Everyone Is Tired. Nobody Is Stopping.
An important note: in a very rare move for us, we used Claude to help write this blog. There was simply too much information to cram into one blog in the limited time span between 10pm and midnight. Much of the following was a Gen AI synthesis of 90 of dictated content. We did, of course, review the output afterwards to verify we agree with the content, but the style and tone may sound somewhat more robotic than our usual writing voice. We hope you will forgive us for taking the lazier path of AI-assisted writing this time.
Everyone we spoke to in September was exhausted. Not one of them was slowing down (or could slow down).
We had roughly 60 conversations in the last three weeks. We spoke with innovation, KM and technology leaders at law firms in the US, the UK and Australia, with in-house legal teams from startups to Fortune 50 companies, and with investors, bankers and other vendors. Some conversations were over coffee. A few were well past bedtime.
This blog isn’t a survey. It is a reflection of the state of the market from what our friends were willing to share. All details have been anonymized and sanitized for the purposes of this blog.
Exhaustion is the New Metric
Innovation teams are thin, and they are carrying firm-wide transformations.
But exhaustion is not the same as disinterest. Several firms made the same distinction: appetite and capacity are different questions.
Lawyers are more open to new ideas than they have ever been. What they cannot do is change how they work at the speed the market changes. As one person put it: you are turning a bus around.
The pile of tools doesn't help. One firm holds four separate products that all help with drafting, and expects to cancel three of them next year. Another rolled out Microsoft Copilot alongside its legal AI platform, and lawyers started asking which one to use for what. Some clients are adding to the load by asking their firms to use the clients' own AI tools too.
To the surprise of absolutely no one, everyone is exhausted.
Harvey and Legora Won the Booth, but they Haven't Won the Market
By all accounts, ILTACON 2026 was the most extravagant yet. Some companies created private meeting rooms on the trade floor, organized headline musicians, and even employed robot baristas. Legal tech now sponsors the US Open and has signage in baseball stadiums.
One CIO summed up the mood on the vendor floor: "If you're not Harvey or Legora, you're scared of Harvey and Legora."
The adoption is real. Many firms are seeing more than three-quarters of their attorneys use Harvey or Legora regularly. Even then, usage was seen as individual lawyers making themselves more efficient, not improving how the firm works.
The loyalty is thin.
“Good enough” is doing a lot of work in both directions:
It is displacing point tools. Firms are cancelling long-established review and drafting products because the Gen AI platform does an adequate job.
It is also undermining the platforms. A handful of firms have gone direct to the frontier models, built their own practice-specific skills, and report better results than Harvey or Legora give them.
Consumption pricing is also coming. Legora has started, and most people expect Harvey to follow within a year. Firms are trying to lock in seat pricing before that happens.
In our view, many buyers see the platforms as a stop gap, not a permanent layer. Microsoft and Google own the screen lawyers work in. The model labs own the intelligence. That leaves a narrowing space in between for everyone else.
Going Straight to the Source
The firms going direct to the labs gave three reasons:
Fewer layers. Every wrapper adds its own system prompt on top of the lab's. Going direct removes one layer of invisible instructions before you have typed a word.
Owning the harness. Firms want to own their skills, workflows and data, not rent them. Several worry about what a platform learns from watching how their lawyers work.
Price. Frontier-model costs are well above what the platforms charge per seat today. Someone is subsidizing the difference. Firms and in-house teams expect that will not last forever.
It seems what holds them back firms from adopting Claude, Gemini, or ChatGPT is governance, not capability. At some firms that have adopted a frontier lab’s models, they still block document uploads to those chatbots, allowing only document uploads to Harvey and Legora instead.
KM Gets the Last Laugh
We have written before that KM finally got some respect once people started calling it "data strategy". It seems that the respect was already paid.
Today, everyone recognizes that every firm has access to the same models, so the model cannot be what sets a firm apart. What sets a firm apart is what it knows. That means its precedents, its negotiation history, and the judgement locked in its partners' heads.
