Monday, June 29, 2026
Perplexity took over an event space in New York on Wednesday for an invitation-only crowd of in-house counsel, lawyers from small and mid-sized firms, and legal tech founders, and launched Computer for Counsel. The product routes more than 20 frontier AI models per task, runs natively within Microsoft 365, and connects to over 400 tools via MCP. The launch was the surface event of the week. The strategic story was what two of the largest legal AI vendors did in the same three days. Thomson Reuters rebuilt CoCounsel Legal from scratch on Anthropic's Claude Agent SDK on Monday. Clio published a positioning piece on Anthropic's legal AI push on Tuesday, then announced a partnership with Perplexity twenty-four hours later.
Three vendors. Three days. The same strategic decision underneath all of it. None of the three is competing on model capability. Each is making a clear call on which layer is worth building, buying, or renting. The model is rented. The content is anchored. The firm's context is where value is produced. Everyone making a serious move in legal AI is now running the same playbook from different angles, and the rest of this issue traces how each of them ran it.
Perplexity Launches Computer for Counsel
Computer for Counsel is Perplexity's formal entry into legal as a vertical, and the product's architectural choices define where Perplexity is competing and where it is partnering. The product is explicitly not a replacement for Westlaw or LexisNexis. Perplexity described it as a research, drafting, and workflow layer that sits on top of the open web, firm systems, and specialized legal data sources.
The product is built around a multi-model engine. Perplexity routes more than twenty frontier AI models across subtasks, choosing the strongest model for each step rather than betting the platform on any single vendor. Nathan Barksdale, Perplexity's General Counsel and Vice President of Operations, named the design choice directly. "We have twenty different models integrated into Perplexity Computer," Barksdale said, "and when you release these sub-agents, it's like releasing a team. The best agent that does the best thing will be assigned a certain task. When you're working within other competitors' similar applications, you're working solely with their models, and you're relying on those models to do generally everything, but we all know that there is a model race out there, and models are very good at very different things." The pitch is that lawyers do not have to pick a model. The platform picks for them, task by task.
Computer for Counsel runs natively inside Microsoft 365, with the agent drafting in Word, retrieving files from SharePoint, and pulling context from Outlook and Teams. Premium legal data sources at launch include Midpage for case law and statutes, with a citator that links every output to the underlying authority, plus LegalZoom and Deel. More than 400 tools connect via MCP, including DocuSign, NetDocuments, Box, DeepJudge, Carta, PandaDoc, and Drata. Integrations with Clio and Ironclad were announced for the near future. Available through Perplexity Enterprise and Max plans, the product is positioned to compete in the small- and mid-sized firm and in-house markets that BigLaw-focused legal AI platforms have served less directly.
The signal is the entry's directness. Perplexity is not building a legal research database. It is building the connective layer above the existing databases and wiring that layer into the productivity software lawyers already use. The position is the most aggressive Big Tech move into legal so far, and it is built on the assumption that Microsoft 365 plus MCP plus the right legal content sources gets you most of the way to a platform.
Clio and Thomson Reuters Define the Vendor Response
Clio ran a two-move play in forty-eight hours. On Tuesday, June 23, Clio published a positioning piece titled Anthropic's Legal AI Push. The argument was pointed. "Anthropic is moving deeper into legal," the piece opened, "but the context that makes AI-generated work defensible lives in the firm, not the model." The thesis took several paragraphs to unfold. Legal work assesses interlocking obligations. Documents tie to matters, matters tie to clients, clients tie to conflict-check histories, billing arrangements, retainer balances, jurisdictions, court rules, calendar deadlines, and chains of prior correspondence. Every piece of legal work sits inside that web, and the work is defensible only when the AI producing it can see the whole web simultaneously. A model accessing legal text, however carefully, cannot see the live state of a firm's practice. The matter, the client, the conflict, the calendar, the billing record, the document history — none of it lives in the model, and all of it comes from the firm's business systems. The argument was a public statement of where Clio thinks the legal AI moat lives.
Twenty-four hours later, Clio announced a partnership with Perplexity. Vincent by Clio, the AI assistant built on the vLex corpus and the Jurisage Canadian dataset Clio bought ten days earlier, will be available through Computer for Counsel via a new MCP connector launching in the coming months. The partnership specifically targets large law firms, the segment Clio has been moving into through Clio Operate and Clio for Enterprise. John Foreman, Clio's Chief Product Officer, briefly framed the move. "We're moving from a world where people learn tech to a world where tech learns what people are trying to accomplish," Foreman said. Vincent's legal research, citator analysis, and authority-backed reasoning will flow through Computer for Counsel, while the firm context Clio holds remains anchored on Clio's side.
