Most European in-house teams now call AI reliable, and some ask it before they call a law firm, in a week when a law firm insured its agents' work, contract systems opened to any AI assistant, and Latham put up $1 million for associates' AI ideas.
Legal Business published its 2026 General Counsel Legal Tech Pulse on November 11, a survey of more than 250 senior in-house lawyers across the UK and Europe, run in association with Thomson Reuters. Close to 60% now say AI is a reliable resource for their team, up from 40% a year ago. Almost none of them want a smaller team as a result.
Read it with the sponsor in mind: Thomson Reuters sells legal AI, and one of the GCs quoted works there. The report gives no spend figures, so "we ask the AI first" is two anecdotes, not a measured fall in outside counsel cost. And trust is not throughput. The answers describe a lawyer asking a question and checking the reply, which is the same lawyer, a little quicker.
Eighty percent want more done without more people. That is a capacity problem, and a tool each lawyer operates does not add capacity, because the work still waits for a lawyer. Expensive people are still doing a steady flow of work that sits beneath their cost. Making them faster at it is not the same as taking it off them.
Crosby, an AI-native law firm, told Artificial Lawyer on November 10 that a $5 million policy covering its agents' work took effect at the start of the month. We covered the plan in August; this is the policy itself. Crosby says human review time has fallen to zero on 20% of its existing legal tasks, such as summarising negotiations and routing items for approval.
Crosby believes it is the first law firm to insure its agents' output, and reports 94% accuracy on whether an agent correctly finishes a task or escalates it. That figure is the firm's own, measured before launch, and it leaves six cases in a hundred going the wrong way. Insurance pays after a miss. It does not show the work was right, and as Bloomberg Law noted in July, it does not settle whether an agent may lawfully give legal advice.
When an agent finishes a piece of your work and nobody reads it, who set the rules it applied, and who on your side would know it had gone wrong? A policy answers who pays. It does not answer either question.
LawVu, a platform for in-house legal teams, said on November 11 that customers can connect five AI tools to its system through a Model Context Protocol server, with no custom development. A lawyer can ask Copilot which vendor contracts auto-renew next quarter, then assign the review to a colleague without leaving it. The same day, Legora and DiliTrust announced a similar link that only reads data.
For buyers this loosens lock-in. If the system of record talks to every assistant, you can change assistants without moving your contracts. The two designs differ where it matters, though. DiliTrust's link reads and changes nothing, with approvals and signatures staying in DiliTrust. LawVu's can update records from a chat window, so ask what gets logged when it does. Both are vendor announcements, and neither has published usage.
A connector makes the question quicker to ask. Say sixty contracts auto-renew next quarter: someone still has to read each one against your positions and decide. The connector found the work and handed it to a colleague. Agents that do the reviews to your own rules, and raise only the few points that need a lawyer, are what shorten the queue.
Latham & Watkins has launched a $1 million awards programme for associates who lead or materially contribute to AI initiatives, City AM reported on November 12. It follows Shoosmiths, which tied a £1 million bonus pot to one million Copilot prompts. Neither scheme measures what a client paid before and after.
The same day, Artificial Lawyer's founder asked whether Big Law is "AI gaslighting" its clients: the grander the announcement, the piece argued, the less clear it is how anyone on the client side benefits. It proposes that firms publish client benefit the way they publish revenue and profit. An internal award is a reasonable way to find good ideas. It is an incentive to adopt, and the firm keeps whatever efficiency follows unless the client asks for it.
Which of your matters cost less this year because your firm used AI, and can the firm show you the number? If it cannot, its AI programme is an investment in its margin, not in your bill.
A survey measured how much lawyers trust AI, an insurer priced an agent's mistakes, two vendors connected assistants to contract data, and a firm rewarded AI ideas. Each is a sign that AI is settling in. None says how much work was done, or who stood behind it.
Owning legal AI software does not get legal work done. Every story this week left the same thing in place: recurring work, specific to your organisation, that your lawyers should not be doing and that nobody has taken off them.
Flank deploys agents that do legal work. They are built to your templates, positions and escalation rules, they finish the work, and your own lawyers sign off the few points they raise.