The AI Strategy Guide logo The AI Strategy Guide at the cusp of AI Subscribe →
Legal AI Platforms in 2026: Harvey, Spellbook, Robin AI, Lexion, Casetext — The Procurement Read — Capabilities illustration

Legal AI Platforms in 2026: Harvey, Spellbook, Robin AI, Lexion, Casetext — The Procurement Read

Tom Prommer · CIO/CTOUpdated 2026-05-3013 min read

The General Counsel I spoke with in February had inherited three legal AI tool subscriptions she did not remember buying. The first was Harvey, signed during a 2024 enthusiasm spike by her predecessor. The second was Spellbook, adopted by the contract-negotiation team because the in-Word workflow was the path of least resistance and the legal-ops lead had championed it. The third was an emerging legal-ops platform whose name kept changing through funding rounds and which was being used by exactly two members of the compliance team for one specific workflow. The annual spend was meaningful, the overlap was real, the governance picture was incomplete, and the CIO had not been in any of the original procurement conversations. The cleanup took six months. The procurement-correct posture, in retrospect, had been to evaluate the category as a stack rather than as three separate purchases — and to have the CIO in the room from the first vendor demo. The General Counsel was now the buyer of AI tooling at a scale her predecessor had not anticipated, and the procurement maturity in the legal department was lagging the maturity of the purchases.

That is the shape of legal AI procurement in 2026 at most large enterprises. The General Counsel, the Chief Compliance Officer, and the Chief Privacy Officer are increasingly the buyers of AI tooling for the operational work that lands on their desks — contract review, EU AI Act compliance review, privacy impact assessment, vendor risk analysis, regulatory submission preparation. Most of this work has a legal-AI tool attached. Most of it was bought without a procurement framework. This page is the operator-voice read on the category: the three archetypes, the named vendors, the fragmentation reality, and the procurement-correct posture for the General Counsel and CIO who are increasingly joint buyers. The broader governance question lives at the governance hub; this page is for the procurement question one level down.

The three archetypes

The legal AI market in 2026 has consolidated into three meaningfully different procurement archetypes. The vendor marketing will conflate them — every legal AI vendor describes itself as a full-stack legal platform — but the operational reality is three different shapes of tool with three different procurement decisions.

Archetype one: contract-lifecycle AI. Spellbook, Robin AI, Lexion. The procurement archetype: AI tooling for the contract workflow — drafting, redlining, playbook enforcement, clause extraction, negotiation tracking, contract-database querying. The buyers are the contract-negotiation team, the legal-ops lead, the commercial-legal function. The deployment surface is typically in-Word or in-document-management-system, where the work already happens. The cost structure is per-seat or per-contract, at materially smaller absolute cost than the legal-research archetype.

Archetype two: legal-research AI. Harvey, Casetext (now part of Thomson Reuters and integrated into Westlaw), the AI features in the incumbent research providers (LexisNexis, Bloomberg Law). The procurement archetype: AI tooling for the legal-research, case-law-analysis, deal-related-due-diligence, and complex-question-answering workflow. The buyers are the litigation partners, the senior corporate lawyers, the M&A team. The deployment surface is typically a web application or an integrated layer in the research platform. The cost structure is per-seat at higher tiers than the contract-lifecycle archetype, reflecting the model and infrastructure cost of frontier-model legal research.

Archetype three: full-stack legal-ops AI. This archetype is more aspirational than realised in 2026. Several incumbents — Ironclad, ContractPodAi, the broader contract-lifecycle-management vendors who have added AI features — have tried to position themselves as full-stack legal-ops platforms with AI as one capability. The procurement archetype is the legal-department-wide platform that covers the contract, the matter, the workflow, and the AI layer. The realised state in 2026 is that the full-stack pitch is real but the AI depth on the specialised workflows (research depth, contract-redlining quality) is materially behind the best-of-breed point tools in archetypes one and two. The procurement-correct read for most enterprises: the full-stack platform is a defensible choice for the matter-management and CLM workflow, but the AI capability is unlikely to match the point-tool depth and most enterprises will end up adding a point tool alongside.

The three archetypes are not interchangeable. Procurement teams that evaluate them on a flat feature matrix produce ties at the top; procurement teams that decide the archetype first and the vendor second produce clean verdicts.

