|
|
> FROM THE EDITORS
Three things happened this week that read better together than apart. Linklaters launched its own in-house AI engineering team. A Colorado court set new standards for which AI tools a firm can responsibly run against client data. And The Lawyer published a piece — and a podcast — asking whether Harvey and Legora have peaked.
None of those are isolated. They describe the same shift: the legal-AI market has moved past the demo phase. The next moat is data control, audit trails, and workflow ownership — not access to the model. Five-minute read.
|
|
> 01 // STRATEGY · CAPITAL FLOWS
|
Linklaters builds its own AI lab — and tells the vendors why
Linklaters · Applied Intelligence · in-house lawyers + data scientists
Linklaters launched Applied Intelligence last week — a team that puts lawyers and data scientists at the same desk to build bespoke AI tools for matters where, in the firm's own phrasing, off-the-shelf products can't deliver. That last clause is the news. It is not framed as ‘in addition to Harvey’ or ‘alongside CoCounsel.’ It is framed as the place those tools stop being enough.
A magic-circle firm with a permanent AI engineering function is a different procurement target than one with an innovation committee. The buying conversation shifts from which platform should we sign? to which capabilities should we build, and which should we rent? For vendors, the worry is not that Linklaters won't be a customer — it will. The worry is that the high-margin, bespoke work is the work the firm now intends to keep.
Watch whether the next two or three top-twenty firms follow with named, staffed, named-not-renamed AI groups. The pattern, if it repeats, is the in-housing of the part of the stack that used to be Harvey's growth story.
WHY IT MATTERS
‘Off-the-shelf can’t deliver’ is the line that re-prices the legal-AI market. Vendors get the long tail. The firms keep the matter-shaping work — and the data that comes with it.
[ Legal IT Insider · launch coverage ]
|
|
|
> 02 // STRATEGY · CAPITAL FLOWS
|
The Lawyer asks if Harvey and Legora have peaked
Harvey · Legora · data access · open-source pressure
Two pieces in The Lawyer this week put the question on the table: do Harvey and Legora have a data problem, and have they peaked? The reporting is not a takedown. It is a careful read of the same thing Linklaters' move implies — the unique-value layer in legal AI is shifting from model access to workflow-specific, firm-controlled data, and the vendors with the strongest market caps are the ones with the least of it.
On the same day, LawNext covered the release of MikeOSS, an open-source project that claims feature parity with both Harvey and Legora — assistant, tabular review, the lot. MikeOSS is not the threat. MikeOSS is the signal: when a single developer can publish a credible clone of your UI in a weekend, the moat is no longer the UI. It is what trains on top of it.
Harvey's response so far has been the Legal Agent Bench — a public benchmark designed to make ‘how good is this thing, really’ an answerable question. Legora's response has been M&A. Both are plausible. Neither is a data-access strategy, which is the strategy the next twelve months will reward.
[ The Lawyer · ‘Harvey and Legora have a data problem’ ] [ The Lawyer · ‘Have Harvey and Legora peaked?’ (podcast) ] [ LawNext · MikeOSS analysis ]
|
|
|
> 03 // RISK · CITATIONS & SANCTIONS
|
A Colorado court starts picking AI tool winners
Morgan v. V2X · work product · confidentiality · discovery
In Morgan v. V2X Inc., a Colorado court set out new standards for how AI tools handle confidential discovery materials — what counts as work product, what counts as a privilege waiver, and which deployment models survive a motion. Clio's read of the opinion calls it a winner-picking exercise: the bench is becoming an active filter on which tools a firm can responsibly run against client data.
Pair that with the new Icertis survey of in-house legal teams: nearly half say they would not detect an unauthorized or incorrect AI-agent action until days or weeks after it happened. The picture is consistent. Courts are tightening the rules at the same moment buyers are admitting they can't see what their agents are doing.
The practical consequence for firms shopping in Q3: any tool whose audit trail cannot answer ‘what did the agent touch, when, and on whose authority’ will fail diligence the second time it is asked, not the first.
[ Clio Blog · Morgan v. V2X analysis ] [ LawNext · Icertis agent-visibility survey ]
|
|
|
> 04 // TOOLS · PRODUCT
|
iManage moves on contract review — and the document layer eats up the stack
iManage · Playbook Analysis · contract AI
iManage added Playbook Analysis to its AI-contract toolset this week. For most of its history iManage was ‘where you put the documents.’ That is now an understatement: the document-management layer is becoming a contract-review competitor, in the same way NetDocuments became a contract-AI distribution channel through ndMAX.
The pattern matches Lexis and Thomson Reuters pitching themselves as ‘foundational infrastructure’ (Issue 003). The infrastructure incumbents are using their distribution — every desk inside the firm, every contract already inside the system — to land features that pure-play vendors had to sell from scratch.
Standalone contract-AI vendors (Spellbook, Ironclad, Gavel Exec, the new entrants like Scissero's open-source Suzie Law) still win on speed and craft. The question is whether speed and craft survive a renewal cycle where iManage and NetDocuments are giving the basics away with a seat the firm has already paid for.
[ Artificial Lawyer · iManage Playbook Analysis ] [ Artificial Lawyer · Scissero Suzie Law ] [ LawNext · Gavel Exec for Web ]
|
|
> THE DOCKET · MOVES THIS WEEK
|
|
VENDORS
Watch: Harvey's Legal Agent Bench is the company's bid to turn the ‘does it actually work’ question into a published number. Whether Lexis and Thomson Reuters submit to the bench is the next signal.
|
|
FIRMS
Next 30 days: watch for a second magic-circle firm to announce a named, staffed AI engineering function in Linklaters' mould. One is a launch; two is a category.
|
|
COURTS
Morgan v. V2X joins a short but growing list of opinions a buyer will be asked to address on diligence. Expect vendor sales decks to start citing it inside a fortnight.
|
|
OPEN SOURCE
MikeOSS — a one-developer feature-clone of Harvey and Legora — is not yet a product. It is a proof. The next twelve months will tell whether the legal-AI ‘secret sauce’ survives a public reference implementation.
|
|
|
> THE HOLD · ONE PARAGRAPH
|
|
The demo era of legal AI is closing. What replaces it is unglamorous: data controls, audit trails, named owners, and the patient work of integrating models into matter workflows that survive a court order. The firms and vendors that will look smart in 2027 are the ones doing that work now — quietly, with engineers on staff, and with the assumption that the next buyer they meet will have read the opinion. — The Editors▌
|
|
Exhibit AI reports on the AI industry. We do not provide legal advice. Sponsorships are disclosed and never shape coverage. Issue N°001 published MON 11 MAY 2026.
|
