Legal teams hit 87% genAI use as agentic workflows force governance-by-contract
AI addenda in vendor agreements are emerging as the control layer as legal agents move into production.
Major law firms and corporate legal departments are running AI agents in production for contract review, regulatory monitoring, and first-pass discovery. With adoption now reported at 87% in 2026, general counsel are shifting from pilot decisions to contract-based governance and vendor controls.
Key Takeaways
- AI agents are being used in production at major law firms and corporate legal departments for contract review, regulatory monitoring, and first-pass discovery.
- A cited FTI Consulting and Relativity study put 2026 generative AI usage in legal teams at 87%, up from 44% in 2025, while formal technology roadmaps rose to 53% from 25%.
- Summarization (83%) and contract clause identification (63%) led reported use cases, alongside e-discovery and document review, contract drafting, and first-pass review.
- Microsoft’s Legal Agent and Harvey’s custom agent-building tooling were highlighted as examples of autonomous legal workflows moving toward mainstream deployment.
From GenAI Pilots to Agentic AI in Legal Production
Legal departments are no longer treating AI as a sandboxed productivity tool. The operational change is that agentic AI systems, meaning systems that can take actions toward goals with less continuous human direction, are now being used in production environments at major law firms and corporate legal departments for reviewing contracts, monitoring regulatory changes, and handling first-pass discovery.
That matters because it reframes the general counsel’s job. The report frames the core question as shifting from whether to deploy AI to how to govern autonomous systems already running. Generative AI, meaning models that create new text like summaries or draft clauses, sped up experimentation. Agentic AI speeds up deployment, which is where legal risk stops being hypothetical and starts being an incident-response problem.
The autonomy is the feature and the failure mode. A system that can draft edits, flag clauses, or route work without waiting for a human prompt can also make decisions faster than oversight loops can catch, creating exposure to legal noncompliance and third-party harms.
Governance-by-Contract: AI Addenda Become the Control Layer
The governance response described in the report is not a new committee or a new dashboard. It is contracting. AI addenda in vendor agreements are described as becoming a primary tool for protecting organizations from AI vendor overreach, specifically by spelling out how responsibility is allocated when an autonomous system makes a consequential mistake.
Mechanically, these addenda function like a control layer that sits above the model and the workflow. They aim to define what the system is allowed to do, what counts as a breach, and who pays when something goes wrong. The report points to coverage areas that include acts and omissions, violations of law, and liability questions tied to autonomous behavior.
This is where “agentic” becomes procurement friction. If a vendor is selling an agent that can take actions inside a legal workflow, the buyer’s legal team is now negotiating the boundary conditions of that autonomy. The report frames contracting for these protections as a core general counsel competency, which is another way of saying the bottleneck is shifting from model capability to enforceable terms.
One important gap in the packet is scope. The report describes AI addenda as becoming primary protection tools, but it does not quantify how standardized these clauses are across the broader enterprise market, or whether they are converging on common language.
Vendor Signals: Microsoft Legal Agent and Harvey’s Custom Agent Push
The vendor landscape in the report reads like convergence: Big Tech and specialists are both pushing toward autonomous legal work as the wedge. Microsoft’s Legal Agent is cited as representative of tools reaching general counsel, with described capabilities including analyzing documents, drafting edits, and reviewing contracts.
On the specialist side, Harvey is positioned as building an “agent layer” for law firms. The report says Harvey CEO Winston Weinberg unveiled a tool that allows law firms to build custom AI agents tailored to specific practice areas. Harvey is described as serving more than 700 customers across 58 countries, with adoption by 70% of AmLaw 10 firms and nearly 50% of AmLaw 100 firms, rankings often used as a proxy for top-tier firm penetration.
The commercial signal is that this is being funded and staffed like a production rollout, not a demo cycle. Harvey raised $200 million at an $11 billion valuation in March 2026, with funds directed toward expanding AI agents and growing legal engineering teams embedded with customers. Embedded teams are the tell here. They usually show up when the product is being integrated into real workflows with real constraints, not when it is being trialed in isolation.
Adoption data in the report reinforces that shift. A cited FTI Consulting and Relativity report, based on interviews with 30 general counsel and a survey of 224 chief legal officers, found 87% of legal teams report generative AI use in 2026 versus 44% in 2025. Formalized technology roadmaps rose to 53% from 25% the prior year, which is the kind of process signal that typically appears when a function is institutionalizing spend and governance.
Use cases skew toward high-volume, high-leverage tasks. The most common were summarization (83%) and contract clause identification (63%), alongside e-discovery, meaning the process of collecting and reviewing electronically stored information for legal matters, plus document review, contract drafting, and first-pass review.
The report includes two quotes that capture the tone shift inside legal ops. “Generative AI has become a fixture in the majority of legal departments,” said Sophie Ross, Global CEO of FTI Technology. David Horrigan, Discovery Counsel at Relativity, described GC use rising from 20% in 2023 to 87% “today,” calling it the end of what he termed “the era of the Luddite Lawyer.”
Signals to Watch for Legal AI shifts from pilots to
The next leg of this story is whether governance terms become portable. If AI addenda and agent-governance clauses standardize across enterprise procurement, expect common language to harden around acts and omissions, violations of law, and liability allocation, because that is what makes an “agent” sale scalable across buyers.
Adoption metrics also need to split into two buckets: experimentation versus production. The cited 87% figure is directionally strong, but the packet does not provide a definition of what qualifies as “use,” and it does not break out how many deployments are truly autonomous agents versus prompt-driven copilots.
On the product side, cadence matters more than marketing. Microsoft’s Legal Agent is described with specific capabilities around document analysis, drafting edits, and contract review. The signal will be whether those capabilities expand in a way that reduces the need for human gating, or whether they remain bounded by review workflows.
Harvey’s next enterprise expansion signals will likely come through its embedded legal engineering teams and the rate of custom agent deployments following its March 2026 $200 million raise at an $11 billion valuation. That is where “custom agents” either become a repeatable product motion or a services-heavy integration story.
My take: Enterprise agent adoption is turning legal into the bottleneck—and that’s investable signal
The threshold that matters is not whether legal teams are “using AI.” The report’s own numbers already answer that, with 87% reported usage in 2026 and 53% reporting a formalized technology roadmap. The inflection is that legal is treating autonomous systems as already-live operational risk, and the enforcement mechanism is shifting to governance-by-contract through AI addenda that allocate liability for acts, omissions, and violations of law.
If AI addenda language starts to standardize and vendors can sell into it without re-litigating responsibility every time, agentic legal workflows start to look like a scalable enterprise category rather than a bespoke integration business. The practical difference is whether autonomy can be deployed at volume without procurement becoming the rate limiter.