Contract-lifecycle AI — the most-adopted archetype

This is the archetype with the most production adoption in 2026 across enterprises of all sizes, because the workflow is concrete, the time savings are measurable, and the per-seat cost is small enough that the procurement decision does not require a board.

Spellbook. The strongest in-Word workflow in the category. Spellbook sits inside Word as an add-in, surfaces draft suggestions, redlines, clause comparisons, and playbook enforcement at the point the lawyer is actually drafting. The procurement-deciding property is the integration with the workflow the contract-negotiation team already uses; Spellbook does not ask the team to change tools, it adds capability inside the tool the team already opens. For organisations whose contract work happens in Word, Spellbook is the default contract-lifecycle AI choice and the alternatives are competing on the same workflow.

Robin AI. Strong on contract review, clause extraction, and the contract-database querying workflow. The procurement-deciding property is the depth of the contract-analysis surface — Robin AI is genuinely competitive on the question of “tell me what is in this contract” and on the bulk-contract-review workflow that the legal-ops team often owns. The trade-off versus Spellbook is workflow surface; Robin’s primary interface is its own application rather than a Word add-in, which is the right choice for the contract-review workflow and a less natural fit for the in-the-flow drafting workflow.

Lexion. Acquired by DocuSign in 2024, now integrated into the DocuSign IAM (Intelligent Agreement Management) surface. The procurement-deciding property has become the integration with the broader DocuSign agreement-lifecycle estate; for organisations already standardised on DocuSign for signature and CLM, Lexion within IAM is the natural extension. For organisations not in the DocuSign estate, Lexion is competitive but the integration argument is weaker.

The procurement read on contract-lifecycle AI. For in-the-flow drafting and negotiation work where the team lives in Word, Spellbook is the default. For bulk contract review and contract-database querying where the team works in a dedicated application, Robin AI is the default. For organisations already standardised on DocuSign, Lexion is the natural extension. The three are competing on workflow shape, not on raw AI capability — the model layer underneath is comparable across the three, and the procurement-deciding axis is which workflow surface fits the team’s existing practice.

Legal-research AI — Harvey and the incumbents

This is the archetype where Harvey has the strongest brand and the incumbents (Thomson Reuters via Casetext and Westlaw, LexisNexis with Lexis+ AI, Bloomberg Law) have the strongest distribution. The procurement reality in 2026 is more contested than the Harvey narrative implies.

Harvey. The strongest brand in the category and a credible product for the workflows it targets — large-firm legal research, deal-related due diligence, complex contract analysis that benefits from a frontier-model layer with deep legal-domain context. The procurement-deciding property is the depth on the specific workflows Harvey has built for; the trade-off is that Harvey is a SaaS-first product with a managed-cloud posture, and the governance review for organisations with data-residency requirements is non-trivial. Harvey has invested in the enterprise-readiness surface (SOC 2, data-handling controls, audit logs) but the deployment remains SaaS-only — not private-cloud, not air-gapped, not on-prem — which creates friction for organisations with strict data-residency obligations. For large-firm clients and for enterprises whose governance accepts the cloud-managed posture, Harvey is the procurement-correct default for the workflows it targets.

Casetext (CARA, now Thomson Reuters / Westlaw). Acquired by Thomson Reuters in 2023 and integrated into the Westlaw platform. The procurement-deciding property has become the integration with the Westlaw research surface for organisations already standardised on Westlaw; Casetext as a standalone product is less prominent now than the Westlaw AI features that incorporate its capability. For organisations with existing Westlaw subscriptions, the Westlaw AI features (including the Casetext-derived workflows) are the natural extension and the procurement decision is whether the AI tier is worth the incremental seat cost. For organisations on LexisNexis or Bloomberg Law, the equivalent incumbent-AI features (Lexis+ AI, Bloomberg Law’s AI assist) are the parallel decision.

Incumbents’ AI features. Lexis+ AI from LexisNexis, the AI features integrated into Bloomberg Law, the Westlaw AI surface. The procurement-deciding property is the integration with the research platform the legal team already uses; the AI capability is competitive with Harvey on most legal-research workflows for the platforms’ native content surface. The trade-off is that the AI features are tied to the platform’s content licensing and pricing, which can be substantial.

