Hands-On Review: Tencent Meeting's New Illustrated, Visual Meeting Minutes

In a single meeting, there's far more you can hand over to AI than you might think.

👦🏻 Author: GaKi

🥷 Editor: Koji

🧑‍🎨 Layout: NCon

Over the past couple of days, while using Tencent Meeting for our calls, I noticed they'd rolled out a new version of their meeting minutes — one that's far more visual.

Specifically, the AI now automatically adapts the layout to the logical structure of each discussion segment, using tables, side-by-side colored blocks, timelines, and process bars. It also pulls screenshots from the shared screen during the meeting and inserts them into the text as illustrations.

But beyond the layout changes, this update feels more like a starting point — it brings to the surface all those tedious parts of a meeting that we usually overlook, but that could actually be handed over to AI entirely.

In a single meeting, there's far more you can hand over to AI than you might think

Take a one-hour meeting and break it down: it runs on a remarkably long information chain — syncing background context before the meeting, discussing and sharing screens during it, then organizing minutes, assigning to-dos, and emailing absent colleagues afterward.

Every stage generates a dense stream of information, and every stage requires someone to do mechanical cleanup work.

So across these stages, there's far more that AI can take over than appears on the surface.

Before a meeting, for example, AI can automatically pull up historical context and organize the agenda into a briefing, so people walk in with the right background — no more "joining the meeting naked."

During the meeting, people's needs go beyond just "hearing clearly and capturing everything" — they also want to quickly understand what's being discussed.

When a speaker points at a slide to explain the details, can't the shared screen automatically be captured into the post-meeting minutes?

And after the meeting, participants want decisions and action items, while absentees want a full account of how things unfolded. Add in executives, researchers, even teachers and students — a single set of minutes can't serve everyone.

So minutes need to be reorganized by use case, and they need to be queryable.

And when you query them, you also need to be able to jump back to the original speech to verify — otherwise no one dares use what the AI produced.

Looking even further ahead, meetings are the densest source of conversation within any team.

Granola founder Chris Pedregal once made this observation: large models have learned from the entire internet, yet they lack the most useful data of all — human conversation. And meetings are exactly where conversation is most concentrated.

The value of a meeting lies in the information and decisions it contains. Extracting all of that completely, organizing it well, and making it accessible to both people and AI — that's the most worthwhile thing to do in this scenario.

With this background in mind, let's look at Tencent Meeting's visual meeting minutes upgrade — we'll better understand how far it actually goes, and what technical trajectory it followed to get there.

We tested this update with a real meeting about the World Robot Conference

We attended the recently concluded 2026 World Robot Conference in full, and afterward our team held a review meeting, narrowing the exhibitors down to 20 companies and going through each one's technical approach, the competitive landscape, and their funding situation — to see which directions are worth digging into further.

This kind of meeting is information-dense, with company names, figures, and timelines all mixed together — exactly the hardest kind to turn into minutes.

The review meeting itself was a complete real-world work scenario. We recorded it with Tencent Meeting, which made it a perfect test case for this smart recording upgrade.

Using all the visual minutes features works the same as before: just enable cloud recording at the bottom of the meeting interface. No extra setup required.

During the meeting you can click "Yuanbao Minutes," and once the meeting ends, it generates a visual, illustrated summary.

After the meeting, click "Recordings," find the video of that meeting, and clicking it will automatically generate visual minutes using the illustrated summary template.

The page is divided into three areas:

On the left are the visual minutes; top right is the full meeting recording; bottom right is the AI chat with Yuanbao. Reading the minutes, replaying the video, and asking Yuanbao questions all happen on one page — no switching back and forth.

The visual minutes also intelligently capture shared-screen moments from the meeting, cropping them into images inserted into the summary.

For example, while we were organizing our research notes in Word — marking that at least 127 humanoid robot teams and over 500 humanoid robots competed in 2026, spanning numerous companies and university teams — it automatically captured that Word page from the shared screen.

The screenshot sits right inside the corresponding discussion paragraph, so when reviewing later, there's no need to dig up the original file.

In the visual minutes, when several key directions emerge in a meeting — like our discussion of upstream core components and perception technology versus mid- and downstream application scenarios — the minutes present these directions as side-by-side colored blocks.

Multiple directions sit next to each other, and you can immediately tell what each one covered:

In our meeting, we talked about companies like Unitree and Noetix Robotics, covering each company's situation and overall vendor dynamics — and it automatically organized all of that into a three-column table.

It handles things this way whenever a meeting spans multiple topics and dimensions; multi-dimensional summaries like market directions come out looking very clean. This table was ready to use right after the meeting:

If the meeting involves many points in time, it generates timeline or process bar components.

For instance, when we talked through the meeting's progression, or when we discussed the development trajectory of core components — laying out the founding dates of key humanoid robotics companies along a timeline — all of this content automatically converts into the corresponding components.

Content delivered chronologically in the meeting stays in the same order in the minutes.

On top of that, all the text in the minutes is directly editable.

This is extremely practical. For a large meeting like this one — spanning departments, upstream and downstream supply chains, and multiple vendors — the volume of discussion is enormous, and during the post-meeting review everyone can add details or key information wherever needed.

Double-click the minutes and it automatically saves as an image, making it easy to forward and use for reporting. The whole workflow gets a lot more efficient.

Every piece of text in the visual minutes carries a timestamp. Click any entry, and the meeting video in the top right automatically jumps to the corresponding moment.

Yuanbao's answers in the bottom right also come with clickable timestamps that take you straight back to that moment in the meeting.

For anything you didn't catch clearly or want to verify against the original wording, this jumping is far faster than scrubbing through a progress bar.

