Translating the Tools & Trends That Shape the Legal Profession
    Rising Above the Noise

    Everybody Picked a Partner

    Laura Kennedy & Chelsey Lambert 5 min read
    Everybody Picked a Partner

    Monday, July 27, 2026

    The handshakes started on Tuesday. Epiq embedded Anthropic's Claude into its agentic e-discovery workflows, and Box shipped a control system built to govern Claude, ChatGPT, and Gemini reaching into a firm's documents. Wednesday, Willkie Farr announced it was building proprietary tools on OpenAI's frontier models. Thursday, Harvey and Microsoft deepened an alliance to the point where Microsoft's own legal department became a Harvey customer. Friday, Spellbook moved its contract review inside Google Docs. Five working days, and four legal companies had each paired off with a different Big Tech firm.

    The trade press covered them as separate deals, which they are. Read in sequence, they are also something more coherent. The legal AI industry spent two years arguing about whether it would build its own stack or lease someone else's. This week it stopped arguing and started signing. The model, the compute, and the distribution surface are being rented from the companies that make them best. The layer nobody is renting is the firm's own practice — the methodology, the accumulated matter data, the judgment that took decades to build. In the language of Adaptive Legal Practice Systems™ — ALPS™, the framework Laura developed at Circle Management Group — AI is the connective tissue that runs through a firm's ten technology layers. This week the field agreed on where that tissue comes from. It comes from Big Tech. What it connects to has to come from the firm.

    July 22, 2026: Willkie Builds Its Own Tools on OpenAI

    Willkie Farr & Gallagher announced a firmwide partnership with OpenAI on Wednesday, and the shape of the deal is more interesting than the headline. Willkie is rolling out ChatGPT Enterprise across the firm, which is the part that reads like every other enterprise AI announcement. The part that matters is what sits on top. Willkie is integrating OpenAI's Codex into Willkie Works, the firm's internal AI and innovation organization, and it expects to develop a "Codex for Legal" capability of its own. The firm is not buying a finished product. It is renting frontier models and a coding agent, and using them to build proprietary tools around the way its own lawyers work.

    Jason Boehmig, who leads OpenAI's legal vertical as head of product, framed the arrangement in exactly those terms. "With OpenAI's frontier models and Codex, Willkie can turn its expertise and ideas into tools designed around how its professionals work," he said. The operative words are Willkie's expertise and Willkie's ideas. OpenAI supplies the horsepower. Willkie supplies the thing the horsepower runs on, which is its own practice.

    That distinction shows up in what Willkie already has in the ground. The firm has been shipping a suite of applications under the Wendell name — Wendell Intelligence, Wendell Financial Insights, Wendell Horizon, a private funds secondaries platform, a class action claimant platform. Those are not products a lawyer could buy from OpenAI, because they encode workflows specific to how Willkie runs particular kinds of matters. Codex for Legal is the tooling that lets the firm build more of them faster. Thomas Cerabino, Willkie's chairman, put the strategic frame plainly: "AI is changing the practice of law," and the firm intends to meet that "through combining exceptional legal talent with transformative technology." John Elbasan, the firm's chief digital and information officer, called it "an important milestone in Willkie's AI strategy."

    The move lands with more weight than it would have a month ago, because OpenAI's newest frontier models only became generally available in early July after a White House security review that had held them back. Willkie is building on a model layer that was gated by an interagency process the firm does not control, a dependency worth naming. It is also the encoding argument this newsletter tracked on July 6, when Kirkland formalized its litigation methodology into a proprietary system. Willkie is running the same play with a different partner. Rent the intelligence, encode the practice, own the tools that result.

    July 23, 2026: Harvey and Microsoft, and the Customer That Sells the Platform

    Harvey deepened its alliance with Microsoft on Thursday, and the detail that made the news is a good one. Microsoft's own Corporate, External, and Legal Affairs organization — roughly two thousand legal and compliance professionals — is deploying Harvey. The company that builds Copilot, that ships its own MAI models, that owns the productivity surface every firm runs on, chose an outside legal AI platform for its own lawyers. In the same breath, Harvey committed to expanding its internal use of Microsoft 365 and Copilot, putting its teams on the same platform its customers use.

    Winston Weinberg, Harvey's chief executive, kept his comment on the relationship itself. "Microsoft's products are foundational to the work lawyers do, so we have always been grateful for our deep relationship with their team," he said. Antony Cook, Microsoft's deputy general counsel, described the value in operational terms, saying the collaboration "will help our teams work more efficiently and focus on the complex, high-impact work that matters most." Neither quote oversells it, which is appropriate, because the significance is structural rather than rhetorical.

