A task that used to eat an associate's week now takes something closer to a long lunch. AI-generated case summaries and statute research have compressed the work that hard, and clients are already noticing on their invoices. A first-year who once spent ten days pulling authorities on a novel evidentiary question can hand a partner a cited draft digest before the second coffee. Partners feel it. Clients feel it too, and they've started asking why a memo that used to take a fortnight now shows up on the bill as three hours.
Most of the conversation about these tools is built on assumptions that no longer hold. Some are hangovers from the early hallucination scandals. Others are wishful thinking from vendors. A few are the polite fictions firms tell themselves while the market shifts underneath them.
Here's a teardown of the ones doing the most damage.
Myth: The Time Savings Are Marketing Hype
The compression is real, and it's larger than most skeptics expect. Thomson Reuters estimates that AI-assisted legal research can pull an average litigation matter down from roughly 17 to 28 hours of research time to something between 3 and 5.5 hours. That isn't a marginal improvement. It's a different order of magnitude, and it lands hardest on the work associates used to be handed to prove they could handle volume: pulling cases on a discrete issue, chasing a statute through its amendments, digesting a fifty-page opinion into something a partner can skim.
Anyone who has watched a competent associate use a modern research tool for a week has seen this firsthand. The pushback usually isn't about whether the hours drop. It's about whether the output is trustworthy enough to stake a filing on.
Myth: The Hallucination Problem Was Solved by Better Prompts
Prompt discipline helps at the margins. It does not fix a tool that generates text without grounding it in a retrieved document. The tools that have meaningfully reduced fabrication rates did it by changing the architecture: pulling from a closed, authenticated corpus of case law and statutes, forcing the model to cite as it drafts, and rejecting outputs whose citations don't resolve. That's an engineering choice, not a prompt trick.
The buying implication is straightforward. When a vendor demos a summary, click the citations. If every case, pinpoint, and statutory subsection resolves to the actual source in one click, the tool is worth a deeper look. If a citation opens a search box instead of an opinion, the reviewer is doing the retrieval the tool was supposed to do.
The ABA's Formal Opinion 512 on generative AI is worth a careful read here. It frames verification, confidentiality, and candid client communication as continuing duties, not one-time procurement questions.
Myth: Associates Will Just Absorb the New Workflow
They won't, and pretending otherwise is how firms end up with expensive seats and mediocre output. Reviewing an AI-generated summary is a different skill from writing one. It rewards a specific kind of skepticism: reading for what the model left out, checking whether a favorable-sounding case was distinguished on facts the summary skipped, catching a statute's operative clause paraphrased into something subtly broader.
Firms that get this right build the review protocol into the workflow itself: which categories of summary get spot-checked, which get full verification, who signs off before a passage moves into a brief. Platforms designed for this, for example, Law.co's AI case summary and statute tooling case summary and statute tooling, are built around attorney approval gates and audit-ready outputs because the review step is the product, not an afterthought.
Myth: Clients Don't Notice What's Under the Hood
General counsel notice. Sophisticated buyers noticed first, and mid-market clients are catching up faster than most firms assume. The tell is in the questions showing up on RFPs and engagement calls: how the firm uses AI for research, what stays inside the firm's environment, who verifies the output, how the bill reflects the compression. Two-week research memos priced as two-week research memos are getting flagged.
That isn't bad news for firms, necessarily. Clients aren't uniformly demanding rock-bottom pricing. They want an honest conversation about where the time went, what the tool did, and what the lawyer added on top.
Firms that can answer those three questions cleanly tend to keep the work. Firms that can't tend to lose it on price they didn't need to compete on.
The compression from two weeks to two hours is genuine. The firms that turn it into durable advantage stop arguing about whether it's happening and start being honest with clients about what to do with the time they just got back.