements.

In December 2024, as practitioners started to have more hands-on experience, there were stronger sentiments expressed. The data found that early adopters were split between being enthusiastic champions and strong rejectors. Some users found the AI feature “one of the best AI programs for work”, others called AI a “plague”, and some advocated for a “no-AI office policy”.

A ‘crisis’ period and recovery

VIEW ALL

Following this, a “crisis” period emerged in March 2025. This resulted from product failures, system instability, hallucinations, and the inability to perform promised tasks. One respondent said: “It is so bad that I do not believe an AI program is behind it. I suspect there to be a sham. It cannot read documents ... If you ask it a basic legal question the response is usually so vague or inaccurate I need to go to Google,” while another said: “Extremely limited document uploads ... cannot accurately do basic economic loss calculations ... when you ask on what basis, it is incapable of telling you why.”

Early recovery occurred in June 2025, where daily usage patterns emerged among satisfied users, while others remained disappointed. Optimistic and frequent users discovered integrated tools that they implemented into their daily workflows, while simultaneously managing their expectations. Users have also begun to differentiate between tech platforms, with certain sector-specific tools gaining credibility. One user said that daily usage of AI saved them time; however, another said that they were careful with the data that they fed into their AI tools.

Fragmentation over uniform progress

The market appeared to stabilise in September 2025; however, there was fragmentation rather than uniform progress. Satisfied users identified the ease of use, quality of output and consistent performance; while negative feedback arose from greater awareness of security concerns surrounding sensitive data. Users reported that their AI tools were “brilliant and time-saving”, and “work[ed] so well as an AI, better than CoPilot”; however, one said that they opted to avoid “all AI products”.

This data revealed a “mismatch” between product readiness and innovation speed, which has the potential to create lasting market segmentation. It found that platforms that emphasised quality over hype were successful in creating a sustainable user base as opposed to the fast rollout that created “permanent detractors”. This journey reveals a profession that “approaches AI adoption with evidence-based evaluation, rather than just an outright rejection,” Agile concluded.