AI News

Anthropic updates Claude Code project features, allowing coordinators to proactively report multi-thread progress

TMTPost · AGIUpdated Nov 16 19:15

AI summary

On November 14, Anthropic made a major update to the Projects feature of Claude Code. The coordinator responsible for orchestrating multi-threaded tasks can now answer progress questions in the project's main conversation and proactively report the progress of each thread. Users can set notification conditions, such as only alerting when a task is completed or encounters an obstacle, and can also have Claude set scheduled reminders; new additions include /btw to open an independent thread, /fork to branch a new line of work from the current thread, and read-only share links. The new version of Projects is gradually rolling out in public beta to subscribers such as Claude Pro and Max, supporting the web, desktop, and the new mobile app. The official notice reminds users that working on multiple threads simultaneously will consume plan quota faster.

Why read this

AI score 68

The original article outlines specific changes such as Claude Code project coordinators proactively reporting progress and thread branching, which can be used to understand the practical use of multi-threaded collaboration.

Topics

Tags

Body · AI translation

Claude is upgrading all the way toward becoming a "Proactive Agent."

According to what Anthropic previously proposed, a so-called proactive Agent means letting Claude gradually shift from waiting for users to issue instructions toward proactively participating in, following up on, and pushing work forward.

On November 14, Anthropic rolled out a major update to the project features in Claude Code.

Now, users no longer need to check one by one how far Claude has gotten. Claude can arrange multiple threads to execute different tasks at the same time, answer users' questions in the project's main conversation, and proactively report the progress of each piece of work.

Before this, Anthropic released Claude Whiteboards, Sheets, and Data, enabling Claude to directly participate in whiteboard, spreadsheet, and data analysis work. Two days later, Claude Meetings officially entered testing, and Claude can even join team video meetings, participating in discussions while producing documents and whiteboards.

From upgrading collaboration tools to participating in video meetings, and then to proactively coordinating multiple threads within projects, Claude is increasingly like a worker who "always has work in sight."

Claude begins proactively reporting on work

Back in September, Anthropic did a major overhaul of the Projects feature in Claude Code.

Users only need to assign a task within a project, and Claude can split the work among multiple independent threads, have them execute simultaneously, and then consolidate the results.

For example, if you ask Claude to build a website, it can assign one thread to handle page design, another to test functionality, and another to organize usage documentation. Several threads can work at the same time without getting in each other's way.

The Claude responsible for coordinating these tasks is called the "Coordinator."

One of the major updates on November 14 was making this "Coordinator" more proactive.

Previously, if you wanted to know the progress of a certain piece of work, you had to open the corresponding thread yourself, but now you can directly ask Claude in the project's main conversation and have it report the progress of different threads.

Even if you don't ask, the Coordinator will proactively report work progress based on your needs.

You can tell Claude under what circumstances you want to receive messages, such as "only notify me when a task is completed or encounters an obstacle." Or, if there are tasks that don't need to be handled for the time being, you can directly ask Claude to set a scheduled reminder, such as "remind me to check this task at 3 p.m. tomorrow." The corresponding thread will temporarily be folded away from the work list and reappear at the specified time, and Claude will also remind you in the project's main conversation.

Beyond the Coordinator, Anthropic also added some small features to Projects to make task management easier.

For example, if Claude is modifying a website and you suddenly think of an unrelated question, you can type /btw to have Claude open a separate thread to handle it, without affecting the original work.

If you're not satisfied with the current approach and want to try another way, you can also type /fork to branch a new line of work off the current thread.

This way, you can try two approaches at the same time and then decide which one to keep.

Now, you can also generate a read-only share link through the thread menu, allowing other logged-in Claude users to directly view this thread—however, threads that have accessed the local computer cannot be shared for the time being. You wouldn't want to leak your privacy, would you?

If you want to show the process of a certain piece of work to colleagues, there's no need to take screenshots or copy the entire chat log anymore.

In addition, Projects has also optimized the results search feature. Recently opened, updated, or pinned result files will be displayed together, and each thread will also show a thumbnail of the corresponding result, making it easy for you to quickly find what you need.

Currently, the new version of Projects is gradually opening public beta to subscribers such as Claude Pro and Max, supporting the web, desktop, and the new mobile app. Users who have not yet received access can apply to join the waitlist.

It is worth noting that when the official team launched the "Coordinator," they reminded users that multiple threads working simultaneously will consume plan quotas faster. The "Coordinator" itself and each execution thread all consume Claude usage quota.

If there are many tasks, you can adjust the model used or reduce the number of threads running simultaneously.

