On Thursday 21st May, Nexian hosted the fifth session of the Legal Modern Workplace webinar series. Matt Newton was joined by Ziyaad Ahmed, Co-Founder of legal AI platform Qanooni, and Sarah Blair, Director of Technology and Transformation at Thorntons Law, to tackle the strategic question every law firm leader is currently trying to answer: where do general-purpose AI tools like Claude and Microsoft Copilot fit, and where do specialist legal AI platforms earn their place?
Here are the key takeaways:
- Most law firm leaders are not short on enthusiasm for AI. They are paralysed by the pace, the noise, and the fear of buying the wrong thing. The honest emotional state of the market right now is excitement with a side of paralysis.
- The single most useful question any firm can ask an AI vendor is whether their tool becomes more valuable or less valuable as the underlying foundation models improve. If the honest answer is "less", that vendor is on borrowed time.
- A five-level value stack offers a practical way to assess any AI tool: from raw model access at the bottom, through workflows, data context, and systems of record at the top. Most tools sold to law firms in the last two years sit at Levels 2 and 3, exactly where the market is squeezing hardest.
- The "build vs buy" framing is now out of date. The combination of frontier models, MCP, connectors, and improving practice management platforms means the real question is “stitch". How the pieces fit together inside your firm.
- Claude and Copilot have moved up the stack faster than anyone expected. They are now genuinely capable of practice-of-law work, not just business-of-law tasks, when used with the right governance and context.
- Specialist legal AI earns its place when it lives inside your systems of record and compounds your firm's matter knowledge over time. The durability comes from integration, not from the cleverness of the underlying model.
- For firms wondering where to start, the order is consistent: workflows first, governance second, tools last. And literacy is built like muscle memory, not in a single training session.
On Thursday 21st May, we hosted the fifth session of our Legal Modern Workplace webinar series. After four sessions focused on the foundations of a modern legal tech estate, this one turned to the topic that has been sitting underneath every previous conversation: AI.
Matt Newton was joined by two guests who bring complementary perspectives on the same question. Ziyaad Ahmed is the Co-Founder of Qanooni, a UK-based specialist legal AI platform built by lawyers, for lawyers, and now in active deployment across UK law firms. Sarah Blair is Director of Technology and Transformation at Thorntons Law, one of Scotland's leading firms. She has spent over two decades in legal IT, from service desk through to transformation leadership, and is currently on a career break that has given her the rare luxury of time to look hard at where the legal AI market actually is.
Between them, they covered why most law firms are stuck, how to think about the AI tooling decision without being swept along by hype, and what a balanced AI strategy looks like across general-purpose and specialist tools.
What follows is a longer-than-usual recap, because this was a longer-than-usual conversation. There is more substance here to revisit than in any session we have run so far.
Where law firms actually are: excitement with a side of paralysis
The session opened with an honest diagnosis. The people who spend their time on LinkedIn talking about AI live in a bubble, and the bubble is genuinely exciting. New tools every week, breakthroughs every month, and an unending sense that the ground is shifting beneath everyone's feet.
Outside that bubble, in the actual offices of actual law firms, the picture is very different.
"There is definitely FOMO coming from within the firm, from fee earners, from partners, from leadership. We have to do something about AI, so let's go and buy something. Nobody has the time for any of this, and it's taking more and more time. The dominant mood I'm hearing is: stop telling me about the tools and tell me what I should do next."
— Sarah Blair, Director of Technology and Transformation, Thorntons Law
Ziyaad recognised the same pattern from the vendor side. Firms feeling pressure from within, hearing about new tools every day, worrying that if they wait six months something better will arrive, and worrying just as much that if they buy now they will have backed the wrong horse.
His advice, and the through-line for the entire hour, was to stop thinking of AI as something to be bolted onto the side of your firm. The systems of record already exist. The matters already live somewhere. The practice management system, the document store, the Microsoft 365 environment, these are the places AI needs to operate, not separate destinations with their own logins and their own learning curves.
The one question worth asking every AI vendor
Before the framework, before the tooling, before the adoption advice, Ziyaad set out the single question that does more strategic work than any other in this market right now.
"Is anything I'm buying right now going to become more valuable, or less valuable, as these models improve? Because the reality is that these models are just going to continuously improve."