    Harvey runs on Microsoft Azure, and its legal intelligence has been embedded as an agent inside Microsoft 365 Copilot since March, extending into Copilot's Cowork surface in June. The two companies have been converging at the infrastructure layer for more than a year. What changed on Thursday is that Microsoft crossed from platform partner to paying customer. It could have pointed its two thousand lawyers at Copilot alone and called it a showcase. It bought Harvey instead, which says the general-purpose assistant and the legal-specific application are different products serving different needs, even when one company could plausibly build both.

    That is the useful signal for firms watching the platform question. The company best positioned in the world to make the model layer and the application layer the same thing decided to keep them separate. Harvey earns its place on Azure by being the thing Azure is not, a system that understands legal work. It reports more than a thousand in-house legal teams now, and Microsoft's is the one that tells you what the arrangement is worth.

    July 24, 2026: Spellbook Rents a Billion-User Front Door

    Spellbook closed the week on Friday by moving inside Google Docs. Its contract review capabilities — Reviews, Playbooks, and the Ask function — now run natively in the document, with redlines applied on the page. The workflow it replaces was clumsy in a way every transactional lawyer will recognize: a contract drafted in Google Docs had to be downloaded, opened in Word, reviewed there, and re-uploaded. Spellbook collapsed that round trip by going to where the document already lives.

    The company's framing was blunt about why. Over a billion people use Google Docs. Spellbook did not build a competing editor or ask its users to adopt a new surface. It rented Google's distribution and put its own reasoning layer inside it. This is the pattern Spellbook's chief executive named back in late June, when the company was pushing into finance systems on the same logic: go to the surfaces the users already work in rather than making the users come to you. On Friday that logic reached the largest document surface on the consumer internet.

    Set the three anchor deals next to each other and the division of labor is hard to miss. Willkie rented frontier models from OpenAI to build its own tools. Harvey rented compute and distribution from Microsoft to place its application in front of every firm on Azure. Spellbook rented a billion-user document surface from Google to eliminate a workflow seam. Three legal companies, three Big Tech partners, one shared decision about which parts of the stack to build and which to lease. None of them rented the thing that makes their product theirs. The model, the compute, and the reach came from outside. The practice stayed in.

    In Other News This Week

    The smaller moves ran on the same current, and several of them supplied the plumbing that makes renting the model safe to do.

    Epiq opened the week on Tuesday by embedding Anthropic's Claude into its Epiq AI Accelerate workflows for e-discovery, making it the fourth legal player in five days to build on a Big Tech model rather than its own. E-discovery has been slower than most categories to move to agentic execution, and Epiq's choice to reach for a frontier model instead of a home-grown one fits the week's larger pattern precisely.

    Box, also on Tuesday, shipped the governance layer the rest of the week assumes. Its new agentic control system, part of the E-Advanced plan, gives administrators guardrails over what external AI agents connected through the Box MCP server are allowed to do — scoped permissions that can allow file creation only in approved folders, block external sharing, detect prompt injection, and require human approval at defined points. The controls apply to Claude, ChatGPT, and Gemini alike. A Box executive framed it as making sure agents "have what they need to function effectively, without accessing, modifying, or exposing content beyond the scope of its intended task." When a firm rents a model and lets it reach into client documents, this is the seatbelt.

    Flank launched Record on Thursday, an autonomous system that maintains contract data by deploying agents to watch where signed contracts land, extract real metadata from executed documents, and flag amendments as they happen. Co-founder Jake Jones described the agents as "the maintenance workforce for the record itself," working underneath the data layer rather than on top of it. It is a small product with a large idea inside it, which is that the firm's record of truth is itself something that needs continuous tending, not a one-time import.

    NetDocuments added three plaintiff-side applications to its ndMAX Studio on Wednesday — a medical chronology app that turns hundreds of pages of records into a structured timeline, a demand generator that drafts jurisdiction-appropriate letters, and an expert witness profile builder that extracts credentials and prior opinions while flagging testimony inconsistencies. The apps run on the firm's own documents, and the studio now offers thirty-nine of them, which is a quiet reminder that the context layer keeps compounding while the interface debate rages above it.

    Two economic data points closed the loop. Artificial Lawyer published a piece on Wednesday asking who captures the savings from legal AI, and the numbers cut against the firms. Thomson Reuters' 2026 corporate law department research found thirty-six percent of general counsel expect to increase outside counsel spend over the next year, against only twenty percent expecting to cut it, even as firm billing rates rose 7.4 percent year over year in the second quarter of 2025 against 2.8 percent inflation. The AI dividend, so far, is flowing to the firms, not the clients, and the piece argued that the clients who redesign their workflows are the ones who claw it back. LexisNexis, meanwhile, reported first-half legal division revenue up from £900 million to £959 million despite the market turbulence earlier this year, a reminder that the companies sitting on decades of proprietary legal data are better insulated from AI shocks than almost anyone else in the field.

