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

    The Agent on the Hook

    Laura Kennedy & Chelsey Lambert 5 min read
    The Agent on the Hook

    August 3, 2026

    The week's sharpest sentence came from a law firm that runs on software. On Tuesday, Crosby said it would take out professional liability insurance for its AI agents so they could do legal work without a lawyer signing off on every output. On Wednesday, Legora bought a litigation fact-intelligence company, its fifth acquisition of the year, because it expects agentic systems to multiply the volume of fact queries by orders of magnitude. The same Tuesday, LegalOn shipped a hundred attorney-built workflows meant to keep a human structure wrapped around the agent rather than take it away. Three moves in three days, each answering a version of the same question from a different seat.

    The trade press read them as unrelated, which they are on the surface. Read together, they mark a shift in what the industry is arguing about. For two years, the question was whether an AI agent could produce legal work good enough to use, and in more and more categories the answer is turning into yes. The question underneath it, the one this week brought into the open, is who carries the consequences when the agent is wrong. In the language of Adaptive Legal Practice Systems™ — ALPS™, the framework Laura developed at Circle Management Group — AI is the connective tissue running through a firm's ten technology layers. This week the field started building the thing that has to exist once that tissue does the work on its own. It started building the accountability layer.

    July 28, 2026: Crosby Insures Its Agents

    Crosby is a NewMod law firm built around AI agents, the kind that raised a Series B from Sequoia and told the market it would compete on deal velocity rather than the billable hour. On Tuesday, it said something no traditional firm has said out loud. It is taking out professional liability insurance for its agents so they can deliver autonomous legal work.

    Chief executive Ryan Daniels framed it as a threshold the firm had reached rather than a bet it was placing. "Today, our lawyers review every single work output," he said. "As agents have improved, it's become clear this won't be necessary in the future." The insurance is what replaces the review. Instead of a licensed lawyer inspecting each contract, memo, and turn of a negotiation before it leaves the building, the agent produces the work and a policy stands behind it. The firm is not describing a tool its lawyers use more efficiently. It is describing an actor whose output carries its own coverage.

    The move matters precisely because Crosby is small and unafraid to say the thing larger firms are managing around. Every firm running agents today has a human check somewhere in the loop, and most of the reassurance the market offers about legal AI rests on that check. The lawyer reviews, the partner signs, and the malpractice exposure sits where it always has. Crosby is proposing to remove the review and price the risk instead, which reframes the agent from a productivity layer into a producer of work product. If the agent's output is insurable on its own, the agent is being treated as something closer to a practitioner than a piece of software.

    The reason this lands now rather than as a stunt is that the economics favor it. When a firm bills by the hour, a lawyer reviewing every agent output is billable time, so the review pays for itself. When a firm bills for outcomes, as Crosby does, that review is pure cost, worth carrying only while the agent's error rate is high enough to require it. Daniels is saying that rate has fallen far enough that the math has flipped, and the firm's insurer will settle whether he is right. The price of that policy is about to become one of the more informative numbers in the industry.

    July 29, 2026: Legora Buys the Fact Layer for Agentic Litigation

    Legora made the week's second move on Wednesday, and it made it with money. The company acquired Wexler, a fact-intelligence platform built for litigation, in its fifth acquisition of 2026. Wexler was founded in early 2023, has 18 people across Europe and the United States, and reported eightfold revenue growth over the past year, with net revenue retention above 400%. Its engineering team will anchor a new Legora engineering center in London, and deal terms were not disclosed.

    Chief executive Max Junestrand explained the logic in terms that could serve as the week's thesis. "As agentic systems take on more litigation work, fact queries won't just grow, they'll multiply by orders of magnitude," he said. "Wexler is the infrastructure built to handle that scale today." A human litigator asks a finite number of factual questions of a record, bounded by the hours in a day. An agent working a case has no such bound. It can interrogate a document set continuously and generate factual queries at a rate no associate ever could. Legora is buying the machinery to answer those queries at that rate, on the theory that the bottleneck in agentic litigation will not be the reasoning; it will be the plumbing that feeds the reasoning reliable facts.