— Ziyaad Ahmed, Co-Founder, Qanooni
If the honest answer is "less valuable", that vendor is on borrowed time. A clever wrapper around a foundation model becomes redundant the moment the foundation model itself can do the same job out of the box, and the last eighteen months have demonstrated this happening over and over again. Tools that wowed law firm leaders in demos in 2024 are now competing with what ChatGPT, Claude, and Copilot can do as standard.
The tools that become more valuable as the models improve are the ones that sit inside your systems, compound your firm's matter knowledge over time, and benefit from every improvement in the underlying model without having to be re-engineered.
It is a deceptively simple test. It cuts through almost every AI pitch a managing partner is currently being given.
The framework: five levels of the AI value stack
To make the assessment concrete, Sarah walked through a framework she has been refining (originally developed by Helen Phan, with versions also published by the team at Iris). It maps any AI tool against a five-level value stack, with capability and durability both rising as you move up.
Level 1: Raw AI capability.
The frontier model itself. ChatGPT in its earliest form, or Claude or Gemini accessed directly. Clever, but no memory, no workflow, no system around it. Question in, answer out.
Level 2: Workflows and templates.
A wrapper that bakes in multi-step processes, playbooks, and prompts. This is where most legal AI tools were operating as recently as a year ago, and where many still sit.
The problem: this is no longer much of a moat. Frontier models are increasingly building these workflows in as standard, which is why Level 2 is the most exposed layer of the stack right now.
Level 3: Data and context.
Tools that bring proprietary data (Westlaw, LexisNexis, Thomson Reuters) or learn from your own working patterns. Copilot, when used inside a mature Microsoft 365 environment, sits here, it has context on your documents, your emails, your meetings. Claude is moving into this territory through projects and skills. The system starts to understand how you work, not just how lawyers in general work.
Level 4: Systems of record.
Tools that live inside your practice management system, your document management system, and your matter data. The AI does not just retrieve information, it compounds. Every new matter teaches the system. Every clause drafted refines the firm's position.
The system can pre-empt: not just answering "how did I draft this clause last time" but "this clause is not how the firm has been drafting since March, you may want to reconsider". This is the layer Qanooni is built for, and where Ziyaad argues the genuine durability of any legal AI investment now lives.
Level 5: Operating layer.
A whole other conversation, as Sarah put it, and one for a future webinar. This is where firms start with the technology and build their service around it, rather than fitting technology to existing services.
Two observations from Sarah closed the framework.
First, the middle of the stack (Levels 2 and 3) is being squeezed. Frontier models are reaching up from below, and systems of record are reaching down from above, with iManage's recent repositioning as one prominent example.
Second, the framework is most useful as a decision tool when you are accountable for outputs. Hallucinations in case law research, for instance, are usually a symptom of using a Level 1 tool for a Level 3 or 4 task. Knowing which level you are operating at is the foundation of accountable AI use.
Where the tools actually sit
With the framework in place, the conversation turned to where the major tools currently fit, and where they are heading.
Claude and Copilot.
Both have moved up the stack faster than most law firm leaders realise. Copilot used to be a pure Level 1 tool with a thin Microsoft 365 context overlay, and was reasonably criticised for being underwhelming. The current generation, particularly Copilot's emerging agentic capabilities and Claude with skills and projects, has pushed both into Level 3, with multi-step workflows and meaningful contextual awareness.
Six months ago, Sarah's advice would have been to use Copilot only for business-of-law tasks. Today, with proper governance, both Claude and Copilot can be used for practice-of-law work as well.
Specialist legal AI (Qanooni, Harvey, Lagora and others).
The market is moving in a clear direction. The early players generally started at Level 2. A separate interface, a vault of documents, riding on the same frontier models everyone else uses. The successful ones are now pushing up the stack toward Level 4, integrating into systems of record and trying to capture the compounding-context advantage.
Ziyaad's position is that Qanooni was built on a Level 4 thesis from day one, three years ago, which is why it runs inside Actionstep and Microsoft 365 rather than as a standalone destination.
Sarah's observation was that the partnership between Qanooni and Actionstep makes the combination "sticky" in a way standalone tools cannot be, because firms change their practice management system roughly once a decade.
The wider point both speakers landed on is that this is no longer a "who wins, who loses" question. It is a "what belongs where" question.