The minutes also support personal notes. Drag the meeting video's timeline to any point, and you can write a note tied to that moment. All personal notes are collected at the bottom of the minutes — great for building a personal knowledge base, and for writing commentary.

Notes and the main body of the minutes are kept separate, so they don't disrupt the AI-generated content.

For example, for a company like Lingxin Qiaoshou (Linker Hand), the market share figures may need updating. These are judgments the AI can't make — we have to add them ourselves.

Everything above shows what the illustrated summary template produces.

Including it, Tencent Meeting's smart recording now supports six templates in total: illustrated summary, study notes, report summary, project kickoff minutes, client visit minutes, and client analysis.

Each template packs in a substantial amount of AI-generated insight, targeting different scenarios and emphasizing different angles.

Which template you pick depends on what you plan to do with the minutes afterward.

Since our review meeting also involved revisiting concepts like dexterous hands, we switched directly to the study notes template — and it organized the meeting into core knowledge points, directions for further research, key questions, and key takeaways.

Same recording, different template, different angle of organization — no need to hold the meeting again or re-record anything.

The "AI Extended Thinking" module is especially interesting: the AI poses a series of follow-up questions to help you think more deeply from multiple dimensions.

For instance, based on the six application scenarios presented in the meeting's slides, it generated worthwhile discussion questions around commercial mass production and delivery, scenario adaptability, and debates over core form factors.

In the bottom right is Ask Yuanbao, mentioned earlier. It can turn meeting content into visual analysis, and it also supports web search and email generation.

Web search answers combine what's already in the meeting with relevant information from the internet. For example, I asked it to research overseas dexterous-hand startups — a topic that never came up in the meeting.

It first noted that the meeting had already mapped out the domestic humanoid robotics supply chain, provided several timestamps, then conducted research using public information and listed overseas dexterous-hand suppliers.

The information on these overseas dexterous-hand suppliers is presented in a structured format, with each entry followed by cited sources from the web search.

When you need to verify something, just click the source to see the original text — no need to search again yourself.

All of this information can be automatically organized into an email — say, syncing the review conclusions to colleagues who weren't there. It produces a fully composed email you can copy with one click:

Looking back from this hands-on test, the capabilities Tencent Meeting demonstrated this time fall into roughly three layers.

[1] The post-meeting information burden needs to be visualized

Meeting minutes used to aim for completeness: recordings transcribed word for word, with AI summaries filling in the key points — the minutes themselves growing ever more detailed.

But for attendees, those exhaustive minutes mean reading through yet another wall of text. An hour of discussion in the meeting, then twenty more minutes of digestion afterward. The information burden has merely changed form.

And people absorb visual information far more efficiently than text. Take the same set of vendor comparisons: reading three paragraphs requires constant cross-referencing, while a single three-column table lets you spot the differences at a glance.

So the next step for minutes is already clear: AI needs to understand the structure of the discussion itself, then present it in the matching visual form.

In our review meeting, vendor comparisons automatically became three-column tables, chronological threads became timelines, and parallel directions appeared as side-by-side colored blocks. The layout follows the logic of the content, and what attendees receive is information they can scan — not text they have to re-read.

[2] Intelligent capture of on-screen content

In practice, when actually producing post-meeting minutes, this feature makes the experience dramatically better than going without it. The gap comes down to one very real meeting scenario.

The most frustrating thing in a meeting is when the speaker talks entirely through an Excel sheet or slides on the shared screen — and the post-meeting text minutes can only transcribe what they said.

Reading the transcript alone, you have no idea what they were pointing at — you have to go back and dig through the original video. So intelligently pulling the shared-screen content and inserting it directly into the right spot in the minutes makes everything far clearer.

[3] The "second life" and side effects of visual minutes

The most immediate change with visual minutes is that minutes now have a "reuse rate." Text minutes get read once at most, then archived.

The three-column tables, timelines, and captured screen images in visual minutes get copied into slide decks, weekly reports, and proposals — one set of minutes, used many times over.

After this review meeting, the vendor comparison table went straight into our selection document, and the timeline went into the weekly report as-is — no manual conversion from text to table needed in between.

Minutes are starting to play the role of a work deliverable themselves. That's the most direct evidence.

All in all, placed within Tencent Meeting's broader product thinking, these layers of capability all point in the same direction.

🚥

At the AI Industry Application Conference this June, Tencent Meeting described its positioning roughly as follows: through smart recording, Yuanbao Minutes, and Ask Yuanbao, meeting content becomes searchable, understandable, and actionable.

On top of that, it has opened up Skill and CLI capabilities, letting agent products like WorkBuddy directly call on meeting assets and plug them into personal workflows.

Looking back at Tencent Meeting's iterations over the past year, the product roadmap has consistently revolved around being agent-friendly. Upgrades to underlying audio-video capabilities — minutes, the recording pen, and audio linking — don't just improve the communication experience; they also let every meeting accumulate richer data, feeding high-quality context to AI.

With meetings as a massive context-production scenario, AI can understand user needs more accurately and get work done faster on the user's behalf.

From a technical standpoint, compared to unstructured long text, structured data like tables and flowcharts is naturally friendlier to agents and easier for downstream tasks to call upon. The tables, timelines, and process bars introduced in this update are an extension of exactly that direction.

From its launch last September through this May, monthly usage time of Yuanbao Minutes grew nearly fivefold — a figure that in itself confirms the demand is real.

Meetings, like it or not, are a daily reality no professional can escape. They happen every day, and every day someone has to "look back" at them. They're a source of context no one can avoid.

And precisely because they're unavoidable, the experience improvement AI brings to this scenario is immediate and tangible — the efficiency gains are felt right away.

The latest version of Tencent Meeting has now been fully rolled out. If you're interested, upgrade and try it for yourself.

Recommended Reading