AI is evolving quickly. For law firms still working out how to get consistent value from the tools they already have, the conversation has moved on again. Now we’re talking about agentic AI.
There is real opportunity here. But the first productivity wave is unfinished. Giving people access to AI and embedding it successfully across a firm are very different things.
Before we move further down this pathway, we need to understand what is changing, where the value might live and whether our firms have the foundations to support it.
From task to outcome
Most people are familiar with conversational AI. You ask it to summarise a matter file, review its response and ask for changes. You remain central to the work.
Agentic AI changes that unit of work from a task to an outcome. An agent is an AI system that can plan steps, use tools and work across connected systems to deliver something you have asked for.
Take a simple example such as: “Prepare tomorrow’s client briefing.”
You specify the client, matter, deadline and template. The agent builds a plan, gathers relevant information from the sources it is allowed to access and prepares the briefing. Much of that activity can happen in the background while you do something else.
At agreed checkpoints, it comes back for your review or approval. Before an email is sent or a record changed, you decide whether it should proceed.
That is delegated work. The opportunity is to reduce repetitive activity and give people more time for work that needs their expertise. The challenge is deciding what you are comfortable delegating and where human oversight must remain.
A standing mandate changes the question again
There is another step beyond asking an agent to prepare one briefing: asking it to “keep the client briefed on this matter”.
That is a standing mandate, a continuing objective that allows an agent to act when defined events occur, without a fresh instruction each time.
It raises further questions. What should trigger action? When must the agent stop? Which decisions require approval? How will someone review what it has done?
Microsoft’s Cowork example that we are discussing in our webinar series helps bring this shift to life. But whichever platform a firm uses, the underlying questions about delegation, boundaries and supervision remain.
Start where the value is checkable
The first opportunity may sit in business services. HR starter preparation, finance reporting or marketing event administration can all offer useful starting points, alongside carefully selected legal workflows.
I would apply five tests, as follows, to any proposed use: repeatable, checkable, supervised, bounded and worth it.
Ask yourself:
- Does the work happen often enough to justify designing the process?
- Can someone establish whether the result is right?
- Is there a named owner with time to review it?
- Is access limited to the information needed?
- Does the value exceed the cost?
All five need to hold. Otherwise, the time unlocked at the beginning can disappear into checking and correcting the output.
Nor should we hand an entire process to one agent. Use clear rules where consistency matters, AI where summarising or drafting helps, checks to identify exceptions and a person to approve the outcome. Documenting what good looks like makes supervision easier.
Understand what an outcome costs
At Stridon, we recently used an example where an AI agent could prepare a client briefing for around £2–£5. The actual cost will vary depending on how much information it needs to review and how complicated the work is.
But the cost of using AI is only part of the picture. You also need to consider how much time someone spends checking the briefing and making corrections.
If a solicitor spends longer checking an AI-generated briefing than they would have spent writing it themselves, there is little benefit. If it saves time and helps them respond to clients more quickly, it could be worthwhile.
Start with a clear budget and spending limit. Then check whether the work is accurate, how much correcting it needs and whether it saves time overall. This will help you understand where AI is delivering value for your firm.
Build supervision into the design
Human supervision needs to be part of the process from the start.
Every AI agent should have someone responsible for its work. Set clear rules about what information it can access, what it can do on its own and when it must ask for approval. Keep a record of its actions, review its work regularly and set a spending limit.
It is also important to check who can access your firm’s information. If people already have access to confidential documents they should not see, an AI agent could use those documents too. AI can make an existing problem bigger.
Before giving an agent access to your systems, check that confidential information is properly protected and that its use follows your clients’ instructions. You also need to know which AI provider will handle the information and where it will be processed.
Prepare before you scale
There are seven areas I would encourage firms to focus on: protecting information, choosing suitable tasks, having people available to check the work, helping leaders understand how AI works, measuring the benefits and developing staff skills.
These all work together. People checking AI’s work need the time and skills to do it properly. Leaders need to understand what they are asking AI to do and what it will cost. Junior colleagues still need opportunities to learn and develop their judgement, even as AI takes on some of the work they would traditionally do.
Start with one clearly defined process. Decide what a good result looks like, then check whether AI delivers it before expanding its use.
For legal leaders, the key question is whether their firm is ready to manage and check the work it hands over to AI. Getting this right will help turn the technology’s potential into benefits the firm can see and measure.
Join our upcoming webinar series for UK law firms, where we’ll explore how law firms can maximise value from existing AI investments, build confidence across their teams, strengthen governance and prepare for emerging technologies such as Agentic AI. Register your interest here > https://go.stridon.co.uk/autumn-2026-webinar.