Straiker announced a kill switch for enterprise AI agents this month. The pitch is simple. If an AI agent starts doing something it shouldn’t, you need a way to stop it immediately, mid-task, without waiting for a support ticket or a vendor callback. For most industries this is a nice-to-have. For law firms, it should be a requirement before you sign anything.

Here’s why. If you’re running AI on intake calls, matter triage, or first-pass document review, that agent is touching client files, privileged communications, and matter data that crosses ethical walls between clients. An agent that pulls the wrong file, mixes up two matters with the same opposing party, or surfaces privileged content to the wrong recipient isn’t a minor bug. It’s a bar complaint waiting to happen. The Straiker news is a useful prompt to ask a question most firms haven’t asked yet: what actually happens when your AI agent goes wrong, and how fast can you shut it down?

The manual work this is really about

Before we get into kill switches, it’s worth being honest about why firms are adopting AI agents in the first place. It’s not curiosity. It’s the math.

Most firms we talk to in the $1M-$25M range are bleeding time in three places. Attorneys lose somewhere in the range of 4 to 6 hours a week to document review, intake admin, and matter housekeeping that never shows up on an invoice. Call it billable-hour leakage. It doesn’t feel dramatic week to week, but multiply it across a partner group and an associate bench over a year and it’s a real number.

Intake is the second leak. A prospective client calls at 7pm or fills out a web form on a Saturday. Nobody picks it up until Monday morning, if then. Industry ranges suggest 30-40% of after-hours intake never converts into a signed client, mostly because the person called three other firms and one of them answered. That’s not a service problem. That’s a staffing-hours problem, and it’s costing you clients who were ready to sign.

The third leak is discovery and first-pass document review. Junior associates spend days working through contracts and document batches at $200-400 an hour of fully loaded cost, flagging clauses and building summaries that a partner then has to re-check anyway. It’s necessary work. It’s also expensive and slow in a way that doesn’t scale with matter volume.

Put those three together and most firms this size are looking at somewhere between $80,000 and $250,000 a year in leakage, depending on headcount and practice mix. That’s the band we usually see once we actually sit down and map the hours.

Why “just add AI” isn’t the answer on its own

The instinct once you see numbers like that is to bolt an AI tool onto intake or document review and call it solved. That’s where the Straiker story becomes relevant again, because bolting on AI without runtime control is how firms create a second, quieter risk to replace the first one.

Think about what a document review agent actually does. It ingests files, reads across a matter, and produces summaries or flags. If that agent’s access isn’t scoped tightly to the matter it’s assigned, and if there’s no way to interrupt it mid-run, you’re trusting a piece of software to self-police privilege boundaries that even trained paralegals get wrong sometimes. An intake agent is similar. It’s on the phone with a caller before you’ve had a chance to conflict-check them. If it books a consultation for someone who’s actually opposing counsel’s client, or if it captures matter details from a caller who should have triggered a conflict flag, you want a way to catch that instantly, not after the fact.

This is the actual argument behind a kill switch model, and it’s the right one. It’s not about distrust of AI generally. It’s about runtime governance. Any agent handling client data needs three things: scoped access so it can only see what it’s supposed to see, logging so you can reconstruct exactly what it did and when, and a hard stop that a human can trigger the moment something looks wrong, without needing engineering support to pull it off.

If your current AI vendor can’t answer clearly how their kill switch works, that’s the conversation to have before the next one, not after.

What this looks like when it’s built right

We build agents for law firms with this exact governance built into how they’re scoped, not bolted on after a scare. Two of them are worth walking through in detail because they map directly onto the pains above.

The Intake Voice Agent answers every call, after-hours, during lunch, on weekends, whenever a human isn’t available. It conflict-checks the caller against your existing matter database before it goes any further, captures the details of the matter, and books a consultation directly into the right partner’s calendar. Critically, it operates inside a scoped data boundary. It can check for conflicts and log intake data. It cannot browse matter files, pull document histories, or access anything outside the intake workflow it’s built for. If a call raises a flag, a partner can pull the plug on that specific conversation instantly, and the transcript and access log are right there for review.

The Matter Triage Agent picks up incoming form submissions and emails, classifies the practice area, scores how well the matter fits your firm’s focus, and routes it to the right partner with a one-paragraph brief attached. This is the agent that keeps good leads from sitting in an inbox for three days. Like the intake agent, its access is scoped to inbound triage data only. It doesn’t get matter file access, and every routing decision it makes is logged with a reason attached, so a partner reviewing it later can see exactly why a lead went where it went.

The Document Review Agent is the one doing the heaviest lifting on the discovery side. It performs first-pass review on contracts, discovery batches, and matter files, flags key clauses, summarizes positions, and produces an associate-grade memo a partner can actually use. This is the agent where scoped access matters most, because it’s touching the most sensitive material. Every document set it reviews is walled to that specific matter. There’s no cross-matter memory bleed, and if you need to shut it down mid-batch because something looks off, that’s a one-click action, not a support ticket.

None of these agents replace judgment. They replace the hours spent on work that doesn’t require judgment, so the judgment work gets a partner’s full attention instead of the leftover 20% of their day.

The dollar reality, one more time

That’s not a hypothetical. It’s the gap between what your team is capable of billing and what actually lands on an invoice, plus the clients who called and never heard back in time. A kill-switch-governed AI setup doesn’t just close that gap faster than a human hiring plan would. It closes it with an audit trail attached, which matters a lot more in a law firm than it does in most other businesses.

Where to start, and what a real audit looks like

If you’re weighing whether to bring AI into intake or document review, or you’ve already got something running and you’re not sure how tightly it’s governed, the fastest way to find out is to actually map it rather than guess. That’s what an Omni Audit does. It’s 60 minutes, no slide deck, and you walk away with three things: a map of where your hours are actually going right now, a dollar estimate of what that’s costing you annually, and a specific recommendation on which agents would close the gap first.

We built the AI audit for law firms around this exact question, because “should we use AI for intake” isn’t the real question most firms need answered. The real question is which specific manual tasks are costing you the most, and what governed, scoped agent setup would actually replace them without creating a new risk in the process.

If you want to see the audit approach in more detail before booking, See Omni for law firms walks through what we look at and how we scope the recommendation to your actual matter mix, not a generic template.

For firms that want to tighten up intake specifically before going further, we put together a practical AI Client Intake Checklist for Law Firms that covers conflict-check timing, after-hours coverage gaps, and the specific data boundaries an intake agent should respect. You can grab the full checklist here and use it as a working document with your intake team this week, whether or not you end up talking to us.

We’ve also written more broadly about how firms are structuring voice-based intake and ops-side automation for matter work, if you want the wider context before narrowing in on your own numbers. And if governance is the piece you’re most focused on right now given the Straiker news, our insights section has more on how runtime controls should factor into any AI vendor decision, not just ours.

The firms getting the most out of this right now aren’t the ones who adopted AI first. They’re the ones who asked the governance questions before they signed, and built agents that a partner can actually stop, review, and trust with client data because the access was scoped correctly from day one.

If you want that mapped out for your firm specifically, with real numbers instead of industry averages, Book my Omni Audit and we’ll walk through it together in an hour.

Sixty minutes. Three outputs. No deck, no pitch, just your numbers on the table. Book a 60-min Omni Audit and see exactly where your firm’s $80K-$250K is going, and what governed AI would look like if you brought it in the right way.