From build vs buy, to stitch
"It's not build or buy anymore. It's stitch."
— Sarah Blair
The strategic shift Sarah named is, in some ways, the single most useful sentence from the whole session. The combination of improving frontier models, the Model Context Protocol (MCP) connecting systems together, and practice management platforms increasingly baking AI into themselves means firms no longer face a binary build-vs-buy choice. The viable strategies now involve combining: a frontier model for some tasks, a system-of-record AI for others, connectors and integrations between them.
This makes the decision more nuanced, but it also makes it less terrifying. You are not betting the firm on a single vendor. You are assembling a stack, and the assembly itself is the strategy.
Governance and compliance: the question to ask vendors
The compliance question came up early in the audience chat and shaped the rest of the conversation. UK law firms operate in a sector where professional confidentiality is paramount, and most AI vendors are US-based with US-based inference.
Both speakers were clear that governance should be treated as an enabler, not a blocker. Three practical principles emerged:
-
Ask vendors what they have anticipated, not what they have caught up to. A vendor that is GDPR-compliant today by design is in a different category from one that is racing to comply. Ziyaad's position on this was uncompromising: build for the regulations that are coming, not the ones that already exist.
-
Understand where the inference happens. Claude is currently doing inference in the US, but is moving to Azure Foundry later in 2026/27, which will be a meaningful shift for UK firms. Microsoft Copilot's enterprise tier already offers zero data retention and standard contractual clauses. The compliance story is genuinely improving, but it is uneven across products and tiers.
-
Use governance to unlock work, not to delay it. Building skills, prompts, playbooks, and the firm's "portable intelligence" can happen now, even while the compliance picture is still moving. When the data residency story improves for a given tool, you can switch on what you have already built.
Adoption: workflows first, governance second, tools last
The hour ended where most AI conversations should begin: with adoption. Sarah's framework is simple and worth committing to memory.
"Workflow first, then governance, then tools. You should be thinking about tools last, not first."
— Sarah Blair
The order matters because most firms do it backwards. They buy a tool, then realise they have to retrofit governance, then realise they have not actually defined the workflows the tool is supposed to serve.
Two other adoption ideas from the session are worth carrying away.
Purpose.
AI will free up time. The question is what you do with the time. Sarah's firm built an AI charter that explicitly committed the freed-up capacity to client care. The example she shared was IKEA, who replaced 8,000 contact centre staff with a bot but retrained them as interior designers, and generated £1.3 billion in additional revenue from the retrained workforce. The point: AI's strategic value comes from what you do with the time you reclaim, not from the cost you cut.
Literacy as muscle memory.
A single training session is not how anyone becomes capable. Sarah's analogy was to a professional athlete: daily training, daily reps, daily pushing yourself a little further. Every time you face a task, ask whether AI could help with it. Over weeks and months, that habit becomes the literacy. There is no shortcut.
Where to start
The session closed with a practical question: for the law firm currently stuck in analysis paralysis, what is the next concrete step?
Ziyaad's answer was a test, not a tool. Pick a live matter. Ask the AI system you are considering where it stands. What is outstanding? What was last agreed? What is coming next?
"Notice what happens then. Do you get a straight answer, or is it something that sounds super generic? If the system cannot answer those questions in a connected way, it is not the right tool, at least not for the legal industry. Precedent is the key IP a lawyer has. Any system you consider needs to know your precedents, the current state of them, and compound with every new precedent that comes in."
— Ziyaad Ahmed
Sarah's closing advice was similar in spirit: don't worry about understanding every acronym. Don't worry about whether the tool you bought last year was a mistake, it doesn't matter, because everything will change again. What matters is building the firm-level capability to keep up: the workflows, the governance, the literacy, the muscle memory. The tools will keep changing. The capability is what compounds.
This was the fifth of seven sessions in the Legal Modern Workplace series. The next session turns to what it takes to actually make any of this work inside a law firm. The internal and external resources, the consultants, the implementation partners, the people who help bridge between the tooling and the day-to-day reality of legal work.
Download the white paper or get in touch with the team if you would like to talk through how your firm should be approaching its AI strategy.
Download the
Whitepaper
nexian.co.uk/legal-modern-workplace-whitepaper
Or get in touch with the team if you would like to talk through where your firm is on its technology journey.