The procurement read on legal-research AI. For large firms and complex enterprise legal work where the frontier-model depth justifies the premium and the cloud-managed posture is acceptable, Harvey is the default. For enterprises already standardised on Westlaw, LexisNexis, or Bloomberg Law, the incumbent AI features are the natural extension and the procurement decision is incremental rather than category-switching. The procurement-correct posture for most enterprises is to evaluate Harvey against the incumbent the team already uses, not in isolation — the integration with the existing research workflow is usually the deciding axis.

The fragmentation reality

The procurement reality that most enterprises confront in 2026 is that the legal AI market is more fragmented than the marketing implies, and the consolidation that would make the procurement decision simpler is not happening on the timeline the optimistic narrative suggests.

The mechanism is structural. The vertical depth required to win a legal workflow is hard to build broadly. Spellbook has invested in the in-Word drafting workflow and is genuinely strong on it; expanding that depth to the bulk contract review workflow that Robin AI owns is non-trivial. Robin AI has invested in the contract-analysis surface and is genuinely strong on it; expanding to the in-the-flow drafting workflow is non-trivial. Harvey has invested in the frontier-model legal-research surface and is genuinely strong on it; expanding to the in-Word contract redlining workflow is non-trivial. Each tool has bet on a specific workflow shape and the procurement reality is that no single platform covers all of them with comparable depth.

The enterprise consequence. Most large enterprises in 2026 end up with two or three legal AI tools — a contract-lifecycle tool for the high-volume contract work, a legal-research tool for the complex research and analysis work, and sometimes a workflow-specific tool for high-volume or specialised workflows (immigration, IP, regulatory submission). The procurement-correct posture is to accept this reality rather than to pursue a single-platform consolidation that the market is not ready to support.

The honest read on Harvey specifically. Harvey has the strongest brand in the category and a real product for the workflows it targets, but the brand has produced an expectation of full-stack coverage that the product is not yet designed to deliver. Harvey is the procurement-correct choice for large-firm legal research and for specific enterprise workflows where the frontier-model depth pays back; it is not the procurement-correct choice for in-Word contract drafting or for bulk contract-database querying. Procurement teams that bought Harvey expecting full-stack coverage are the most common source of legal AI procurement disappointment in 2026, and the disappointment is a marketing problem, not a product problem.

The General Counsel and CIO joint procurement

The single most important procurement-process change for legal AI in 2026 is moving from General-Counsel-only procurement to joint General-Counsel-and-CIO procurement. The mechanism is dull and recognisable.

The legal-domain depth and the workflow knowledge sit in the General Counsel’s organisation. The General Counsel knows which workflows matter, which tools fit which workflows, which integrations the legal team will actually use, and which procurement criteria are deciding from the legal-work perspective. None of this knowledge sits with the CIO.

The data-residency, security, integration, and AI-governance considerations sit with the CIO and CISO. Where the data goes, how it is encrypted, who has access, how it integrates with the identity surface, how it complies with the EU AI Act for high-risk system obligations, how it interacts with the broader AI-governance framework — none of this knowledge sits with the General Counsel.

Legal AI tooling that lands as a General-Counsel-only purchase consistently underperforms because the integration with the enterprise data estate, the identity and access surface, and the AI-governance review is under-built. The contract-lifecycle tool that does not integrate with the CLM the legal-ops team has already standardised on does not get used. The legal-research tool that the CISO has not signed off on for data-handling does not get used. The full-stack platform that the IT organisation cannot connect to the identity surface does not get used.

Legal AI tooling that lands as a joint General-Counsel-and-CIO purchase consistently outperforms because both the workflow knowledge and the technical-and-governance constraints are represented at decision time. The procurement-correct posture in 2026 is to treat legal AI as a joint purchase from the first vendor demo, with the General Counsel as the deciding voice on workflow fit and the CIO as the deciding voice on integration, governance, and data-handling.

What I would procure in 2026, by enterprise shape

A pragmatic short list, scoped to the realistic starting positions of an enterprise legal AI programme.

Mid-sized enterprise, high contract volume, mature legal-ops function. Spellbook for in-Word drafting, Robin AI for bulk contract review and database querying, the existing legal-research platform’s AI features (Lexis+ AI or Westlaw AI) for research. Three tools, three workflows, joint General-Counsel-and-CIO procurement with formalised data-handling review.