Read together, the two moves describe the playbook in compressed form. Argue publicly that the model alone is undifferentiated and that the firm context is the moat. Then partner with the largest Big Tech entrant in the same week, so the firm context flows through the Big Tech platform as well. Critique and partner are not in tension. They are the same strategic position rendered as two complementary actions.
Thomson Reuters made the same architectural choice with a different mechanism. On Monday, June 22, the company opened early access to the next generation of CoCounsel Legal, the most substantial rebuild of its flagship legal AI since the Casetext acquisition in 2023. The product was rebuilt from scratch on Anthropic's Claude Agent SDK. Ragunath Ramanathan, Thomson Reuters' president of the legal professionals business, named the decision directly. "We abandoned all of the old code lines, and we said this feels like a shift that we need to rebuild the product from scratch," Ramanathan said. "Of course, we've done it on the basis of Anthropic and the latest models, which in retrospect was a great decision." Beta users reportedly told the company they "F#@%ing loved" the product, as quoted by CEO Steve Hasker. General availability in the United States is planned for August 2026, with Canada, the United Kingdom, and Australia to follow.
The new CoCounsel Legal includes workspaces that hold matter context across sessions, a Brief Builder agent for complex briefs and motions with Westlaw and Practical Law citation checking, and an organizational intelligence layer that lets firms encode their own expertise into the platform. The integration list is broad: Microsoft 365, HighQ, iManage, NetDocuments, SharePoint, Icertis, DeepJudge, Supio, and Smokeball. An MCP integration with Anthropic also lets lawyers working in Claude reach CoCounsel Legal without switching platforms. Hasker also confirmed that Thomson Reuters is developing its own LLM, called Thomson, which suggests the company sees the model layer as something it wants optionality across rather than dependence on.
Thomson Reuters released a second piece of news the same day. Its Future of Professionals Report 2026 warned of a widening gap between AI adoption and the value organizations are getting from it, and noted that mid-career lawyers are considering leaving their organizations over the next two years due to disillusionment with AI. The report sits alongside the CoCounsel rebuild as the framing argument. Adoption is now broad. Value is uneven. The platforms that close the gap will win, and Thomson Reuters is betting that grounding agentic AI in Westlaw and Practical Law content is the way to close the gap.
The pattern across Clio and Thomson Reuters is identical in shape. Rebuild on a Big Tech model layer through MCP or an Agent SDK. Anchor the differentiation in proprietary content. Hold the firm context as the part of the architecture that the platform owns, not Big Tech. Both vendors made the same architectural move in the same week, in response to the same Big Tech entry. The legal AI category has agreed on where the moat lives. The model is rented. The content is owned. The firm context is the working surface where the value gets produced.
In Other News This Week
The Bloomberg Law piece that landed Monday, June 22, titled "Law Firms Adopt AI Tools at Unheard-Of Pace as Enthusiasm Grows," called the curve. Legal-specific AI tools, the piece said, are now as commonplace as other software that has long been a mainstay of an attorney's tool box. The article does not break new ground on a specific vendor or firm move. It marks the moment the adoption story stops being remarkable.
BARBRI, the legal education company, acquired Lega on June 24 to accelerate practical AI fluency across the legal profession. The deal extends the training and education theme announced earlier in the month by Clio's Legal AI Accelerator and the Florida Bar Preferred Bar Program. Three different organizations now treat AI fluency as a baseline professional credential rather than an optional skill. The competency floor is moving.
Jackson Lewis won the 2026 Litera Foundation Innovation Award on Thursday, June 25, for turning matter data into a firm-wide growth asset. The recognition fits the same pattern Issue 5 covered around content audits. Firms that treat their own matter data as a strategic asset, rather than a passive byproduct of practice, are the ones being recognized for what they have built around AI.
