Harvey, Legora, and Thomson Reuters: where the legal AI market is actually going

Three vendors are currently shaping the trajectory of legal AI, and understanding their distinct operating models matters more than comparing feature lists. The conversation inside most law firms and legal operations teams still centres on capability demonstrations: which platform drafts better, which one summarises faster, which one hallucinates less. These are legitimate questions, but they are the wrong starting point for a strategic infrastructure decision. The more consequential question is what each vendor's architecture, commercial model, and institutional positioning implies for the firms and operators that adopt them. Harvey, Legora, and Thomson Reuters are not simply competing on product quality. They are competing on fundamentally different theories of where legal work will be performed, by whom, and under what commercial arrangement. Getting that distinction wrong is expensive.

What the market usually gets wrong

The dominant misconception in legal AI procurement is that the market is a feature race. Buyers attend demonstrations, score outputs against sample documents, and select the platform that performs best on a curated test set. This approach is understandable. It is also structurally misleading.

Feature parity in generative AI compresses quickly. A capability that distinguishes one vendor in one quarter is frequently matched or exceeded by competitors within two or three product cycles. Competing on features alone therefore optimises for a snapshot rather than a trajectory. The more durable differentiators are data access, workflow integration depth, pricing architecture, and the institutional relationships that determine how a vendor will behave when a client's needs diverge from the vendor's commercial interests.

The legal AI market also suffers from a category confusion that conflates general-purpose large language model access with purpose-built legal infrastructure. Several platforms marketed as legal AI are, in operational terms, thin wrappers around foundation models with legal-specific prompting and some document ingestion capability. That is not a criticism of the underlying technology, but it does affect how buyers should think about vendor dependency, data governance, and the realistic ceiling of workflow integration. Purpose-built legal infrastructure requires sustained investment in proprietary data, domain-specific fine-tuning, and deep integration with the systems of record that legal teams actually use. Not every vendor claiming the legal AI label has made that investment.

For a broader orientation to how AI is reshaping legal practice, the Legal AI and Technology pillar sets out the structural context that informs the vendor analysis below.

What actually changes when you look at the operating layer

Harvey entered the market with a clear institutional positioning: a platform built for elite law firms, trained on legal reasoning at a level of sophistication that general-purpose models do not reach by default, and backed by investment from firms with direct interests in its success. That institutional alignment is both its strength and its structural constraint. Harvey's commercial model has historically been oriented towards large firms with the appetite and the budget to deploy AI at the practice-group level rather than the individual-fee-earner level. The implication is that Harvey's roadmap is shaped by the priorities of its most influential clients, which are not necessarily the priorities of mid-market firms, in-house legal teams, or litigation funders evaluating portfolio-level document analysis.

Legora has taken a different approach. Its architecture is oriented towards workflow integration at the matter level, with an emphasis on making AI assistance available across the full lifecycle of a legal matter rather than as a discrete drafting or research tool. The distinction matters operationally. A platform that sits alongside existing workflows requires users to context-switch and manually transfer outputs. A platform that integrates into the matter management layer can, in principle, surface relevant AI assistance at the point of need without requiring the fee earner to initiate a separate process. Whether Legora has fully delivered on that integration promise at scale is a question that prospective buyers should test rigorously, but the architectural ambition is directionally correct and commercially significant.

Thomson Reuters occupies a structurally different position from both. It is not a legal AI startup. It is an incumbent data and information business with decades of investment in legal content, a substantial installed base across law firms and courts, and the commercial infrastructure to bundle AI capability into existing subscription relationships. The acquisition of Casetext and the development of CoCounsel represent Thomson Reuters' attempt to convert its content advantage into an AI advantage. The strategic logic is sound: proprietary legal data is a genuine moat in a market where foundation model access is increasingly commoditised. The execution risk is the one that afflicts most incumbent technology businesses attempting to absorb a disruptive capability: the organisational and commercial incentives that protect existing revenue streams can slow the pace of genuine product transformation.

Understanding these operating-layer distinctions is not an academic exercise. It directly affects which vendor is the right fit for a given firm's workflow, risk profile, and commercial trajectory. The writing archive contains adjacent analysis on how AI procurement decisions interact with professional conduct obligations and data governance requirements, both of which bear directly on vendor selection.

Commercial consequences

The commercial consequences of vendor selection in legal AI are more durable than most buyers appreciate at the point of procurement. Legal AI platforms are not interchangeable utilities. They embed themselves into workflows, training habits, and institutional muscle memory in ways that create genuine switching costs over time. A firm that deploys a platform across its associates for eighteen months and then attempts to migrate to a competitor faces not just a technical migration problem but a retraining problem, a change management problem, and a period of productivity loss that is difficult to quantify in advance but very visible in retrospect.

