iia/1400 Personal Agents
Personal Agents
- meta
- discussion
- Person/Jordan Tian developer of ZeroClaw/Labs
- so how do you think about safety?
- Person/Uzma Farheen simulate human interactions to see where human and models interact, then creating safeguards and frameworks.
- Person/Gunjan Sinha power in agent swarms is in prediction
- Person/Greg Raiz how many people have personal agent here?
- Person/Niresh Agarwal I can AI/Agent/Personal Consumer Agents
- Person/Gunjan Sinha are there technical problems we need to solve? what's the hard part?
- Person/Niresh Agarwal there are tons of problems, but consumer trust is the hardest.
- Person/Greg Raiz memory, as you give AI/Agents context, that's hard to manage. lots of techniques - trees of knowledge, Person/Andrej Karpathy's coming up with new methods (is he talking about llm-wiki obsidian?)
- Person/Uzma Farheen the quality of data is important, and defining that is important, and having access to that is hard
- Person/Jordan Tian infra problem. right now, it will just go to amazon. things need more APIs or MCPs to make AI/Agents to doing things
- Person/Maria Gorskikh Costof hosting agent on Cloud and cost of Inference
- what would success look like w.r.t. the future
- Person/Maria Gorskikh everyone in the future will have agents, and they will have multiple specialized agents for each part of their life or profession. these will be owned by young StartUps. It might even be geographically specialized.
- Person/Jordan Tian - he did a demo - find me a date tonight and schedule it - scope will grow as apps are tailored to allow them to do things. it can do things more autonomously over time
- Person/Uzma Farheen - for me, an agent which is autonomous where I can be the observer and there is delineation and accountability. there should be difference between steps I did and actions I took and steps my agent did and steps my agent took. that has an in-built concept of AI/Safe/ty in it Good point
- Person/Greg Raiz - Paper/Attention is All You Need is Iron/ic because the result of this has resulted in people getting their attention fractured, not protected.
- Person/Niresh Agarwal - most people think of CursorAI, ReplitAI, coding agents have had a major breakthrough with every software engineer. we need something to come out that every person is talking about it, like what Uber did. it may be one small thing, but it will make people trust it.
- Person/Gunjan Sinha - long term view, what will we do to make the agent better? let's flip it on how will we make the humans better? is it there for agents to solve our ADHD that they created in the first place?
- tbd topic
- Person/Maria Gorskikh I don't trust my Claws to do my Email. but maybe in the future I will
- Person/Jordan Tian when you use apps like social media, once it understands your preferences, it can curate what you want.
- Person/Uzma Farheen let's be realistic - are AI Agents going to augment us or are they just going to engage us? that's the crux of the problem. Good point
- Person/Gunjan Sinha we've got to start to think in 2026 about the genuine human needs. most llms were trained on datasets created over the last 30 years. a lot of companies are more oriented towards ...
- walled garden app stores vs open solutions
- Person/Gunjan Sinha a person a YCombinator Person/Gary Tan just Open Sourced his intelligence
- Person/Maria Gorskikh I don't want to live in a world where everyone is working with Claude/Code or Codex/CLI, ideally, it should be Decentralized, one registry that is a registry of registries so you can discover any type of agent. that's project NANDA
- Person/Jordan Tian - if we get it to run on edge, then it's baked into your OS. If Gemini or Siri can actually do things for you
- Person/Uzma Farheen - right now we are keeping it closed, but to keep children safe, we want to make it open eventually
- Person/Greg Raiz centralizing concext of my life in a standardized way is the thing
- Interest/ing/Point about Context Engineering