    Wexler's own function makes the logic concrete. It establishes what actually happened across a body of evidence, the unglamorous groundwork under every litigation argument, and its chief executive, Gregory Mostyn, framed the fit around the work rather than the technology. That is the layer Legora decided was worth owning. Not a smarter model, which it can rent, but the infrastructure that turns a pile of documents into a queryable set of established facts an agent can build on without hallucinating the record.

    Set Legora's move next to Crosby's and the shared premise shows through. Crosby is betting the agent is ready to work without supervision. Legora is betting it is about to work at a scale that breaks the tools underneath it. For Crosby the layer worth owning is accountability, priced as insurance. For Legora it is reliable fact infrastructure, bought as a company. Two pieces of the same emerging stack.

    July 28, 2026: LegalOn Keeps the Human in the Structure

    The same Tuesday Crosby moved to remove the human review, LegalOn moved in the opposite direction, and the contrast is the useful part. LegalOn released a library of more than a hundred prompt workflows for in-house teams, spanning ten practice areas from commercial contracts and data privacy to employment, M&A, and litigation, in US, EU, and Singapore variants. The workflows are attorney-built, customizable by the team using them, included at no extra cost, and pointedly not metered by token usage.

    Bari Williams, LegalOn's head of legal, described the design as getting to a finished product safely. "Prompt Workflows are crafted by attorneys, structured to get customers closer to a finished work product," she said. Rather than hand a lawyer a blank agent and expect them to become a prompt engineer through trial and error, LegalOn encodes attorney judgment into the workflow itself, so the human expertise sits in the scaffolding around the agent rather than in a review step after it. A data-breach response sequence or a worker-classification test arrives with a lawyer's reasoning already built into how the agent proceeds.

    That is a different answer to the same question Crosby is asking. Crosby's answer is that the agent has gotten good enough to trust without a human check, so replace the check with insurance. LegalOn's answer is that the agent earns trust by building the lawyer's judgment into its structure from the start, so the human shapes the work rather than merely inspecting it afterward. One firm prices the risk of autonomy. The other engineers the conditions under which autonomy earns trust. That two serious companies took opposite approaches in the same week is the clearest sign the industry has moved past whether to trust agents and into the harder work of deciding how.

    In Other News This Week

    The smaller moves ran on the same current, and several of them filled in the stack the anchor deals assume.

    Clio launched Clio Work in Canada on Monday, bringing its AI workspace home to the country where the company was founded, and called it the fastest-adopted product in its eighteen-year history. The Canadian version was built on the Jurisage acquisition Clio closed in June, positioned as an assistant that understands a firm's matters, their context, and the governing law rather than a generic chatbot. When the product that reaches into a firm's own matter data is the fastest thing a company has ever shipped, the appetite is for context, not novelty.

    Intapp reached general availability with Celeste, an agentic AI coworker aimed at the business of the firm rather than the practice of law, handling deal screening, conflicts, and the operational work underneath client matters. It puts an agent to work in exactly the place where a firm's own relationship and matter history is the differentiator. Casepoint, meanwhile, launched a Model Context Protocol server that opens its ediscovery platform to outside AI tools rather than trapping its data behind a proprietary assistant, giving the agents Legora is building fact infrastructure for another source to reach into.

    LexisNexis gave the week its economic anchor. In an interview published Tuesday, chief executive Sean Fitzpatrick reported the fastest growth in the company's history and put numbers behind the platform question this newsletter has tracked for weeks. Token costs, he said, run at less than one percent of parent RELX's revenue, and the subscription model lets profit outpace rising token consumption. LexisNexis works across Anthropic, OpenAI, and open-weight models and grows through bolt-on acquisitions such as Doctrine in France rather than large platform buys. The model is a swappable input priced near zero against the business. The curated data underneath is the moat.

    Two more items squared the picture. Artificial Lawyer asked on Monday whether clients should expect price cuts from legal AI and answered that, so far, they should not, because whether AI lowers prices turns on whether clients treat elite legal work as a utility or a luxury. In BigLaw, where clients keep paying rising rates regardless of efficiency, the savings accrue to the firm as profit, and only the NewMod firms are built to pass them on, which is why a firm like Crosby exists. And Ashurst Perkins Coie became the largest firm to hold an alternative business structure license from the UK regulator, a reminder that the question of who is permitted to deliver legal services keeps moving alongside the question of what an agent is allowed to do inside a firm.

