FeaturesSecurityPricingBlogSign in
Blog/Learning/Agentic AI
Blog/Learning/Agentic AI

Agentic AI vs Legal Tech

Alan Yahya3 min read

Lawyers are increasingly waking up to the utility of agentic harnesses like Claude Code for general purpose legal work.

The thesis for specialised legal tech companies was always simple. Optimise a context gathering strategy, AI prompt, and user experience around a single task (like document redlining). Bonus points if you optimised the underlying model with reinforcement learning or fine-tuning.

Agent harnesses are encroaching on this territory for one simple reason: flexibility. Rather than following a narrow, pre-defined path, an agent runs context assembly and task execution at runtime, leaning on specialised utilities. These utilities, like skills (allowing an agent to execute a pre-defined task consistently) or MCP (allowing an agent to access remote remote data) help a general agent to achieve specialised goals.

Moreover, attention is at a premium for high intensity knowledge work. A lawyer in Claude Code can touch dozens of documents at once, without having to change windows. This is often not the case for legaltech solutions. And, if a legaltech tool only gets you 90% of the way there, requiring manual checks anyway, why not use a general agentic solution to get 86% of the way there with a smoother UX?

However, there are key problems with agentic harnesses. There is more of a learning curve, as the agentic system can use their increased autonomy to fail dramatically. They also take a lightweight approach to the existing relationships in your knowledge base, which increases runtime and the probability of hallucinations.

The end result is that legaltech companies need to provide governance, UX or accuracy that enables at least one of the following:

  • Heightened specificity to a firm's internal requirements, through a deep data integration.
  • Tasks executed accurately enough to require minimal checks, transitioning towards a production line approach.
  • A UX which bakes specific legal processes into the core workflow.

Ultimately, lawyers will use what they feel is the best tool. If legaltech companies aren’t able to beat a meandering general purpose agent to the punch, firms will move away. And that is not a fundamental technology problem, it’s an execution problem.