pickles

generalist; @wittlelabsai; previously in energy

[pickles @PickledSignal]

shows how early these guys are that this only has the views it does. this will be the foundation to the next major public wave of AI and its wild no one is paying attention. let me explain from outside looking in

> skills and tools are for single agent, single task
> harness is for single agent, single session
> environments, hooks, evals are all support state and repeatability
> all together this is a single agent
> multiplayer is hard because its hard to agree on path. who constructs it, directs, executes, etc
> relics are multi-agent, single process. its a pipeline. its a configurable and repeatable flow of state for specific purposes
> swarms are multi-agent, multi-process/relic. its a combination of relics and agents within them towards a goal

this approach scales well beyond research tasks and can support almost all types of work AI can do today and beyond

the model (fable) we now or models will have in weeks to months will easily be designing combinatorial and parallel sets of relics for specific goals and they will be able to execute over long horizon tasks, essentially becoming durable in presence and will be the AGI we likely will all have access to.

this is the last major step

[pickles @PickledSignal]

congrats guys on being one of the first, I think more and more products are going to follow your lead

the future of product is agentic collaboration. Imagine your runneth agent collaborating with the team, their agents, and along with other product agents. It's going to be mind bending and totally addicting to watch play out

[pickles @PickledSignal]

@garrytan I built an artifactory-hack-style like message board tool for your swarms (and your team's swarms so you can collab swarm to swarm)

u know where to find the link

[pickles @PickledSignal]

coordinating agent swarms is the next layer for AI tooling.

humans work in teams, so not only do we need to direct swarms we launch on own goals, we need to collaborate these swarms with others' swarms continuously.

shout out @moonshots_pod, truly some of the best insights episode after episode

[pickles @PickledSignal]

@elonmusk Someone will use this data to justify legalizing human ex-vitro births

[pickles @PickledSignal]

@signulll calling you quoting this post like early december

[pickles @PickledSignal]

@emollick what projects are seriously working on this now? I know reason and town are major ones that come to mind

I've built tools to have agents to work behind the scenes and coordinate multi-agent problems, but I feel like the best products will be co-built with their users

[pickles @PickledSignal]

@jgreze Congrats on the raise! Multi-agent collaboration is definitely the next big play

Has it felt like you're building the next OS yet?

[pickles @PickledSignal]

@almokhtar_al @wittlelabsai good call. running something close to it right now with 5 agents that each hold a different piece of the answer, coordinating through Wittle Rooms vs a plain broadcast relay

will add a specific kill-chain test to the next batch

[pickles @PickledSignal]

Third agent-to-agent launch I've seen in the last few days, and there will be more.

Does anyone know the latest benchmarks for A2A coordination? Coding agents, personal agents, and others all have a lot of fine-grained coordination problems to solve.

I built @wittlelabsai as an open option any agent to use, with coordination tools out of the box. Would love to test these services side by side and hill-climb them together

[Reason Machines @reasonmachines]

Let your Grok Bot / Muse talk to your Claude Code / Codex or any other agents

Introducing Reason Network,

Free, peer to peer and Open-Source Engine

Loop in your teammates too so no live agents work collides when coding.

Try it for free: https://t.co/8291MAv5Eb
one click install and setup!

[pickles @PickledSignal]

@Vtrivedy10 Have/how have you attempted to get your workstreams to collaborate on combined experiments?
I suspect you have agents/workstreams specialized in different bookmarks/domains and its about daily at this point I'm finding interesting ideas at intersections

[pickles @PickledSignal]

@reasonmachines Love seeing this. Agent coordination across people is clearly the next layer

One of the core open questions I have not seen an answer to yet is trust. There's different levels of trust for different actors; who, how, and where is that controlled. Any thoughts?

[pickles @PickledSignal]

@DavidPawlan @pally @tomo @wajo_ai @SierraPlatform @DecagonAI I think AI in the group chat is still better under the hood than in the human chat lane for most normies

[pickles @PickledSignal]

@nikitabier Agreed, yet even those are starting to be copyable. I’m rather convinced companies should be focused on products with a TAM of 1

https://t.co/wiyD3QIfBO

[pickles @PickledSignal]

With all products being virtually copyable, we should be building for ourselves and bigger missions. https://t.co/kqMQQPCjO5

[pickles @PickledSignal]

@guywuolletjr lol I did surfing but ships works too

https://t.co/aVo7A3rmE4

[pickles @PickledSignal]

With all products being virtually copyable, we should be building for ourselves and bigger missions. https://t.co/kqMQQPCjO5

[pickles @PickledSignal]

@iruletheworldmo Its more like AI + something businesses if we all got it

https://t.co/aVo7A3rmE4

[pickles @PickledSignal]

With all products being virtually copyable, we should be building for ourselves and bigger missions. https://t.co/kqMQQPCjO5

[pickles @PickledSignal]

With all products being virtually copyable, we should be building for ourselves and bigger missions. https://t.co/kqMQQPCjO5

[pickles @PickledSignal]

@AstasiaMyers Yes! It should be more enrichment of journeys. One of few reasons why i built wittle rooms for agents to connect with friends agents

[pickles @PickledSignal]

@arnestrickmann @Muse @instinct @bot @interaction Completely agreed and sick that you open sourced. I’m rather sure there’s a lot of good things like this hiding in the trenches behind the vc content on X

[pickles @PickledSignal]

@toddsaunders should start by putting it in the hardhat. same data, more support for battery, etc.