I've never been a big fan of Microsoft's Copilot. For most of its life it has trailed the other AI tools I use. I rarely found it useful for real work. So it has frustrated me that most of the law firms I work with chose Copilot as their AI tool. I understand why. It sits inside the Microsoft environment their IT and security teams already trust. The functionality just never matched that trust.
I haven't tried the new release yet. I'm excited to, because if the early reports hold up, it could change a great deal about how law firm marketing and business development teams work.
On September 25, Microsoft announced a new version of Copilot built to keep working without being asked. The new Copilot app has three parts. Home combines Chat with Cowork, the mode for work you hand off, and puts Word, Excel, and PowerPoint inside the same window. Code lets people who don't write software describe a tracker or a small internal app and have Copilot build it. Autopilot, previously called Scout, is what Microsoft calls a "digital teammate." You give it a role and a goal, and it keeps working in the background, inside Teams and Outlook, without waiting for you to type anything. When Microsoft listed examples of tasks you can delegate to Cowork, the top one was "an RFP response."
This is the first version of Copilot built to do the job on its own rather than answer when you ask. That's the source of my excitement. It's also the source of my worry, because an agent operating on its own needs access to the same sensitive material you do.
A department built on standing jobs
Law firm marketing and BD runs on recurring work. Directory submissions with deadlines that arrive the same weeks every year. Pitch prep that starts with a meeting appearing on a partner's calendar and ends with a scramble for context. Award nominations. Follow-ups with partners who owe you a matter description. The client news monitoring that someone was supposed to own. All of it is relentless. That relentlessness is the reason talented people in our field spend their weeks chasing deadlines rather than deciding which opportunities deserve chasing at all.
Autopilot is designed for exactly that category. Microsoft's featured example is an agent that runs a full supplier review: it builds the schedule and the workback plan, then handles prep and follow-ups on its own, "right down to reaching out to stakeholders for updates." Apply that to a Chambers submission, and you have an agent that builds the timeline and follows up with partners for their matter descriptions until the submission goes in.
Ryan Sloan, reporting from the launch for Every, described how Charles Lamanna, Microsoft's executive vice president for Copilot, uses his own Autopilot, which he named Apollo. Lamanna gave it one goal: make sure he's prepared for customer meetings. It watches his calendar around the clock, finds those meetings, and pulls the relevant context from Teams and email before he walks in. That is a pitch-prep assistant. It's the thing BD teams have wanted to give every rainmaker for twenty years and could never staff.
I think about this kind of agent the way I think about the load-in crew at a concert. By the time the band walks on at nine, the crew has been at it since noon, the stage built and the sound checked. Nobody in the crowd sees any of it, and there's no show without it. For as long as I've been in this business, BD teams have been the crew and the band on the same night. An agent that handles load-in means the people who know the audience get to spend more of the day on the stage.
Code covers a different kind of standing work: the spreadsheet that should have become an app years ago. You probably have one. The pitch tracker, the directory calendar, the running list of which partners have spoken where. Code puts building that app in the hands of the person who already maintains the spreadsheet and hosts it inside the firm's own tenant rather than on some tool nobody in IT approved.
The permissions that protect you can also stall you
The caution starts with a permission wall Sloan hit on the demo floor.
He asked the new Copilot to pull a table from a Word document into a new tab in an Excel workbook. Copilot found both files, then told him it lacked permission to edit the workbook. A handoff from Chat to Code hit a similar wall. Microsoft's team assured him both would work "in the right configuration." Sloan's conclusion was that the controls that make Copilot trustworthy inside a company are the same ones that can keep it from finishing a routine task.
Now picture a law firm's configuration. Ethical walls and matter-centric document security, sitting on top of confidentiality obligations that IT and the general counsel's office take personally, as they should. Marketing and BD hold unusually broad access inside that structure. You touch pitch materials and client lists across every practice group, often more of them than any single attorney sees. Sloan reports that Autopilot draws on the last 30 days of your work, your place in the org chart, and your calendar, and that it can access most of what you can.
So the configuration can go wrong in either direction. IT scopes the agent tightly enough that it can't finish the directory submission because it can't open the matter list. Or IT scopes it loosely enough that an agent with a friendly name is reaching into material on the far side of a wall. Satya Nadella said as much at the launch: "The new massive insider risk is the AI agent."
The answer is for marketing and BD to be in the room when Autopilot gets configured, instead of finding out on day one what IT decided the job was. In my experience, MBD is rarely the first call when a firm sets AI policy. With this release, it needs to be an early one, because nobody else in the building knows what the agent would be asked to do.
There's also the question of whose name is on the follow-up. Autopilot appears in Teams under a name you give it (Sloan's example was "Dot (Ryan's Autopilot)"). Colleagues can tag it like anyone else. Microsoft's own demo has the agent reaching out to stakeholders. In our world, the stakeholders are partners. I would think hard before letting an agent nudge a practice group chair about a late bio at 7 a.m. The goodwill BD people spend years building with partners is not something to hand to an agent in its first week on the job.
Somebody is going to ask what it cost
Microsoft split Copilot pricing in two. Chat and the Office apps stay on the per-user subscription your firm already pays for. Cowork, Code, and Autopilot run on usage-based billing, credits consumed as the agent works. Autopilot is off by default. Lamanna told Sloan it has to "clear a higher bar" than a tool that only works when you ask, because it's spending credits around the clock.
That's a fair standard. It also means that the first time the firm's finance team sees a usage bill generated by marketing's agents, someone will ask what the firm got for it. "It felt faster" will not survive that conversation.
Measure the job before the agent does it
The opportunity for MBD is to define and measure the work before an agent is set loose on it.
Microsoft spent a surprising share of the launch on evaluations. Nadella called for "evals specific to your company," tests built from the tasks your people do. Lamanna said Microsoft runs about 16,000 internal evals. He also described what successful customers do: the person who does the job every day builds the eval. Before the agent touches anything, the team measures how humans do the job today, tracking outcomes like error rates or the time it takes to get a contract out the door. That gives the agent a real baseline to beat. When a test fails, Lamanna's advice was to "embrace the red," because the failure shows you what to fix.
Could your department say, with a number, how long an RFP response takes from intake to submission? How often a directory submission goes in with a stale matter? If the honest answer is a guess, that's the problem. It's also the opening, because the person who can measure those jobs and write their evals is you. IT can set permissions, but the test for what a good pitch looks like at your firm takes your expertise. Once the agent is running on credits around the clock, your eval is what decides whether to keep it on.
Autopilot is expanding to private preview at the end of September. Home and Code are rolling out through Microsoft's Frontier early-access program first. Your firm probably won't have any of it switched on for a while. Pick one standing job now. Write its job description as if you were hiring a person to do it. Time how long it takes today and note where it breaks. That document becomes both your first eval and your configuration request to IT. You'll have it ready before anyone asks.
Where Im coming from
I am a Cofounder at InnovAItion Partners, an AI consultancy, and most of the workflows we build for clients run on platforms other than Copilot. Weigh my caution with that in mind, and weigh my excitement the same way, since it comes from a longtime skeptic.
For marketing and BD teams at Copilot firms, this is the first version that looks like it could run a standing job, an RFP response or a directory submission, from start to finish. That's exciting, as long as the people who do the work help scope its permissions and write the test it has to pass.
I still haven't tried it, and I'm looking forward to it. For years, the missing piece at Copilot firms was the functionality. If the reports hold up, the missing piece becomes the job description, and that's one marketing and BD are better placed to write than anyone else in the building.