Article 4.1 · The Agentic Revolution: When AI Stops Assisting and Starts Doing

T3PS Legal Dynamics · Series 4: Practical AI for Legal Operations · Refreshed June 2026

Published as part of the "Practical AI for Legal Operations" series

There is a moment in most technology transitions when the thing stops being interesting and starts being inevitable. We are at that moment with agentic AI in legal services. And if you are still treating AI as a clever search engine or a faster way to draft a letter, I say this with professional respect: you are already behind.

I have been in legal project management for 37 years. I have watched the fax machine arrive, the voice dictation replace the typing pool, the DMS replace the filing cabinet, and cloud computing replace the server room. Every time, the same conversation: it's interesting, but it won't change the fundamentals. Every time, the same outcome: it absolutely did.

Agentic AI is the biggest shift I have seen in this profession. Not because it is clever. Because it is capable.

What "Agentic" Actually Means

Let me cut through the marketing. Every major legal tech vendor shipped something called "agentic AI" by early 2026. Not all of it is equally impressive, and not all the claims will survive contact with the reality of your firm's messy data and half-built processes. But the underlying shift is genuine and it matters enormously.

Traditional AI — the kind most of us have been using for the past two or three years — is essentially a very well-read assistant. You ask it a question, it gives you an answer. You paste in a document, it summarises the document. You write a prompt, it generates a draft. One task at a time, in direct response to your instruction. Think of it as the junior associate who can process enormous amounts of information quickly but needs to be told what to do next at every step.

Agentic AI is different in kind, not just degree. An agentic system receives a high-level objective, breaks it into a sequence of tasks, executes those tasks using multiple tools and data sources, monitors its own progress, and delivers a finished work product — all without you having to hold its hand through each individual step. Think of it as the experienced paralegal who takes a brief, runs the whole project, flags the issues that need a solicitor's eye, and lands it on your desk ready to review.

That is not a trivial distinction. It is the difference between a tool and a colleague.

The Platforms That Are Already Doing This

Thomson Reuters CoCounsel Legal launched in the UK on 26 January 2026, and the press release language was deliberately chosen: this is "agentic AI that helps UK law firms and legal departments future-proof their practices". The platform now serves over one million users and ships with more than 300 pre-built workflows covering litigation, transactional work, and regulatory analysis.

The detail that caught my attention is Tabular Analysis. CoCounsel can now accept up to 10,000 documents in a single workflow, processing them in parallel and presenting the results in a dynamic, filterable table format. For anyone who has ever spent three weeks on a document review exercise with a team of six, let that number land. Ten thousand documents. One workflow. The platform also brings Deep Research on both Practical Law and Westlaw Advantage — a UK-first — meaning the AI is not just processing documents you upload, it is simultaneously drawing on the most authoritative legal content available and reasoning across both sources.

The integration angle matters too. CoCounsel is embedded into Microsoft 365, document management systems, and Thomson Reuters HighQ. This is not a standalone portal you log into separately. It lives where the work already happens.

LexisNexis Protégé launched its general availability in February 2026, replacing Lexis+ AI entirely. Sean Fitzpatrick, CEO of LexisNexis Global Legal, described the shift as a response to what legal professionals are "increasingly seeking: integrated legal AI work environments." The platform is built on a living legal knowledge graph spanning more than 200 billion interconnected documents, with four million new documents added daily and continuous citation validity signals. The no-code workflow builder — allowing firms to design, test, and share their own multi-step processes without writing a line of code — is the feature I find most interesting for mid-sized and smaller practices. A firm's institutional knowledge, its preferred contract standards and playbooks, can be turned into repeatable, consistently executed workflows. Protégé also supports a multi-model approach, running across AI systems from Anthropic, Google, and OpenAI, which gives it a degree of flexibility that platform-locked tools simply cannot match.

Harvey AI raised $200 million at an $11 billion valuation in March 2026, co-led by GIC and Sequoia. The number is striking but the operational metrics are more telling: Harvey's CEO Winston Weinberg described the direction plainly — "AI isn't just assisting lawyers. It's becoming the system through which legal work gets done." The platform now has over 25,000 custom agents deployed across customer organisations, and its Agent Builder tool allows firms to construct bespoke workflows tailored to their specific practice needs. Harvey powers more than 100,000 professionals across 1,300 organisations, including DLA Piper (5,000 licences), Ashurst, and Baker Donelson. The platform reclaims an estimated seven to ten hours per lawyer per week.

