Editorial illustration for AI Agents Store Team Workflows and Meeting Summaries
AI Agents Now Store Team Workflows and Meetings
Good morning, and welcome to our 6,908 new readers. Tavus just gave people one more reason to second-guess what's on the other end of a video call. The startup's new Griffin model, built on what it calls a "Human Interaction Model," doesn't just generate a face that talks. It watches the screen, listens for pauses, nods at the right moments, and folds details from what it sees into its responses in real time.
That matters because video calls have been the last holdout for AI mimicry. Text can be faked, voice can be cloned, but a live face reacting naturally in the moment has stayed hard to fake convincingly, until now, apparently. Tavus ran Griffin-Lite through a face-to-face study and tracked how many participants assumed they were talking to a real person. The jump from earlier models to this one is the headline number here, and it's not subtle.
Before getting to what that number actually was, here's how Tavus framed the stakes and the comparison to NVIDIA's own benchmark for natural video chat.
Tavus tested Griffin-Lite in a face-to-face study, with 48% of participants believing their conversation partner was human, up from just 2.4% in previous models.
Why this matters
Griffin's half-convinced testers aren't a parlor trick, they're a preview of what's coming to every sales call, support line, and job interview you take over video. For founders building on this, the gap between "looks human" and "is trustworthy" is where the real work starts: disclosure norms, consent for recorded interactions, and ways to flag synthetic participants before someone finds out after the fact. The meeting-summary and workflow-storage use case is the quieter story here, and arguably the more durable one.
Teams archiving institutional knowledge through AI agents, so a new hire can query six months of decisions instead of digging through Slack, is a genuine productivity unlock with none of the uncanny-valley baggage. Both threads point to the same shift: AI is moving from a tool you consciously invoke to infrastructure sitting underneath calls, meetings, and memory itself. Researchers and builders should be asking less "can we make this convincing" and more "who's accountable when it is."
Related Reading
Business & Startups Nordic pilot adds Gemini for Education, NotebookLM to boost AI literacy
Business & Startups Kling launches Video O1, all-in-one model with MVL bridge using transformer
Business & Startups DeepSeek Seeks More Capital Weeks After USD 7B Funding Round
From Archives OpenAI's Altman Says New Agent Is a Premium Product
From Archives Photon Raises USD 4.5M to Replace Mobile Apps with AI Agents
Common Questions Answered
What is Tavus's Griffin model and how does it differ from previous AI video models?
Tavus's Griffin model is built on a "Human Interaction Model" that goes beyond just generating a talking face. It actively watches the screen, listens for pauses, nods at appropriate moments, and incorporates details from what it observes into its responses in real time, making it fundamentally different from earlier AI video generation approaches.
What were the results of Tavus's face-to-face study testing Griffin-Lite?
In Tavus's face-to-face study with 48 participants, 48% believed their conversation partner was human when using Griffin-Lite, a dramatic improvement from just 2.4% in previous models. This significant increase demonstrates how much more convincing the new model is at mimicking human interaction in video calls.
Why does the article emphasize that video calls have been the last holdout for AI mimicry?
Video calls represent the final frontier for AI mimicry because they require real-time interaction, visual cues, and contextual awareness that are much harder to replicate than text-based AI. Unlike text generation, video AI must convincingly manage facial expressions, timing, and responsive listening simultaneously.
What are the key concerns the article raises about Griffin's deployment in real-world scenarios?
The article highlights that the gap between "looks human" and "is trustworthy" requires addressing disclosure norms, obtaining consent for recorded interactions, and implementing ways to flag synthetic participants before users discover they're interacting with AI. These safeguards are essential for sales calls, support lines, and job interviews where deception could have serious consequences.
How could Griffin impact common professional video interactions according to the article?
Griffin is positioned to transform every sales call, support line, and job interview conducted over video by making AI participants increasingly difficult to distinguish from humans. This widespread potential application underscores why establishing trust and transparency protocols is critical before the technology becomes commonplace.
Further Reading
- AI Meeting Summaries: Best Tools, Workflows, and ROI for Leaders - Grain
- AI for work: 15 ways teams use AI to get more done - monday.com
- 13 Best AI Meeting Summarizer Tools in 2026 (Tested for Teams) - Taskade
- Top 8 AI Agents Every Office Team Should Deploy Now - Sana
- Best AI Agent Platforms for Meeting Follow-Ups & Audit Logging - Airtable