This switching cost dynamic has a direct implication for how vendors price and structure their commercial relationships. Vendors with a high-quality product and a clear path to workflow embedding have an incentive to price aggressively in the early adoption phase in order to capture market share before switching costs crystallise. Buyers who interpret aggressive early pricing as a signal of long-term value rather than a customer acquisition strategy may find themselves negotiating from a weaker position at renewal.

For litigation funders and legal operations functions, the commercial consequences extend beyond the firm level. Portfolio-level document analysis, due diligence on potential matters, and the monitoring of active litigation all benefit from AI capability, but they require a vendor relationship that can accommodate non-standard use cases, variable volume, and data governance requirements that differ from those of a conventional law firm deployment. The major vendors have not all invested equally in serving these use cases, and buyers from outside the traditional law firm market should probe vendor roadmaps and reference clients with particular care.

There is also a regulatory dimension that is beginning to materialise. Regulators in multiple jurisdictions are developing guidance on the use of AI in legal practice, with particular attention to supervision obligations, disclosure requirements, and the professional responsibility implications of AI-assisted work product. Vendors whose architecture makes it easier to audit AI contributions to legal work product will have a compliance advantage as regulatory expectations become more concrete. This is not yet a decisive factor in most procurement decisions, but it is moving in that direction faster than many buyers have anticipated. The about page sets out the analytical framework that informs this site's approach to regulatory and commercial risk in legal technology.

Where the market is likely to move next

The legal AI market is moving towards consolidation, but not in the way that the term is usually understood. The consolidation that matters is not primarily about vendor mergers and acquisitions, though those will continue. It is about the consolidation of AI capability into the systems of record that legal teams already use: practice management platforms, document management systems, matter management tools, and the data environments that underpin legal operations functions.

The vendors that win the next phase of the market will not necessarily be those with the best standalone AI product. They will be those that achieve the deepest integration into the operational infrastructure of legal practice. That is a distribution and partnership problem as much as it is a technology problem, and it favours vendors with existing relationships in the legal technology ecosystem over pure-play AI entrants, all else being equal.

At the same time, the foundation model layer is becoming more capable and more accessible at a pace that continues to compress the moat available to any vendor whose differentiation rests primarily on model quality rather than data, integration, or workflow design. The vendors most exposed to this dynamic are those that have not invested in proprietary legal data or deep workflow integration, and that are relying on the current generation of model quality as their primary competitive argument.

For buyers, this trajectory has a practical implication: the evaluation criteria that matter most in 2025 are integration depth, data governance architecture, vendor financial stability, and the quality of the vendor's roadmap for regulatory compliance, not the quality of the output on a demonstration document set. Firms that build their vendor selection process around the latter will find themselves revisiting the decision sooner than they expect.

The emergence of agentic legal AI, in which AI systems take sequences of actions across multiple tools and data sources rather than responding to individual prompts, will accelerate this dynamic further. Agentic capability requires deep system integration and robust governance by definition. Vendors that have invested in integration infrastructure will be better positioned to deliver agentic capability safely than those that have not, regardless of the quality of their underlying model.

What this means in practice

The practical synthesis for any firm or legal operations function evaluating the legal AI market is this: treat vendor selection as an infrastructure decision, not a software procurement exercise. The question is not which platform produces the best output today. The question is which vendor's architecture, commercial model, and institutional trajectory aligns with where your practice needs to be in three to five years.

For large law firms with the budget and the appetite for a premium relationship, Harvey's institutional positioning and depth of legal reasoning capability represent a credible choice, provided the firm is clear-eyed about the switching costs and the commercial dynamics of the relationship over time. For firms and legal operations functions that prioritise workflow integration and matter-level AI assistance, Legora's architectural approach is worth serious evaluation, with rigorous testing of integration depth against the firm's actual systems of record. For buyers who value content depth, regulatory familiarity, and the ability to bundle AI capability into an existing vendor relationship, Thomson Reuters' trajectory is the most institutionally legible, even if the pace of product transformation carries execution risk.

None of these is the universally correct answer. The best legal AI platform for a given organisation is the one whose operating model, commercial structure, and integration architecture best fits that organisation's workflow, risk profile, and strategic direction. Arriving at that answer requires a more rigorous evaluation process than most firms currently apply, and it requires buyers to ask harder questions of vendors than most vendors are accustomed to answering.

The firms that get this right will have a durable operational advantage. The firms that treat it as a feature comparison will be back in the market sooner than they planned. For further analysis on the intersection of legal technology, professional conduct, and commercial strategy, the Legal AI and Technology pillar is the right starting point.