AI law firm market map
AI law firms, legal-tech vendors and adoption partners: three operating models, one decision
The direct answer: “AI law firm” can describe three different purchases
The same phrase is being used for businesses with materially different responsibilities. In the first model, you buy legal work from a law firm that uses AI in delivering it. In the second, your firm buys software and remains responsible for using it in legal work. In the third, your firm buys help changing its workflows, controls and evidence around technology. These are not three versions of one product. They allocate the regulated service, client relationship, implementation burden and risk differently.
That makes the choice an operating-model decision before it is a feature comparison. Start with the outcome: do you want another provider to deliver a legal service, a tool for your own people to operate, or support redesigning how your existing firm works? A firm may use more than one model, but each engagement should still be identified correctly. A software licence does not become legal representation because its interface discusses law. A consultancy does not become a regulated law firm because lawyers work there. A regulated provider does not become a conventional software vendor merely because software is central to its service.
This market map stays deliberately narrow. For the inward process of adopting AI, use the legal AI adoption guide. For assessing organisational maturity, see the law-firm AI readiness maturity model. Here, the question is what kind of business relationship the firm is entering.
Model 1: the AI-native or AI-driven law firm: you buy legal work
In this model, the organisation delivering the matter is a law firm or works through identified regulated legal entities. The client buys legal services. Technology may shape intake, production, review or price, but the central transaction is still legal work. The engagement letter, regulator record and identity of the responsible legal provider matter more than the label “AI-native”.
Garfield.Law Limited is a clear UK example. The SRA described it as the first purely AI-based law firm it authorised and said it was designed for small-business debt recovery. The regulator’s ABS record identifies Garfield.Law Limited under SRA number 8010904. Those are regulator-sourced facts, not performance claims: SRA release, accessed 15 September 2026 and SRA ABS record, accessed 15 September 2026.
Lawhive Legal Ltd appears on the SRA register under number 8003766. Lawhive’s own UK about page describes regulated solicitors and fixed-fee services. That service and pricing description is company-stated; the regulatory identity is independently checkable on the regulator’s record: SRA organisation record, accessed 15 September 2026 and Lawhive about page, accessed 15 September 2026. In the United States, Lawhive’s own terms say legal services are provided through Lawhive Legal US LLC and co-counsel relationships. That structure is company-stated in its US terms, accessed 15 September 2026.
Other entrants use similar category language while describing different practices. General Legal calls itself an AI-native law firm; Crosby describes itself as an AI-powered or agentic law firm; and Norm describes Norm Law as an AI-native law firm. Each is a company-stated description from its own site, not an independent assessment of capability or results: General Legal, accessed 15 September 2026, Crosby, accessed 15 September 2026, and Norm Law, accessed 15 September 2026.
The practical point is not that these firms are interchangeable. It is the opposite. Check the exact entity, jurisdiction, authorised activities, engagement terms and people responsible for the matter. When you instruct a law firm, you are a client buying legal work. The provider’s use of AI may influence delivery, but it does not remove the need to identify who owes which professional duties.
Model 2: the legal-tech vendor: you buy software and remain the lawyer
A legal-tech vendor supplies a product for your firm to operate. The software may assist drafting, review, research, matter management or another task, but your firm keeps the client relationship and remains responsible for the legal service it provides. The relevant commercial documents are usually a licence or subscription, data-processing terms, service levels and product documentation rather than a legal-services engagement letter.
For this model, assess the licence, data terms and allocation of responsibility rather than treating the interface as legal representation. Our legal-tech profiles keep individual products separate. The legal AI tool evaluation guide covers the vendor-assessment question; this article does not rank or re-review products.
A tool may be impressive without being suitable for a particular matter, team or data class. Conversely, a narrow tool can be useful when its role and limits are explicit. The neutral test is whether the licence, data terms, workflow and human review collectively support the result your firm needs. Buying software transfers capability into the firm; it does not transfer the firm’s professional responsibility to the interface.
Model 3: the adoption partner: you keep the firm and change how work is done
An adoption partner helps a firm decide how technology should fit into existing work. The deliverable might include a workflow map, governance decisions, evaluation criteria, controlled testing, training or an evidence record for a go/no-go decision. The firm normally keeps its client relationship, technology accounts and responsibility for legal work. The partner’s value therefore depends on implementation discipline and clarity about boundaries, not on pretending to be either the software provider or the regulated legal provider.
