AI just handed every lawyer 240 hours a year. The firms that pull ahead will be the ones who can trust the output.

The 2026 Thomson Reuters AI in Professional Services report — its fourth annual, drawing on more than 1,500 professionals across legal, tax, and accounting — landed with a number worth sitting up for: AI now frees the average lawyer roughly 240 hours a year, about five hours every week, and more than 80% of organizations use generative AI weekly. Adoption didn’t creep. It nearly doubled in twelve months — GenAI use at law firms went from 28% to 41%, and in corporate legal departments from 23% to 47%.

That is the good news, and it’s genuinely good. Work that used to eat a junior associate’s evenings — first-pass contract review, pulling authorities, drafting the routine memo — is now the kind of thing a well-built system does in the time it takes to get coffee. A capability that a few years ago required a bank’s budget is now sitting inside off-the-shelf tools.

It’s worth remembering how far that is. Back in 2017, JPMorgan’s COIN contract-intelligence system became the headline example of AI in law precisely because it was exotic — it reviewed commercial-loan agreements at a scale reported as the equivalent of 360,000 lawyer-hours a year, and doing it took a bank with a research lab. In 2026, a two-partner firm can buy a version of that on a monthly subscription. The frontier didn’t just advance. It came down off the mountain.

So if the capability is settled, here’s the number that actually decides who wins with it: only 17% of legal professionals say they’re ethically comfortable letting AI provide legal advice, and only 18% of firms measure the return on their AI at all — with roughly 40% unsure whether anyone is tracking it.

That gap is the whole opportunity.

The question changed while everyone was busy adopting

For three years the competitive question in professional services was are you using AI yet? As of this report, that question is closed — nearly everyone is, weekly. Being an AI adopter is no longer a differentiator; it’s table stakes, like having a website.

The question that replaces it is quieter and much harder: can you trust what it produced, and can you prove it to a client, a partner, or a regulator? A tool that drafts a beautiful memo in ninety seconds is worthless — worse than worthless — if a partner still has to read every line to catch the one confidently-wrong citation. The 240 hours only become real hours saved when someone stops checking, and no one stops checking a system they can’t trust.

That’s why the 17% trust number and the 240-hours number belong in the same sentence. The savings are gated entirely by the trust. And the trust is gated by something almost no one is doing yet: measuring whether the thing is actually right.

What separates the firms pulling ahead

The firms turning AI from a demo into billable leverage aren’t the ones with a secret model. Everyone has access to the same models. They’re the ones who built the layer that makes the output trustworthy enough to rely on without re-checking — and who can show the number. In practice that’s a short list:

• Ground every answer in authoritative sources, not the model’s memory. The system should answer from the actual statute, contract, or filing — retrieved and shown — not from whatever the model half-remembers from training. In our own legal-RAG demo, the model only ever answers over a vetted corpus; it is structurally unable to “recall” something that isn’t in front of it.
• Cite everything, and verify the citations before the answer ships. Every claim should carry a pointer to the exact source passage, and those pointers should be checked automatically — not trusted because they look plausible. If a citation doesn’t map to something real, the answer is flagged, not sent.
• Let it say “I don’t know.” The single most trust-building thing a legal AI can do is abstain when the corpus doesn’t support an answer, instead of inventing a confident one. A system that refuses gracefully is one a partner can actually stop double-checking.
• Measure accuracy on your work, with a number you can show. This is the 82% of the game almost no one is playing. A scored test set — real questions with known-good answers, graded on precision and faithfulness — turns “the AI seems good” into “the AI is 94% accurate on our contract questions, and here’s the evidence.” That number is what earns the right to save the 240 hours.

None of this is exotic, and none of it is on the model card. It’s the operational layer around the model — and it’s precisely what converts “we use AI” into “we trust AI, and we can prove it to you.”

What to do this quarter

Three moves, in order of leverage:

1. Pick one high-volume, well-bounded task and make it trustworthy end to end. Not “roll out AI everywhere” — one workflow (say, NDA review or a specific research pattern) where you ground it, cite it, and measure it until a partner will sign off without re-reading. One trusted workflow beats ten impressive demos.
2. Stand up a small scored test set for that task. Even 30–50 real questions with known-good answers is enough to replace opinion with a number — and to catch a regression the day a tool updates instead of the day a client does.
3. Put the trust story in front of clients. Fewer than 20% of clients currently mandate AI — which means being the firm that can show how its AI is grounded, cited, and measured isn’t a compliance chore, it’s a pitch. Trust, provable, is a selling point almost no one is using yet.

The bigger picture

The most encouraging line in the whole report is the one hiding in plain sight: the technology is no longer the hard part. The models are here, the adoption is here, the 240 hours are real and sitting on the table. What’s scarce — what the data says four out of five firms haven’t built yet — is the discipline to make the output trustworthy and the willingness to measure it.

That’s a wonderful problem to have, because scarcity is where advantage lives. When everyone has the same capability, the edge goes to whoever can be trusted with it. The firms that win the next two years in legal won’t be the ones who adopted AI first. They’ll be the ones who can look a client in the eye and say here’s what our AI did, here’s the source for every line, and here’s the number that proves it’s right.

The capability came down off the mountain. The trust is still yours to build — and right now, it’s the most valuable thing in the room.

---

Sources: Thomson Reuters Institute, “2026 AI in Professional Services Report” (fourth annual; 1,500+ professionals) and “AI adoption has hit critical mass,” 2026; Thomson Reuters Legal, “Highlights from the 2026 AI in Professional Services Report and what it means for legal teams,” 2026. JPMorgan COIN figures as originally reported, 2017.