Legal teams are the bottleneck. They don't have to be the villain.

Every company has a queue everyone complains about, and at a remarkable number of them, the queue is legal. Sales says legal kills deal momentum. Marketing says legal waters everything down and takes a week to do it. Product says legal is where launches go to age. Legal, meanwhile, is a team of four serving a company of four hundred, drowning in NDA redlines and "quick questions" that are neither, and privately certain that everyone complaining would be the first to sue-proof their own decisions if they spent one day in the seat.

Both sides are right, which is what makes this interesting. The complainers are right that legal is the bottleneck. Legal is right that the bottleneck exists because the company routes four hundred people's risk questions through four people's judgment, and then acts surprised by the queue. The villain here was never the team. It's the topology.

Look at what's actually in the queue, because the composition is the whole story. A general counsel who audits the intake honestly finds something like this: a large majority of items are variations on questions that have been answered before. The NDA with the standard terms. The contract redline where the counterparty touched the same three clauses counterparties always touch. The marketing claim that's fine because it's substantively identical to the claim approved in March. The "can we say this" question whose answer is in a policy doc nobody reads. Then, buried in that pile, a small number of items that genuinely require a lawyer: the novel deal structure, the ambiguous regulatory question, the clause that looks standard and isn't. The queue's tragedy is that these two categories wait in the same line, and the routine items, by sheer volume, set the wait time for everything, including the item that's actually dangerous.

The standard fixes have all been tried. Self-serve playbooks, which nobody reads under deadline. Contract templates, which work until the counterparty edits them. Training sessions, whose half-life is about three weeks. Hiring, which is expensive and just moves the ratio from 400:4 to 400:5. Every fix fails the same way, because every fix still ultimately depends on either a human lawyer's attention or a non-lawyer's willingness to correctly self-diagnose a legal question, and the second thing is famously the thing non-lawyers cannot do. The salesperson doesn't know their weird clause is weird. That's the point of lawyers.

What's changing now is that the routine majority of the queue can be handled where it originates, before it ever becomes a ticket. An agent that has absorbed the company's playbook, its precedent, its past approvals and rejections, can sit in the flow of work and deal with the recurring shapes on sight. The standard NDA gets reviewed against the playbook in minutes, with the two non-standard clauses highlighted and everything else confirmed as within policy. The marketing claim gets checked against what legal has previously approved, with the answer linked to the precedent. The salesperson's "quick question" gets an actual quick answer when it truly is one.

And, critically, when it isn't one, the agent's job flips: it escalates to a human lawyer, with the analysis already assembled. This clause deviates from the playbook in this way, here are the three past deals where something similar appeared, here's what was decided then. The lawyer starts from a briefing instead of a blank contract. Some legal teams run this in chains, one agent doing first-pass review, another checking its work against the newest policy before anything reaches an attorney, and the attorney's queue transforms from four hundred undifferentiated tickets into a short list of genuinely hard questions, each arriving pre-investigated.

Now, the sentence every lawyer reading this has been waiting to deploy: an AI's confident wrong answer about a contract is worse than a slow right one. Correct, and the objection deserves a real response rather than a dismissal, because it's the response that separates the teams doing this well from the teams generating malpractice anecdotes. The design principle is that the agent's authority is bounded by precedent, not by its own confidence. It handles what the playbook and past decisions clearly cover. Anything novel, ambiguous, or high-stakes doesn't get a creative answer; it gets escalated, and the escalation threshold is set by the lawyers, conservatively at first, loosened only as evidence accumulates, exactly the way a GC extends trust to a new associate. The agent isn't replacing legal judgment. It's replacing the queue in front of legal judgment. The judgment stays human, and gets to spend itself on questions worthy of it.

The part that should actually excite legal leaders, though, isn't the shrinking queue. It's what the queue was hiding. Every GC knows the uncomfortable truth that being the bottleneck comes with a shadow: the risk you never see, because people learned to route around you. The deal terms that never got reviewed because "legal takes too long" and the quarter was ending. The marketing claims that shipped unchecked. Self-help legal work by non-lawyers is the most dangerous legal work in any company, and it's a direct product of queue length. When routine review becomes fast and ambient, the routing-around stops, and legal's actual coverage of the company's risk surface goes up precisely as its ticket count goes down. Fewer tickets, more control. That trade doesn't come along often.

There's a role shift buried in this for the profession, and the in-house lawyers who see it early will define it. The job stops being "person who reviews things" and becomes "person who designs how the company handles risk": writing the playbooks that agents enforce, deciding the escalation thresholds, auditing the boundary between what's routine and what deserves an attorney, and spending recovered hours on the strategic work that four-lawyer teams never reach, the regulatory horizon, the contract terms that keep causing disputes, the counsel-as-advisor function that the queue ate years ago.

The bottleneck era gave legal teams a strange deal: maximum blame, minimum leverage. The next era offers the reverse, and the only price of admission is letting the routine eighty percent go, with guardrails, to something that never gets tired of NDAs. Most lawyers, offered that trade in those words, sign quickly.