The Only Associate Who Loves Redlines: AI’s Big Day at the Firm
At 8:59 on Monday morning, the litigation floor smelled like optimism and burnt espresso. By 9:00, optimism had left the building. A client had just sent over a politely frantic note: an internal investigation had uncovered 10,000 emails and 400 supplier contracts potentially relevant to a regulatory inquiry, and could we please summarize the landscape by Friday “so our Board may sleep”? The partner, Margaret—who believes deeply in precedent and single-spaced memos—looked at the pile, at the calendar, and then at Leo, the associate whose bravest professional act to date was liking a Supreme Court quote on LinkedIn.
“We’ll use the new platform,” Leo said, voice steady in the way of people who haven’t yet read the documents. The new platform was an AI toolkit the firm had onboarded: eDiscovery that did more than sort by date, research that could cite actual law, and a contract assistant that extracted clauses with the dedication of a genealogist at a family reunion. Margaret raised an eyebrow. “We’re not inviting a robot to client meetings,” she said. “But it may review documents—quietly.”
Quietly turned out to be fast. By lunch, the AI had chewed through the email universe using a technology formerly known as predictive coding and now often referred to—more diplomatically—as active learning. It identified pertinent threads around something mysteriously called “Project Bluebird,” flagged a cluster of messages involving a non-standard indemnity promise, and politely suggested that the exchange titled “Subject: Taco Tuesday” was unlikely to be responsive. It also marked likely privileged communications for human review, avoiding the professional nightmare where one congratulates opposing counsel on their delightful access to one’s internal advice.
On the contracts side, the AI read through supplier agreements with the patience of a saint and none of the judgment. Using the firm’s playbook—actual, written-down preferences about caps, baskets, termination rights, and dispute resolution—it extracted and normalized key clauses across the portfolio. It produced a dashboard no one knew they wanted: the number of contracts with “termination for convenience” (17 more than expected), the outliers with indefinite auto-renewals, and a handful attempting to define force majeure as “any situation in which we are not in the mood.” It generated redlines against the firm’s standard positions that could reasonably be called helpful, and unreasonably be called enthusiastic.
By afternoon, the client called for an update. Margaret briefed them with heroic calm. The AI, meanwhile, politely sat in the background not being invited to speak, but it did live-summarize the meeting and created an action list that didn’t miss a single “we’ll circle back.” Leo then asked the research assistant to pull a primer on how force majeure clauses have been treated in the jurisdictions relevant to this client’s facilities—citing recent cases, not nostalgic ones. The system returned a neatly footnoted mini-memo, complete with jurisdictional nuances and the kind of “notwithstanding” usage that had clearly been to law school. Leo checked the citations (trust, but verify), and found them pleasantly real.
Because compliance waits for no one, the team used the platform’s policy analyzer to map the contracts against the client’s internal code of conduct and new regulatory guidance. It identified where notice obligations were missing, surfaced anti-corruption language that was surprisingly shy, and suggested a standardized rider for high-risk jurisdictions. Crucially, it did not issue legal advice to the client; it equipped the lawyers to do so faster, with color-coded confidence levels and links to source documents they could actually open.
At 5:00 p.m., a printer jam threatened to derail everything—a reminder that the future may be intelligent, but the paper tray remains analog. The AI continued working, unbothered by hardware. Overnight, it prioritized deposition candidates based on communication patterns, located a forgotten side letter hiding in an attachment, and suggested a more defensible search term strategy that did not rely on guessing what “Bluebird” might rhyme with. It also performed a sanity check for privilege using the firm’s own names and domains rather than “all emails with the word counsel,” which is the eDiscovery equivalent of dragging a net through the ocean and being surprised at the octopus.
By Wednesday, the tone had shifted. Margaret began sentences with, “Let’s ask the system to…” and ended them with, “…and then we’ll confirm.” The client received a concise interim report: key issues, likely exposure, recommended negotiations for contract remediation, and a timeline that measured in days rather than dog years. The AI had drafted a clean summary addendum with tracked changes aligned to the client’s playbook, and Leo adjusted phrasing in that mysterious dialect called Partnerese—adding just enough “hereto” to feel complete.
There were, of course, moments when the AI tried a little too hard. It once highlighted “promise” in a Christmas card as a potential indemnity. It prioritized “Bluebird” but required a human to understand that a reference to “our feathered friends” was, in fact, code. But these were the kinds of errors that improve with tuning, and more importantly, they occurred in a workflow designed for human oversight. The firm used a secure, enterprise deployment, kept confidential data inside a walled garden, logged every decision, and made sure the only hallucinations were of the caffeinated variety.
By Friday, the Board slept. The lawyers did, too—well, eventually. The deliverable was solid: a defensible review protocol, a contracts heat map, a set of negotiation-ready redlines, and a research appendix that wouldn’t embarrass anyone in front of a judge. The cost came in under budget, because the machine did what machines do well—pattern recognition, summarization, extraction—while the lawyers did what lawyers do brilliantly: judgment, strategy, and the delicate art of saying “no” in a way that sounds like “yes, if.”
The moral, beyond “replace the printer,” is simple. AI in law works best when it’s treated not as a partner replacement, nor as a magic wand, but as the only associate who loves redlines, never sleeps, and doesn’t bill for thinking. It thrives on clear instructions, firm playbooks, and measured skepticism. It invites lawyers to spend less time spelunking in inboxes and more time advising clients. And it delivers something surprisingly radical in our profession: time returned to judgment.
If The Future of Law were a closing argument, it would sound like this: Your Honor, the facts show that when supervised and properly deployed, AI reduces risk, accelerates work, and elevates the uniquely human parts of legal practice. We rest—briefly. The AI will keep reviewing.
