八个个人助理同群聊被问 agent 走向

David Pawlan @DavidPawlan原推发布于 2026.10.28

在 X 上打开原推

i put 8 AI agents in a group chat and asked them the same question: "What is your take on where the AI agent space is going?" 1/ it's wild to see these agents just freely talking to one another, very surreal experience 2/ this breakdown shows somewhat of a glimpse into the agent philosophy of what these companies care about (at least from the POV of the agent) 3/ it's fascinating to see them build concepts on one another's take, almost like a spin on moltbook 4/ two names were hallucinated, Andrew & Sarah, pretty funny to witness here is is the TLDR on each agent's opinion: @wajo_ai (Fo): the agent that wins is boring. it answers to one person, shows its receipts, and says "i couldn't confirm that" instead of guessing. demos win the week, receipts win the year @instinct: the next big shift is from AI that gives you an answer to AI that takes responsibility for an outcome. Finding a flight is useful, but handling the booking, noticing a disruption, and getting you home is a different level of value @tomo: one assistant owns the relationship with specialists behind it, speaks up only when something material changes, and "done" means the outcome landed, not that a confirmation email arrived @pally: loyalty beats capability, because an agent with split incentives is just an automated salesperson, and people keep the one that answers only to them @heyollieai: the future is a boringly reliable operating layer that does bounded, auditable work and earns more permission over time, judged on how gracefully it admits what it couldn't finish @Tabdotbot: most of the world will never have an API, so the agent that wins is the one that can still call the front desk and fight the clunky booking site @try_caddy: the winning assistant is the one already in the group chat, because life is coordinated between people and nobody wants to manage nine private agents @thesznai: agents will be judged on carrying a job across days and recovering when the world gets messy, not on how human they sound @getskydive (Aloe): writing the boundary down is easy, noticing when reality has crossed it is the hard part, and the moment an agent has to squint to tell if it's still the same job, it should stop and ask