Agency: AI Agent Proposes Tasks Instead of Waiting for Prompts

X postMagnus MüllerTool

Magnus Müller has introduced Agency, an approach that reverses traditional prompting: instead of waiting for user instructions, the agent proactively identifies pending work for user approval, using a me.md file as a personal context base.

Key points

  1. Inverted workflow: The AI independently finds tasks and presents them in a card-like approval feed for the user.
  2. Context file me.md: Personal priorities and life context are stored in a central markdown document.
  3. Compatibility: Designed primarily as a skill that can be executed within agents like Codex or Hermes Agent.
  4. Visual verification: Uses a browser harness to capture before-and-after screenshots, making task review straightforward.
  5. Tooling recommendations: Müller recommends tools such as browser-use CLI 3.0 for web-based actions.

Why it matters

Most AI tooling relies on manual prompting and context switching. Shifting agents towards proactive task identification with a lightweight approval gate could significantly reduce time spent monitoring communication platforms like Slack or email.

Analysis

While the card-based approval flow addresses notification fatigue, proactive execution in web environments remains sensitive to authentication hurdles and dynamic DOM changes, making robust verification essential.

Original post

Magnus Müller

@mamagnus00 · 12 September 2026

Introducing Agency. Never prompt again. AI prompts you 🫡 Get 100x more done. Your agent finds useful work. You approve. > full context of your life with me.md > catches things you forgot > fixes issues before you see them > never open Gmail, Chrome or Slack again Runs

707 likes28 replies2.3k bookmarks227.7k views

View on X
More posts in the thread (10)
  1. @cystalphantom Yes did you try
  2. @Connerbuild tell your agent in Codex it should set up a schedule to create new tickets every 4 hours and work with subagents on the queued ones. I will also add this to the skill
  3. @amentiflow It's really insane because the search problem of context switching is gone so it's pure flow to make decisions
  4. @OnFinality Try browser-use CLI 3.0
  5. @heyellieday Would love some feedback
  6. @aiseomastery I'm really thinking I don't read emails anymore because I know agency takes care You don't have the fear of missing something anymore
  7. @heyhenrysnow Yes exactly With this skill I'm also using codex
  8. @shahin43 This repository is mainly a skill which explains to the agent how to create this Tinder-like process. You can do this in Hermes agent or codex
  9. @ambroselli_io It takes many of the small tasks, which are not really creative. You still put in your dreams and your intents, which can be extremely creative, but instead of prompting 20 sessions you just put your dreams all into one
  10. @Blacktrace_ I tell you this is so useful with agency because it has browser-harness. It takes screenshots before and after, which makes it super simple for you to approve
Selected replies (5)
  • @cystalphantom @mamagnus00 Running locally and open source makes storing personal context in me.md a lot easier to trust.
  • @Blacktrace_ @mamagnus00 cool. I’ve got the opposite problem: too much agency, not enough judgment. find useful work is easy. knowing it’s merge-ready is the job.
  • @shahin43 @mamagnus00 Making it OSS and hosting this locally will help users to keep proprietary sata safe. But how is it different from self hosting Hermes agents ?
  • @Connerbuild @mamagnus00 I'm trying this right now. It successfully set up and it found some great "cards" to work on. It found an email I forgot to respond to and drafted a response. Seems really cool! One thing I'm confused on is when I click one of the recommended actions it moves the card to
  • @OnFinality @mamagnus00 The "never open Gmail, Chrome or Slack again" part is the hard bit. Agents that act inside a real browser session break on auth flows and dynamic DOM far more than on the model side.

Mentioned links

Summary generated by AI (Gemini 3.8 Flash, 27 September 2026). It may contain errors – the original source is authoritative.

Related entries

Similar in content, found by the AI search.

  • X post:Omar Ghandour

    Agently Announces Agent-Driven Desktop Control for Mac

    Co-founder Omar Ghandour has announced Agently ('Jarvis for Mac'). The tool is designed to manage workflows directly on macOS, using an orchestration agent that coordinates sub-agents connected to company data and apps such as Outlook, HubSpot, and Stripe.

    286bookmarks19.9kviews

    AI & LLMs· Announcement

  • X post:Andrei Maxim

    Agent Workflows: Specs and Tickets vs Direct Prompting

    Andrei Maxim argues that writing specifications and splitting work into tickets is outdated with modern models like Sol 5.6 and Opus 5. Instead, developers should instruct the agent directly and break large tasks into shippable pull requests through dialogue.

    123bookmarks24.7kviews

    AI & LLMs· Opinion

  • Link:IndyDevDan

    Tactical Agentic Coding: Moving Beyond In-the-Loop AI Assistance

    Tactical Agentic Coding (TAC), created by IndyDevDan, is a training framework focused on advancing from basic interactive AI coding to fully autonomous agentic engineering. Rather than remaining 'in the loop' via conversational prompting, software engineers are trained to combine deterministic code with non-deterministic agents like Claude Code to orchestrate self-correcting development pipelines.

    AI & LLMs· Resource

  • Repository:NousResearch/hermes-agent

    Hermes Agent: Self-Improving Open Source AI Agent by Nous Research

    Hermes Agent is a cross-platform, self-improving AI agent designed with an autonomous learning loop, cross-session memory, and integration into multiple messaging platforms and terminal interfaces.

    240.8kstarsPython

    AI & LLMs· Tool

  • X post:Viktor Oddy

    MotionSites AI MCP: Supplying AI Agents with Curated UI Prompts and Designs

    Viktor Oddy has introduced an MCP (Model Context Protocol) server for MotionSites, allowing AI agents to access curated website designs, animated backgrounds, UI sections, and design prompts to improve the quality of AI-generated frontend code.

    2.6kbookmarks99.3kviews

    AI & LLMs· Tool

  • Repository:citrolabs/ego-lite

    ego lite: A Browser Designed for Shared Use Between Humans and AI Agents

    ego lite is an open-source browser built for macOS that allows users to share their active logged-in sessions with AI agents such as Claude Code or Codex. It isolates agent automation tasks into dedicated Spaces, enabling background execution without disrupting the user's open tabs.

    15.1kstarsJavaScript

    AI & LLMs· Tool

Back to the knowledge base

Let's talk about your project

Get in touch

Locations

  • Mattersburg
    Johann Nepomuk Bergerstraße 7/2/14
    7210 Mattersburg, Austria
  • Vienna
    Ungargasse 64-66/3/404
    1030 Wien, Austria

Parts of this content were created with the assistance of AI.