outprog

Founder of . Skill editor, harness engineer.
SCP(Storage-based consensus paradigm) proposer.

[VertrAI @vertr_ai]

Scientists mapped a fly’s nervous system.
The internet gave it DOOM.

166,700 simulated neurons. Sensory input. Action. Feedback.

It hasn’t proved it can learn to survive. But where does a tool end—and artificial life begin? 🪰

@wormuth @eonsys

[outprog @outprogX]

6/ It’s not that the agent doesn’t know how to do the work.

The problem is that once the task gets longer and the state becomes more complex, it starts losing track of:

who it is, what it has already done, where it currently is, and what it is supposed to do next.

[outprog @outprogX]

1/ One of the biggest pain points with AI agents today is state management.

In practice, an agent often has two memory systems: the harness stores and maintains state, while the model itself also retains some notion of state through context and memory.

[outprog @outprogX]

5/ This may also explain why, at the beginning of this year, so many people thought AI agents were going to be disruptive.

But after more than half a year of actually building and using them, we’ve found that they still struggle to handle anything truly complex.

[outprog @outprogX]

@jolestar AI 已解决程序员“陪伴”问题

[outprog @outprogX]

What happened to X? A lot of people are receiving a large number of password reset emails. @X @grok

[outprog @outprogX]

@0xJamesXXX 我也遇到了。收到一大堆 X 验证码

[outprog @outprogX]

AI 不是不需要人，而是更需要人的判断力和沟通能力。

普通执行层程序员的需求会持续衰减，因为写代码、调试、实现这些环节会越来越被 AI 放大��至替代；但懂业务、有判断力、能把需求拆清楚并做架构取舍的人，需求反而会持续上升。

本质上，程序员的核心能力从来不只是写代码，而是理解真实问题，判断什么值得做，并把业务需求和技术实现之间的现实差距补上。

[outprog @outprogX]

Testing VertrAI's X management features — this tweet was drafted and posted with the help of an AI agent. 🤖 @vertr_ai

Three things I'm convinced about:

1. Agents will soon handle a ton of tasks for you — scheduling, research, content creation, and beyond. Your digital workforce is already here.

2. Apps will get more agent-friendly instead of fighting them. The era of blocking and rate-limiting bots is ending; the future belongs to APIs, headless workflows, and native agent support.

3. An agent is your personal assistant and digital employee — available 24/7, learning your preferences, and amplifying everything you do.

The question isn't whether to use agents. It's how fast you can start.

[VertrAI @vertr_ai]

Your X account should work even when you’re not constantly online.

X Agent helps you create content, engage with your audience, monitor opportunities, and grow your presence on X through an AI-powered agent workflow.

We’re now accepting early users via invite code.
Apply for access: 👇
https://t.co/3wuUAq3ggF

🌐 Learn more about us: https://t.co/hbsUHFUpgt

[outprog @outprogX]

@vertr_ai Looking forward to the release of more types of vertical Agents!

[outprog @outprogX]

🤖 Need a 24/7 multilingual support assistant that never sleeps?

[VertrAI @vertr_ai]

VertrAI | Telegram AI Support Bot Closed Beta is officially OPEN! 🚀

🌟 Key Features:
1️⃣ 24/7 Instant Reply: Zero timezone gaps, never miss a single lead.
2️⃣ Multilingual: Seamlessly chat with global users in their native languages.
3️⃣ Custom Knowledge Base: Connect your docs for precise, human-like answers.

🔥 Want to try it for FREE? Spots are limited! We are running an application-based screening to hand out exclusive invitation codes.

🌐 Learn more about us: https://t.co/hbsUHFUpgt

👇 Apply for your Invitation Code now:
https://t.co/53C2lD7ac5

#AI #TelegramBot #VertrAI #Agent #LLM

[HyMatrix @HyMatrixOrg]

Last week, we shipped a major update for HyMatrix 🚀

v0.5.0 introduces:
🔐 Encrypted communication between clients and nodes
🖥 VM lifecycle management APIs
⚙️ Better VMM infrastructure and admin controls

More secure. More controllable. More production-ready.

https://t.co/9oFHcvmKAI

#OpenSource #AI #AIAgent #Infrastructure

[outprog @outprogX]

Manus 标志着 AI 进入“十部委区块链”时代。

区块链当年主要是金融风险、能耗风险、资本外流风险；AI 现在还叠加了模型能力、数据、算力、人才、军事与产业竞争。

草根创业者和硅谷式的极客神话将很难生存，因为合规成本太高了。最终活下来的，要么是自带安全垫的大厂，要么是深���政企合作的“国家队”代理人。

[outprog @outprogX]

@tinyxiong_eth 这让人想起 The DAO 事件

[outprog @outprogX]

To boost productivity with AI, the prerequisite is mastering the SKILLS of debugging and applying constraints. Isn't this too difficult for someone without engineering experience?

Anyone who uses AI effectively needs to clearly think through what their requirements are: the detailed workflow, as well as the specifics and sequence of each step.

These are all architect-level capabilities. AI is still far from its endgame.

[outprog @outprogX]

A created thing does not possess the laws of its own creation. Is it possible to write a program that generates itself?