Did you see the news about Claude joining Google Meet earlier this week? Anthropic just announced a new feature called Claude Meetings that will allow its AI assistant to participate in video meetings alongside human participants. According to the report, Claude will be able to follow conversations, retrieve information from connected applications, and begin working on assignments while everyone is still talking.
I’ve been writing a bit lately about AI agents joining our workplace conversations, so this latest announcement got me thinking more about what comes next for the practice of law.
Consider an attorney preparing to take a deposition. He sends out the notice advising everyone that the deposition will be conducted remotely using Google Meet. Nothing unusual about that. But then he sends opposing counsel another email. “By the way, I’m also going to invite my Claude AI agent to attend the deposition with us. Do you have a problem with that?”
The idea isn’t as far-fetched as it might sound. Companies already offer AI tools that listen to depositions, compare testimony against case documents, and suggest follow-up questions in real time. What caught my attention about Claude’s announcement, though, was the possibility of combining those capabilities with an agent that can actually carry out assignments. And because Claude is an AI assistant many people already know and use, I suspect more lawyers will begin experimenting with these possibilities.
Imagine that this attorney has already connected Claude to his case file and instructed it to monitor developments in the litigation and prepare appropriate work for his review. When the agent joins the deposition, it follows the testimony, comparing answers against the pleadings, correspondence, prior depositions, and exhibits. The witness says something that contradicts an earlier email, and the agent identifies the discrepancy and sends the attorney a private message through Teams or Slack, along with the document and suggested follow-up questions.
The agent’s work doesn’t stop with helping him question the witness. When the witness mentions a previously undisclosed meeting, it searches the discovery materials, identifies documents that may be missing, and prepares supplemental discovery requests without being asked. During a break, the attorney reviews the requests, makes a few changes, and authorizes the agent to sign on his behalf and email them to opposing counsel.
As the testimony continues, the witness makes an admission that may substantially change the case. The agent recognizes its potential significance and begins preparing a memorandum explaining how the testimony might affect the litigation strategy, including which arguments have become stronger and what additional evidence might be needed.
The admission might even support a motion for summary judgment. Without being asked, the agent begins drafting one, retrieving the relevant authorities and assembling the supporting exhibits. Much of the motion could be prepared before the deposition ends, leaving the attorney with a draft to review alongside the discovery requests and strategy memorandum.
Weeks later, when the transcript becomes available and any corrections have been considered, the agent updates the motion with the appropriate citations. The attorney reviews the arguments and supporting materials, makes a few changes, and asks the agent to revise several sections. Once satisfied that everything is accurate and ready to file, he tells the agent, “Looks good. Sign it on my behalf and file it.” Assuming the agent has been given the proper credentials, it applies the attorney’s electronic signature and submits the motion through the court’s e-filing system.
I explored some of these possibilities in The Autonomous Law Office back in January 2025. What hadn’t really occurred to me at the time was that an agent would be able to do much of that work while attending the deposition itself, without waiting for the attorney to assign each task.
And opposing counsel might have an agent attending the same deposition. Suppose a question is pending, and that agent privately sends the defending attorney earlier testimony that may contradict the anticipated answer. The attorney can object, request a break, or simply allow the witness to respond. Even the timing of these communications may begin to matter.
I can see enormous possibilities here, particularly in complicated cases involving thousands of documents and multiple witnesses. But there’s also something about the speed that gives me pause. An agent might recognize an important admission that the attorney overlooked, or it might place too much weight on an answer and begin preparing work based on an incomplete understanding of the testimony. I’ve seen enough cases to know that the significance of testimony isn’t always apparent when it’s first given. Sometimes the next answer changes everything.
And as I explored in But I Told My Agent Not To, an agent may do something consistent with its understanding of an assignment even though the person directing it never intended that particular action. When those agents can initiate multiple assignments on their own, a misunderstanding can spread well beyond a single document.
For years, lawyers have taken depositions, reviewed the transcripts afterward, discussed the testimony with colleagues, and decided what to do next. Much of that work may soon begin while the witness is still talking, with the lawyer reviewing and directing work that has already been prepared.
Which brings me back to that email to opposing counsel. “By the way, I’m also going to invite my Claude AI agent to attend the deposition with us. Do you have a problem with that?” I wonder whether the answer changes when opposing counsel realizes the agent isn’t simply coming to listen.
