CharlieKey takeaways from your favorite podcasts
← 28 October 2026In Depth

The multiplayer bet that could win the agent wars | Jean-Denis Greze (Town Co-founder & CEO)

Jean-Denis Greze argues that multiplayer, work-focused AI assistants are a huge opportunity, and that with copying measured in weeks, startups must stay model-neutral, trusted and fast at shipping.

Watch on YouTube ↗

Charlie's summary

  1. Work, not personal life, is the bigger market for assistants, because spending and product complexity have no ceiling there

    Jean-Denis thinks personal assistant tasks like booking a restaurant are close to solved, so there is a limit to what people will spend. At work, the more AI can generate revenue, the more companies will want to buy. Work is also more complex because you are always dealing with other people. Work and personal data also need different privacy and ownership rules, so he expects people to carry two assistants.

    ▶ 3:35
  2. Assistants took off now because models got good at white-collar work, then at browser and computer use

    Jean-Denis says Anthropic's models from November and December last year could autonomously handle inbox labeling, scheduling and code generation. Around June or July, models got good at browser and computer use, so agents can run tools like Figma. Reliability rose from about 70-80% to 90-95%, and errors became less annoying. Proactive agents that offer to automate repeated tasks are the last differentiator he names.

    ▶ 9:12
  3. Defensibility has collapsed: competitors copy features in weeks, so you get a quarter of clean air, not years

    Jean-Denis says competitors can dissect and relaunch a feature in “weeks to months,” where it once took years. Town felt untouchable after its June launch, then rivals appeared within months. He says you must be right four or five times over 18 months. His comfort is that incumbents lack network effects too, so they look “much less scary,” and startups can catch up with giants.

    ▶ 12:46
  4. Don't bet on a single AI lab: businesses should buy from a model-neutral product layer

    Jean-Denis asks customers who will have the best frontier model in one, two or five years, and nobody can say. Some customers who bet on one lab picked the wrong one for 12 to 15 months and felt burnt. Price matters too, since open-weight models can cost 10 to 20 cents on the dollar. He argues the product layer should route to the best model for the task and price.

    ▶ 15:46

Give your email to unlock the rest of the key points — for every episode.

Do more with Charlie

Free — sign in with your email, no password.