Large enterprise, mixed legal workload including complex corporate work. Add Harvey to the stack for the large-firm-style research and due-diligence workflows. Keep the contract-lifecycle tools separate. Treat Harvey as the legal-research-and-deal-diligence tier and the contract-lifecycle tools as the volume tier.

DocuSign-standardised enterprise. Lexion within DocuSign IAM as the contract-lifecycle default, the existing research platform’s AI features for research. The integration argument is the strongest in the category for organisations already in the DocuSign estate.

Enterprise with strict data-residency or self-host requirements. The category is harder. Most leading legal AI tools are SaaS-first with limited self-host options. The procurement-correct posture is to evaluate the incumbents’ AI features (which inherit the incumbents’ enterprise-deployment options) over the SaaS-first specialists, and to defer the most ambitious legal AI workflows until the self-host options mature. This is a real constraint in 2026 and the market response is incomplete.

The honest signal of a working legal AI procurement is that each workflow has the tool that fits its shape and the General Counsel and CIO both endorsed each purchase. The signal of a failing one is that the tools accumulated by enthusiasm in the legal department without the CIO in the room, and twelve months in the integration with the broader enterprise governance and IT estate is the bottleneck nobody planned for.

None of this is sponsored, none of the vendors named pay for inclusion, and the joint-procurement framework for General Counsel and CIO is published under CC-BY-4.0 alongside the governance hub. The legal AI market in 2026 is fragmented, brand-loud, and procurement-immature; the procurement-correct posture is the one that accepts the fragmentation honestly and treats every purchase as a joint decision with the technical and governance side.

Sources

Methodology: archetype and procurement-process analysis drawn from fractional CIO and CTO advisory engagements (2023–2026) on enterprise AI programmes where legal AI was a meaningful procurement surface, cross-checked against published vendor architectures and the realised legal-team adoption data the operating organisations shared on the condition of anonymity. The joint-procurement recommendation reflects realised outcomes from engagements where both procurement models were trialled.

Frequently asked questions

Is Harvey actually the best legal AI in 2026?
Harvey has the strongest brand in the category and a credible product for the workflows it targets — large-firm legal research, deal-related due diligence, and complex contract-analysis tasks that benefit from a frontier-model layer with deep legal context. It is not the best fit for every legal-AI workload. For contract-lifecycle work specifically — drafting, redlining, playbook enforcement, negotiation — Spellbook and Robin AI both compete credibly and often win on workflow fit. The procurement-correct read is that Harvey owns the legal-research and large-firm-workflow segment and the contract-lifecycle segment is more fragmented, with three or four credible alternatives depending on the specific workflow.
How does this fit with the AI governance hub?
Legal AI tooling sits at the operational surface of work that the governance hub covers at the policy layer. EU AI Act compliance reviews, privacy impact assessments, vendor-contract risk analysis, model-card review — these are increasingly the day-to-day work of the General Counsel, the Chief Privacy Officer, and the Chief Compliance Officer, and legal AI tooling is the operational substrate that work runs on. The procurement-correct read is to evaluate legal AI tools as part of the governance capability stack, not as a standalone legal-department purchase, because the workflows they enable are increasingly governance workflows.
Why do most enterprises end up with two or three legal AI tools?
Because the vertical depth required for each legal workflow is hard to consolidate across a single platform. The model and the workflow design that work well for contract redlining are different from the model and workflow that work well for case-law research, which are different from the surfaces that work for due diligence or compliance review. Each tool in this category has bet on a specific workflow shape and the procurement reality is that no single platform covers all of them with comparable depth. Most enterprises end up with a contract-lifecycle tool (Spellbook, Robin AI, or Lexion), a legal-research tool (Harvey, Casetext, or the incumbent research providers' AI features), and sometimes a workflow-specific tool for high-volume or regulated workflows.
Should the General Counsel or the CIO buy legal AI?
The General Counsel buys, but the CIO has to be in the procurement conversation. The legal-domain depth and the workflow knowledge sit in the General Counsel's organisation; the data-residency, security, integration, and AI-governance considerations sit with the CIO and CISO. Legal AI tooling that lands as a General-Counsel-only purchase consistently underperforms because the integration with the enterprise data estate, the identity and access surface, and the governance review process is under-built. Legal AI tooling that lands as a joint General-Counsel-and-CIO purchase consistently outperforms because the workflow knowledge and the technical and governance constraints are both represented at decision time.