AI agents with persistent memory

What the claim actually covers

• A running summary of past conversations, which is cheap and degrades as it grows
• A persistent store or memory graph, which is a structural claim rather than a prompt-engineering one

What these connect to

Named by the agents that state this capability. An agent that names none is not necessarily unconnected, it may simply not say.

What people have posted about these

• C

  Crockett Ford, sunbeam Research

  Our research · 2026-10-15

  “Unlike ChatGPT, Claude, or Meta, Instinct doesn't send its own user agent or signature.”

  InstinctRead it

  Nothing Instinct publishes names a search or browser provider, and this is the first outside account of what its traffic looks like to the site on the other end. The consequence sunbeam draws is the one that matters to anyone running a site: Instinct's visits cannot be told apart, so they cannot be counted, allowed or blocked as Instinct's. The method is reproducible, links a site controls and timing it knows, and sunbeam sells search optimisation, which is why it is looking.

• K

  Kush Thakker and Tharun Anandh, Brace

  Our research · 2026-10-01

  “Out of the 456 people who asked for Instinct, only 31 posted about using it in the days that followed.”

  InstinctRead it

  It is the only measurement anyone has published of the hype itself rather than of the product, and it cuts against the account Instinct's founder gives of his own invite system: a compute shortage, not a demand tactic. It measures attention, not use, and says so: it cannot see who got an invite or who used it without posting. The claims about who knew whom and which posts were paid are Brace's, from its own capture, and are not checked here.

• 

  Matt Robb

  @MattRobbt · 2026-09-29

  Update on letting Muse run Facebook Marketplace for me. So I’ve spoken to @dps and the Muse team tonight and we went through the log to figure out what went wrong. Regarding the address leak. The first thing that popped up from Muse when asking it to handle my Facebook marketplace was an option with ‘Allow One Time’ or ‘Allow Always’. I clicked the latter thinking it would still send approvals to accept offers later down the line (it didn’t so be careful). By doing that it granted Muse permission to send messages on my behalf going forwards using a template it put together using information it asked from me. Which also included the pickup address that I did give to Muse (again I didn’t think it would send it out to everyone that gave me an offer so it’s worth checking). I mentioned to the meta team it could be easily overlooked as it was with me and they said they will be making this more clear moving forwards. Regarding the lowball offer situation. It accepted an offer of $600 and sent out my address to a buyer to come collect the item after I strictly set a minimum of $700 with Muse. The Meta team informed me this was an error on their end which caused a display issue removing the ‘7’ from the output to the buyer. And with it stating “Sounds good, 00 it is!” after the buyer asked for $600 it made it look to the buyer like his offer was accepted. I suggested to Meta that each message sent by Muse should have a little “Sent By Muse” badge underneath so everyone knows it wasn’t sent by a human as there was really no way of knowing who sent what in the chat which is very concerning. As for the buyer, we made up and he even came by tonight to buy something else. All ended well. Take care everyone, and double-check what permissions AI has in your life.

  MuseBuying37 likes · 11k viewsRead it

  The approval prompt is where it went wrong, and the user says so: one tap on 'Allow Always' turned an agent that asks into one that sends, to strangers, with his address. Meta's own pages name emails and purchases as the actions Muse confirms first; selling on someone's behalf to people who never agreed to talk to an agent is not on that list. His suggestion afterwards, a 'Sent by Muse' badge on every message, is the third-party disclosure question this directory records and almost nothing answers.

• 

  anu

  @anuatluru · 2026-09-28

  A few thoughts on Instinct (been waiting for consumer AI): - Simplicity is clearly the north star (a prestige product philosophy, Apple-esque) - This is the new interface for commerce, and likely where all their money gets made. - Invite-only consumer growth naturally draws Clubhouse comparisons. It’s a good proxy for the Silicon Valley story arc, but far more solo utility and less social/media network - We’re still deep in unlocking personal superpowers via AI, long phase of tool before network (fits my “Self is the Platform” AI thesis from last yr) - And yet, this could be the first multiplayer consumer AI product to truly scale. ‘Trusted Person Network’ is a great use case and starts with family & close friends, like Life360 (what I’ve called Social Utilities). - Meta is coming, OpenAI will too … and new devices will enter the mix, but hardware is hard and it’ll be hard to beat iPhones anytime soon - Instinct will try to stay free, personality-agnostic, and low gooner / gambling index, which is good for a very mass utility play (also fits the prestige philosophy) - High EQ founder, a good signal for adapting to users, competing, fundraising, and media interviews … - Paradigm novelty is underrated for hiring. Many people want to work on the original new thing more than on copycats, so hiring shouldn’t be hard even at steep valuations, now $10B . - The first tweet from Noah that caught my eye was when he said there are no new interfaces. I 95% agree with that, and have been bullish since …

  Instinct195 likes · 15k viewsRead it

  It is the clearest outside statement of the shape this category is betting on: a long stretch where an agent is a tool for one person, with the network arriving later if at all. This directory can check part of that. {{talksToOtherAgents}} of {{total}} listings state that one person's agent can talk to another person's, which is the whole multiplayer claim, and it has been three for weeks. Nothing here is a fact about the product, and none of it moves a row.

Everything people have got these agents to do · The argument in public

Other jobs people hand over

Most of these agents do several of these, so the same product turns up under more than one.