It can even join meetings together

On November 10, Anthropic had just launched Claude Whiteboards, Sheets, and Data, corresponding to whiteboards, spreadsheets, and data analysis, respectively.

When we saw Claude Whiteboards at the time, we were still thinking, isn't this just the whiteboard that exists in almost every meeting room?

As it turned out, two days later, that idea really came true—Claude was moved into online meetings together with that whiteboard.

On November 12, Anthropic released the Claude Meetings beta, gradually opening it to Team and Enterprise subscribers.

Currently, only Google Meet is supported, but the official team said that other video conferencing platforms will also be supported later.

All you need to do is connect Google Calendar in Claude, find the meeting you're about to attend, and click "Add Claude" to invite it to join. You can also set it up in advance so that Claude automatically attends meetings on your schedule.

Once it enters the meeting, Claude appears in Google Meet like any other participant and listens to the discussion. You can call its name directly to ask it questions, or assign it tasks in the meeting chat box.

However, Claude currently mainly replies via text and does not speak up to take part in the conversation directly.

Anthropic provides a fairly detailed use case in its official documentation:

Suppose your team is in a meeting, planning a client event. After discussing freely, you can simply tell Claude: "Help us create an event proposal document."

Claude will draft the document based on the discussion just now, and other participants can edit the content in the document while the meeting continues.

If, when talking about the budget, someone suddenly wants to know how much last year's event cost, Claude can, after being authorized, query the relevant materials through connected tools, send the results to the meeting chat box, and then compare them with this year's budget.

When it comes time to discuss the event schedule, you can also have Claude organize the plan on the whiteboard so everyone can adjust it together.

The whole process requires no pause in the meeting.

Claude also automatically records the meeting content and compiles discussion summaries and to-do items.

After the meeting ends, the relevant records, transcripts, and the documents and whiteboards just created are all saved on Claude's dedicated meeting page. Members within the same organization who were invited to attend can continue to view and edit these outputs.

If, a few days later, you forget what was discussed in the last meeting, you can also just ask Claude to help you review the previous meeting content.

According to the official documentation, Claude can attend a single meeting for up to 8 hours and will remind you in advance before leaving.

If there are people from outside the company in the meeting, whether Claude can attend depends on the administrator's settings. Even if permitted to attend, Claude will not answer questions directly in the Google Meet chat box as it does in internal meetings; it will only listen to the discussion, organize records, and post prompts in the chat box. External participants also cannot access Claude's dedicated meeting page.

In addition, the official team reminds users that because Claude saves meeting text transcripts, summaries, and descriptions of screen images, before inviting Claude to a meeting, you need to ensure that all participants consent to being transcribed.

During the testing period, meeting records, transcripts, and Q&A without invoking external tools are all included in the Team and Enterprise plans. Documents, whiteboards, and other outputs that Claude creates during meetings can be used for free until February 10, 2027.

Proactive Agent

Anthropic had actually revealed long ago that it wanted to make Claude more proactive.

On October 21, when Anthropic launched the Claude Duet research preview, it explicitly positioned it as a "proactive Agent."

It can continuously monitor apps such as email and calendars that users have authorized it to connect to, proactively organize important matters every morning, execute tasks in the background, and proactively send messages when it finds something worth the user's attention.

Anthropic shared a case in its announcement: on a real estate agent's calendar, a viewing date had been recorded incorrectly. After Duet detected the anomaly, it proactively sent her a reminder. The whole process did not even require the user to make a request.

When introducing Duet, Anthropic also specifically emphasized one trait: judgment.

Claude needs to judge for itself which matters are worth the user's attention, which can be set aside for the time being, and when it should proactively contact the user.

The model used by Duet is Opus 5.5, which is already excellent at long-task execution and multi-Agent collaboration.

On November 9, when Anthropic launched Fable 5.2 (and Mythos 5.2), it showed an even more exaggerated case: an engineer asked Fable 5.2 to develop a prototype operating system, and it mobilized more than 1,000 sub-Agents and worked continuously for 8 days, managing to build a system that could boot on a laptop and had a file system and graphical interface.

Although this was just an internal experiment shown officially, the work Claude can coordinate on its own is indeed increasing.

Looking back at Anthropic's updates over the past few days, from whiteboards and spreadsheets to attending meetings with people and proactively reporting project progress, Claude is gradually taking over some work that originally required people to worry about.

According to Anthropic's vision, Claude in the future should increasingly have a sense of what matters: you only need to assign the work, and it can carry it out on its own; it reports proactively when it should, and when something comes up, it knows to come to you.

Source

TMTPost · AGI