    The CTO Gap Zone™ of the Week: The Accountability That Doesn't Transfer

    Last week's Gap Zone named the layer a firm can't rent, the encoded practice that stays inside every deal even as the model, the compute, and the distribution get leased from Big Tech. This week draws the next boundary, because the industry spent five days building the machinery of agent autonomy and barely addressed what autonomy raises: where the responsibility lands when the agent acts alone.

    Crosby's insurance move is the blunt version of the question. A professional liability policy pays out after something goes wrong. It absorbs the cost of an error rather than preventing one, and it does nothing about the professional-responsibility obligation that sits on the licensed lawyer regardless of who or what produced the work. A firm can insure the financial consequence of an agent's mistake. It cannot insure away the duty to supervise, because in every US jurisdiction that duty attaches to the lawyer and the license, not to the tool. The industry is building ways to price the downside of autonomous agents faster than ways to discharge the supervisory duty those agents don't remove.

    The two answers on offer this week bracket the problem without closing it, and neither resolves the middle ground where almost every real firm actually lives, running agents on real matters with a human check that is already thinning under volume pressure and no clear account of where supervision ends and delegation begins. A partner who signs off on a hundred agent-drafted documents a day without reading them is exposed in a way the malpractice framework has not caught up to. The review exists on paper. Whether it exists in fact is the question the insurer, the bar, and eventually a court will ask.

    This is why accountability belongs in the same conversation as the encoded practice the last few Gap Zones have circled. A firm's owned layer was never just its data and its methodology. It is also the judgment about when the work is fit to leave the building, the one thing in the stack that cannot be rented, encoded, or insured onto someone else's balance sheet. Remove the supervision without replacing it with something real, and the firm has automated its exposure along with its output. The agent got faster. The accountability did not move.

    What This Means For Your Firm

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

    First, treat the human-in-the-loop as a design decision that needs a real answer, not a phrase that ends the conversation. Crosby just demonstrated that the loop can be removed and the risk priced instead, which means keeping the loop is now a choice a firm has to justify rather than a default it can assume. Where exactly is the review in each agent workflow, what is the reviewer actually checking, and does that check still function when the volume triples? A firm that cannot answer those has a human-in-the-loop on the org chart and an autonomous agent in practice, the worst of both arrangements.

    Second, the accountability layer is now part of the owned layer, and it has to be built with the same discipline as the data underneath it. This is the connective tissue in ALPS™ doing the work, which makes supervising that work a first-order design problem rather than a compliance afterthought. Most firms have never mapped where an agent's output is genuinely reviewed versus where it merely passes through, and that map is now a strategic asset. Building it is ongoing work, 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™, and its methodology, The Circle Method™, encodes not just a firm's knowledge but the governance around how that knowledge gets used, so an agent produces the firm's work under the firm's supervision rather than around it. More on the framework at cmgconsultants.com.

    Third, watch the price of autonomy as it becomes public, because it is about to tell you what the market believes. Crosby's insurance premium, once it exists, prices the real-world error rate of unsupervised legal agents better than any vendor benchmark. Legora buying fact infrastructure rather than building it prices the belief that agentic litigation volume is coming fast. Each is a market signal about where the risk and the value actually sit, and a firm that reads them has a clearer view of the stack than one still evaluating which agent demos best.

    The industry spent this week building for a future in which agents do the work on their own, shipping the insurance, the fact infrastructure, and the workflow scaffolding that future needs. What it did not build, because no vendor can sell it, is the supervisory judgment that stays with the licensed lawyer no matter how good the agent gets. As Laura's book, Escaping the CTO Gap Zone, argues throughout, the firms that treat that judgment as a strategic asset and build the discipline to exercise it at scale are the ones that pull ahead. The rest will ensure the output, automate the review, and find out at the worst moment that accountability was the layer they were supposed to own.

    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.