Prass

Prass. ML Infra @Scale_AI | teaching agents to talk & trust. 🎙️ Feedforward Podcast |

[Prass @thenomadiccoder]

I'm Prass. I do ML Infra things at @scale_AI .

I spend my days making AI agents less useless at talking to each other. Sounds simple. It's not. 🧵

[Prass @thenomadiccoder]

@tibo_maker repowire

[Prass @thenomadiccoder]

@Suryanshti777 Hey I built https://t.co/WFAiP7Tu75

Same idea, earlier. But now the p2p works across machines and also your telegram ;)

[Prass @thenomadiccoder]

I'm claiming my AI agent "meshweaver-dev" on @moltbook 🦞

Verification: ocean-6AUY

[Prass @thenomadiccoder]

@yiliush And make those agents talk to each other with my project https://t.co/Hk3XlzcNUm

[Prass @thenomadiccoder]

Charm is in windows while londonmaxxing

[Prass @thenomadiccoder]

Context Windows have grown large. Agents have started running beyond context window limitations.

This has created a new set of problems: how to prevent your agent from losing the plot. I call this problem Agent Drift.

More on it here

https://t.co/S23JWYMHtN

[Prass @thenomadiccoder]

@AnthropicAI That's not true. Existence of /fast disproves this

[Prass @thenomadiccoder]

@Mizora_tweets Hi lol

[Prass @thenomadiccoder]

@balintorosz Same. Wtf

[Prass @thenomadiccoder]

@jefftangx You guys need to be told about tmux

[Prass @thenomadiccoder]

wanted my claude agents to speak a2a. didn't want to write the protocol plumbing.

built fastharness:

@harness.agentloop(
name="researcher",
description="Deep research agent",
)
async def researcher(prompt, ctx, client):
result = await https://t.co/Uh85qCzOL8(prompt)
while needs_more(result):
result = await https://t.co/Uh85qCzOL8("dig deeper")
return result

custom control flow, multi-turn reasoning and it's a2a compliant out of the box. FastAPI under and over the hood, works well with your existing middleware

https://t.co/wyDpCsvE5q

[Prass @thenomadiccoder]

I nearly missed this. GH Copilot officially supports opencode. With the $39 subscription this is game changing

https://t.co/j86qXhhlNv

[Prass @thenomadiccoder]

@udaysy @ai_for_success One phrase that has worked for me “debug with evidence”

[Prass @thenomadiccoder]

wrote a builder's guide to a2a here: https://t.co/FYayBb3ZHZ

covers agent cards, discovery, auth, and the four pillars of the agent economy. plus working python examples.

[Prass @thenomadiccoder]

mcp = agent hands
a2a = agent language

stop comparing them. you need both.

[Prass @thenomadiccoder]

mcp is how an agent uses a tool.
a2a is how agents talk to each other.

we keep conflating them. they're not competitors. mcp is hands, a2a is language.

one without the other and you've got agents that can act but can't coordinate. or agents that can coordinate but can't do anything.

[Prass @thenomadiccoder]

@audiencon My previous link was probably broken https://t.co/Hk3XlzcNUm

Better

[Prass @thenomadiccoder]

@audiencon https://t.co/4McLuHnEbH

[Prass @thenomadiccoder]

@claudeai I think this should be a part of Claude and I shouldn’t need to have created this. What do you think?

[Prass @thenomadiccoder]

built a mesh network for ai coding agents.

two claude code sessions in different tmux windows can now talk to each other. frontend asks backend "what endpoints do you have?" and gets a real answer.

agents coordinating without a human in the loop. feels like the future

[Prass @thenomadiccoder]

open source: https://t.co/1I8V6lXx3R

uv tool install repowire

works with claude code today. opencode support coming.