    The CTO Gap Zone™ of the Week: The Layer You Can't Rent

    The last several Gap Zones have circled one asset from different sides. The model and the content are rented. The firm's own context is the layer it owns. Owning it is not the same as encoding it, and once encoded, someone has to be able to read the meter on what it costs. This week draws the boundary line itself, because the whole industry spent five days publicly deciding what it will rent and what it will keep, and the answer is remarkably consistent.

    Everything that can be bought from a hyperscaler is being bought from a hyperscaler. The frontier model comes from OpenAI or Anthropic. The compute comes from Azure or Bedrock or Google's cloud. The distribution comes from Microsoft 365 or Google Docs. These are not the layers where a law firm builds advantage, because they are exactly the layers where a company with the scale of OpenAI or Microsoft will always outbuild a firm. This week four legal companies conceded the point by signing for them rather than replicating them, and that concession is the correct call, made in unison.

    The layer that stayed in every deal is the one that cannot be rented. Willkie kept its practice methodology and built tools to encode more of it. Harvey's whole value to Microsoft is a legal reasoning layer Microsoft chose not to build itself. Spellbook's playbooks are the firm's accumulated negotiation positions, not Google's. The reason none of these companies outsourced that layer is that there is nowhere to outsource it to. No hyperscaler has a firm's matter history. No frontier model has a firm's judgment about when to concede a point and when to hold. That knowledge exists in exactly one place, and it is the place most firms have never turned into anything an agent could use.

    That is the gap this week made visible. A firm can now rent a world-class model, world-class compute, and a billion-user distribution surface with a few signatures. What it cannot rent, and what none of its vendors can supply, is a structured, current, queryable version of its own practice. Point a rented model at that asset and it produces the firm's work. Point the same model at a shared drive nobody has ever organized and it produces a competent average of the public internet with the firm's name on top. The model is identical in both cases. The difference is entirely the layer the firm was supposed to own and usually hasn't built.

    What This Means For Your Firm

    Three implications for any firm running an AI program through the back half of 2026.

    First, treat the build-versus-rent question as settled at the model and infrastructure layers, because the market just settled it for you. Standing up your own foundation model or your own compute stack is not where a firm competes, and four sophisticated legal players demonstrated as much this week by renting all of it. The energy a firm spends evaluating which frontier model is marginally smarter this quarter is energy not spent on the layer that actually differentiates it. Rent the engine. Move your attention up the stack.

    Second, the layer to build is the one every deal this week kept in-house, which is the firm's own encoded practice. This is the connective tissue finally getting something worth connecting to. Most firms own the raw material — the matters, the redlines, the deal histories, the institutional judgment — and have never structured it into a form a model can query. That structuring is ongoing strategic work, not a one-time project, which is why Laura co-founded Circle Management Group, a legal technology strategy firm that partners with law firms to build and maintain the technology ecosystem their practice runs on. CMG delivers it as a continuous partnership it calls Technology Strategy as a Service™, or TSaaS™, and its methodology, The Circle Method™, encodes a firm's institutional knowledge and keeps it current so a rented model produces the firm's practice rather than the market's average. ALPS™ names what the ecosystem is; The Circle Method™ is how the owned layer gets built. More on the framework at cmgconsultants.com.

    Third, mind the governance and economic seams the rentals create, because they are where the risk and the value both leak. Box shipped its agent control system this week for a reason. The moment a firm lets a rented model reach into client documents, it needs scoped permissions, audit trails, and human checkpoints, or it has handed a probabilistic system unsupervised access to privileged material. And the AI dividend research is a warning on the other side: if a firm captures real efficiency from these tools and its clients notice before the firm decides how to price it, the savings conversation happens on the client's terms. Renting the model is cheap and fast. Governing it and pricing around it is the work that separates the firms that benefit from the ones that get surprised.

    Everybody picked a partner this week. The partners supplied the model, the compute, and the reach — the parts of legal AI that were always going to come from somewhere else. What no partner could supply, and what stayed inside every one of these deals, is the firm's own practice. As Laura's book, Escaping the CTO Gap Zone, argues throughout, the firms that treat that practice as a strategic asset and invest in encoding it with discipline are the ones that pull ahead. The rest will rent a brilliant model and wonder why it keeps giving them everyone else's answers.

    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.

    The Daily Brief

    Every morning, the most important legal tech developments — curated and context-rich — delivered before 8 AM.

    Free. No spam. Unsubscribe any time. Read by 28,000+ legal professionals.