Epiq AI won the 2026 Legalweek Leaders in Tech Law Award for Best Use of Artificial Intelligence in eDiscovery and Litigation. The recognition reflects Epiq's Laer™ platform, which orchestrates models, agents, and humans together in dynamic workflows. The stated headline numbers are striking: legal teams can conduct fact research up to 45 times faster and automate over 80% of the review process. More than 130 clients are already running live matters through it.

From "Smart Search" to "Autonomous Workflow"

What makes these platforms agentic — as opposed to merely capable — is tool use and data integration. An agentic system does not just process text. It uses tools: it searches Westlaw, queries Practical Law, analyses uploaded documents, and generates work product, all within a single coherent workflow. The system understands what it needs to do next and reaches for the right tool to do it.

To make the eDiscovery example concrete: under the old model, a solicitor receives a disclosure request. She searches the DMS, forwards batches of documents to a review team, chases for status updates over three weeks, assembles the privilege log manually, and drafts the disclosure statement at the end. Hours of coordination, multiple handoffs, significant risk of missed documents or inconsistent privilege determinations.

Under an agentic model: the system receives the request, formulates a workflow plan, executes document retrieval and relevance analysis across the entire corpus, flags privilege candidates for human review with supporting reasoning, drafts the privilege log entries, and surfaces a proposed disclosure statement. The solicitor reviews and approves. The coordination overhead collapses.

This is not science fiction. This is what the platforms above are being used for today.

The Impact on the Market Structure of Legal Services

The commercial consequences are already visible. According to the Wolters Kluwer 2026 Future Ready Lawyer Survey, just over half of all respondents — 51% — expect that tasks including legal research, document automation, and contract drafting will increasingly be reallocated to Alternative Legal Service Providers. When AI can execute these workflows autonomously and cheaply, the case for paying partner-rate time on them becomes very hard to sustain.

This creates a genuine strategic fork. Large firms can absorb the investment in agentic AI platforms, redeploy capacity upward into advisory work, and maintain market position. Small firms face the opposite risk: if they cannot access the same tools, they cannot compete on cost with the ALSPs, and they may not have the matter volume or advisory depth to compete on quality at the top end either.

The counter-argument — and I believe it — is that agentic AI is also the great equaliser. A two-partner litigation boutique with excellent process discipline, clean data, and the right platform can now execute document review at a scale that would previously have required a large team. The capabilities that were once the exclusive preserve of the Magic Circle are becoming accessible. The question is not whether small firms can afford agentic AI. The question is whether they can afford to be without it.

What This Demands of Us

Being clear-eyed about this requires acknowledging something uncomfortable: agentic AI does not just make lawyers faster. It potentially restructures what lawyers are for. If the system handles the workflow, the lawyer's value lies entirely in judgment, strategy, client relationship, and accountability. Those things matter enormously. But they require different skills from the ones legal training has traditionally emphasised.

The profession that thrives in the agentic era is one that has done two things: first, built the process maturity to specify clear objectives and review AI outputs rigorously; second, invested in the human skills that AI genuinely cannot replicate — contextual judgment, ethical reasoning, the ability to tell a client something they do not want to hear.

I have been saying for years that AI success in legal practice depends on data quality, process maturity, and human skill development — not just software adoption. The agentic era makes that argument not more complex but more urgent. The tools are extraordinary. The question is always whether the people using them are ready.

Questions worth sitting with:

1.    If agentic AI can execute an entire document review workflow, what is the billable justification for charging hourly rates for each task within it — and when does your firm need to have that conversation with its clients?

2.    The platforms above are built on proprietary legal knowledge graphs and content libraries. How dependent does your firm want to be on a single vendor's data ecosystem, and what is your exit strategy if that relationship changes?

3.    If 51% of routine legal work is expected to shift toward ALSPs, which of your current fee-earning tasks fall into that category — and what are you building to replace that revenue?