The new KM work is capturing tacit knowledge. Firms are turning partner expertise into "skills" that models can use, and they are going about it in three ways:
Lawyers write it down themselves. Very few are willing to do this.
Structured interviews. An interviewer draws the expertise out.
"Digital twins". While powerful in theory, this has not yet gained traction.
The raw material is a mess. It is extremely hard to distinguish good precedents from bad precedents.
The document management system is a dumping ground, and curating it is the hard problem. Some firms are building data lakes. Some are paying consultants. At least one firm is building a universal knowledge graph. The rest are working out where to begin.
"Agentic" Is Mostly a New Label
Most of what firms call agents today are multi-step workflows that were renamed at some point this year. Genuine agentic autonomy inside law firms is rare. Where it works, it is in research and horizon scanning, where the agent can take whatever route it likes and a wrong turn costs little in terms of risk.
The unsolved problem is not building agents. It is reviewing what they produce. Nobody has found a review step that saves time without amounting to doing the work again. One innovation leader reminded her lawyers that they have to check what the AI left unchanged, not just the markups. Damien Charlotin's tracker of AI hallucination cases has now passed 2,000 entries.
Firms have also built plenty of agents that nobody uses. One firm has more than 500 workflows, of which a handful get regular use. The firms doing this well treat each agent as a small product, with an owner, a requirements template, and a training plan. As someone put it: "A product that is not used might as well not exist."
Nobody Can Measure ROI; Everybody Is Worried About Pricing
"We're struggling to wrap our arms around what the ROI is."
We heard some version of that in nearly every law firm conversation.
For several firms, pricing has overtaken technology as the biggest headache. One AI leader is writing memos to his partners on moving from hourly rates to fixed fees. He has run into the obvious problem: "the entire DNA of the law firm is predicated on hourly rates."
The pressure comes from three directions:
Insourcing. Clients now have the same AI platforms their firms do, and they ask what the firm adds on top.
AI-native firms. They are going after emerging-company and other commoditised work at flat fees, and they turn it around faster.
Mid-tier firms moving up-market. One pitch we heard was “80% of the quality for 50% of the price.”
The big banks are now pushing publicly for savings from their panel firms.
Record profits make change harder.
The junior pipeline cost cutting may be where this will show up. One partner told a friend he has stopped briefing junior associates because Harvey gives him better output. Firm leaders are starting to worry about a “mentorship problem”, and about associates who say they can’t practise without AI. We made this argument in February. The work AI does best is the work that trains junior lawyers.
In-House Is Not One Market
In-house teams are not a smaller version of law firms. Their risk tolerance is higher, and “directionally correct” is often good enough. As one in-house lawyer put it, when she needs precision she pays outside counsel to carry that risk.
Their biggest unsolved problem is not drafting. It is knowing what they have already agreed to. Contract repositories are still unreliable after years of CLM spending. When a data breach happens, which customers need notice within 12 hours and which within 24? Which suppliers agreed to which terms?
The tooling is split in four: buy-side and sell-side, before and after signature, each with its own tools. They are struggling in their own way.
The Money Looks Late-Cycle
Seed and Series A rounds are getting done. Later rounds are hard to raise for anyone outside the two leaders. Acquirers are returning after a quiet summer. The going rate for a solid, growing legal software business is much lower than the 10x founders expect.
Bubble talk was everywhere. Several people expect a correction within 12 to 24 months. Nobody we spoke to expects AI to disappear. What they doubt are the valuations.
So What?
In our recent Vespas vs Ferraris post, we argued that every technology investment is really a bet on where AI's capability lands. What struck us in September is how many firms have not placed a bet at all.
The tools are converging. Everyone has the same models and increasingly the same platforms. What will separate firms is what the tools cannot supply: their own knowledge, their own data, and a clear view of which work AI should not be doing at all.
Clients don't pay law firms for getting it done. They pay law firms for getting it right. AI seems to be making getting it done easier, but getting it right much harder.