This is Margo Legal’s intended lane. Margo is in controlled development, is not a law firm and does not provide legal representation. It is working on contextual legal AI and workflow support. Nothing in this article presents Margo as equivalent to a live legal-service provider or a released software product.
Ask what the partner will deliver, which systems it will access, who makes the final decision and what is handed back when the engagement ends. Legal judgment stays with the firm’s qualified people.
A neutral decision table
| Operating model | What you buy | Regulated provider | Client relationship | Typical pricing shape | Evidence to request |
|---|---|---|---|---|---|
| AI-native or AI-driven law firm | Legal work on a matter | The identified legal entity, subject to the stated jurisdiction and authorisation | The provider holds the legal-services relationship for the work it accepts | Matter, scope, fixed-fee or other engagement terms | Regulator record, engagement letter, scope, responsible people and review route |
| Legal-tech vendor | A licence or subscription to software | Your firm remains the regulated provider to its clients; the vendor is not made one by the licence | Your firm keeps its client relationship | Subscription, seat, usage or enterprise licence | Product terms, data flows, retention, security, testing, export and termination evidence |
| Adoption partner | Implementation, governance, workflow or evaluation support | Your firm remains responsible for the legal service unless a separate regulated provider is expressly engaged | Your firm keeps its client relationship | Project, workshop, retainer or defined deliverable | Scope, access boundaries, decision rights, deliverables, acceptance criteria and handover record |
These are patterns, not substitutes for reading the actual contract. A business can operate across models or use partners, affiliates and co-counsel. Ask which entity is doing each part of the work rather than inferring the answer from branding.
Six questions to ask before signing anything
1. Which exact entity is providing what?
Record the legal name, jurisdiction and role of every relevant entity. If legal services are included, verify the regulator record and authorised scope yourself. If the offer is software or consultancy, make sure the contract does not blur that boundary.
2. Who owns the client relationship and legal judgment?
Name the person or firm accountable to the client. Ask who reviews outputs, who can reject them and who signs off before work is sent, filed or relied upon. “Human in the loop” is too vague unless the human, check and consequence are defined.
3. Where does information go, and how long does it stay there?
Map collection, hosting, subprocessors, model providers, support access, logs, backups, retention and deletion. Separate public marketing statements from binding contractual terms. Do not put confidential material into a system merely to discover how it behaves.
4. Which claims are independently verified and which are company-stated?
Regulatory status can often be checked with the regulator. Product capability and service descriptions may come only from the provider. Label the difference, ask for evidence tied to your use case and avoid turning a provider’s positioning into an independent conclusion.
5. Are the important terms public, private or still to be agreed?
A public webpage can explain an offer without containing the controlling terms. Ask for the documents governing price, scope, service levels, data, liability, termination and change control. Record unresolved items before approval rather than relying on a sales conversation.
6. What evidence will support the decision after a trial or engagement?
Agree what will be observed, who will assess it and what would stop or expand use. Evidence might include error findings, review effort, data-handling confirmation and completion against defined acceptance criteria. For a concrete legal-work example, the AI contract review guide shows why workflow boundaries and verification matter.
Where the category is going
The cited company and regulator sources show several organisations describing themselves as AI-native, AI-powered or agentic law firms, alongside regulated records that identify particular providers. That is enough to expect the label to remain broad. It is not enough to claim that every entrant has the same regulatory structure, technology, client base or evidence of performance.
For buyers, the recurring task is classification. New interfaces and business combinations may change what an offer looks like, but the underlying questions remain stable: are you instructing a legal provider, licensing a tool, or engaging help to change your own operation? Which entity holds the client relationship? Who reviews the work? What do the binding terms and retained evidence actually show?
If your firm is deciding which of these three relationships it needs, you can book a fit call. We will classify the operating problem before discussing an engagement, not to treat three different models as a product shortlist.
Source note: Sources were checked on 15 September 2026. Company descriptions above are expressly labelled company-stated; regulator facts link to the SRA. No named company is endorsed or rated here.
Lawhive and LegalOS are Lawhive marks. Margo Legal is not affiliated with any company named here.