Smaller moves continued to compound. Courtroom Insight launched an MCP server on June 23, powering AI tools with what the company called clean legal professional intelligence. Firm Prospects partnered with UniCourt the same day on a broader dataset of attorney and law firm profiles. Litera published a customer story about Cvent's legal team reviewing 360 contracts in minutes during a high-stakes acquisition. Centari launched Amendment Awareness and Deal Maps to help transactional teams understand how documents relate to one another over the life of a deal. Docusign released its Intelligent Agreement Management platform for Perplexity Computer and Computer for Counsel on June 24, joining the connector ecosystem Perplexity built the product on. Sapphire Legal launched a platform for the fractional general counsel market, a category almost no one else is chasing and one that will become more visible as the in-house front opens up further.
The CTO Gap Zone™ of the Week: The Rented Model
A firm running legal AI in 2026 is now renting two of the three pieces it depends on. The model is rented from Anthropic, OpenAI, Perplexity, or another foundation company. The content is rented from Thomson Reuters, LexisNexis, vLex, Wolters Kluwer, or whichever legal data provider the platform integrates. What the firm owns and operates is the third piece. The matter context. The firm playbook. The historical correspondence. The conflict-check history. The pricing data. The clause precedents. The institutional memory of how the firm actually practices.
Most firms have never thought about their AI deployments as a rental relationship layered on top of an ownership relationship. The procurement conversation has been framed around tool selection, which obscures what is being purchased and what is being rented at each layer of the stack. The gap zone is the lack of clarity about that distinction. When a firm renews its CoCounsel subscription, it is renewing a license to access Westlaw content through an Anthropic-powered platform. When a firm subscribes to Clio Work, it is paying for access to vLex content through whatever model Clio routes to in a given task. When a firm signs on with Computer for Counsel, it is paying Perplexity for routing across twenty models and access to integrated content from Midpage, LegalZoom, Deel, and others. None of those rented layers belong to the firm. The firm-context layer is the only one that does.
The strategic question this raises is not which platform to choose. The question is whether the firm has done the work to make its own context layer a real asset. A firm with disciplined matter data, structured playbooks, codified clause libraries, and clean conflict and pricing records can plug that context into any platform and get differentiated work product. A firm with messy matter data, inconsistent playbooks, scattered precedents, and undocumented institutional knowledge will get generic work product from the same platform, regardless of which Big Tech model sits underneath. The architecture has commoditized the rented layers. The firm-owned layer is what the architecture has made more valuable, not less.
The practical move for a firm leader is to start treating the firm-context layer as the asset to invest in. That means content audits plus a parallel investment in the discipline of capturing and structuring matter context as the firm produces it. The platforms can be swapped over time. The model layer will keep improving. The content providers will keep adding sources. What does not get easier is building a clean, current, organized firm-context layer from a starting point of disorder. That work is slow, unglamorous, and the only thing that produces durable advantage in the architecture that just settled into place.
What This Means For Your Firm
Three implications for any firm running an AI program in the next eighteen months.
First, the model is now a rented utility. Treat it that way. Legal AI platforms have decided not to compete with Big Tech on model capabilities, which means the firm should not either. Pick the platforms that route across the strongest models and let the model question take care of itself.
Second, the content layer is also rented, but the rent structure varies by platform, and that matters. Some platforms own the legal content outright through acquisition, as Clio does with vLex and Jurisage. Some platforms license content through partnerships, as Legora does with Wolters Kluwer regulatory data. Some platforms ground in their own proprietary content libraries, the way Thomson Reuters grounds CoCounsel in Westlaw and Practical Law. Each rent structure has implications for pricing, for access durability, and for what happens when the underlying relationships shift. A firm leader should know which structure the firm is buying into and what it costs if the structure changes.
Third, the firm-context layer is the only thing the firm actually owns in the new architecture. Invest there accordingly. The platforms compete with each other and partner with Big Tech. Big Tech competes with itself and partners with the platforms. The firm sits underneath all of that, paying for both layers and producing the work the layers exist to support. The firms that recognize that the architecture has settled and that the only durable asset is the part the firm controls will spend the rest of 2026 deliberately building that asset. The firms that do not will keep buying tools and wondering why the tools are not producing the differentiation they were sold on.
The model is rented. The content is rented. The firms that understand which layer is still theirs to own, and invest there with discipline, will be the ones with something to show for it by the end of the year.
Laura Kennedy and Chelsey Lambert are Co-CEOs and Co-Founders of Legal Tech Decoded. Rising Above the Noise lands every Monday at legaltechdecoded.com.
