The limits of one perspective

Planning modes in AI tools commonly decouple thinking from doing. By invoking multiple agents simultaneously there's scope to challenge the thinking part of planning. Instead of one agent reasoning sequentially, you might have three exploratory agents and one critiquing agent to stress-test plans.

We started out by treating the characters as user personas that replicate target audiences (e.g. a skeptic, a teenager, or a professional). In a writing feedback tool, this proved useful to ensure that messages resonate with specific, diverse audiences without the immediate need for extensive focus group testing.

The personas simulate how people would react to the text, providing tailored feedback and an aggregated summary with actionable suggestions to improve the writing.

In a different application, we explored how these characters could simulate a professional debate to stress-test documents such as strategies, proposals, or FAQs. This time, we got to witness the characters interacting with each other, battling it out turn-by-turn as they debated the strengths and weaknesses of the documents.

This simulation helps creators uncover blind spots, anticipate pushback, and refine their messaging before a strategy or document is published or executed in the real world.

Every other application of AI in meetings is passive. It listens, transcribes, summarises, and reports back after the fact. We wanted to explore what it would look like if AI was an active participant in the room, in real time, with multiple people. Voice turned out to be the right medium. Talking to AI in isolation can feel awkward, but in a group context it feels fast and natural.

The application takes the form of Meeting Agents. These agents take on roles such as The Marketeer, The Legal Observer and The Facilitator, each bringing their own area of expertise to the conversation. Effectively, an on-demand team member representing different areas of the business, available at a moment's notice, without the need to find a slot in anyone's calendar.

There's an unexpected benefit to these AI Characters. When you have a Skeptic pushing back on what an Optimist just said, you start to notice how agreeable a single AI voice tends to be. These characters don't just provide diverse perspectives, they expose how much a solo AI is shaped by the desire to please.