Stenobird

Episode

Pwning Your Friends' Agents Is Good Manners, Make Something Agents Want, OpenAI Buys TBPN, MCP Goes Stateless, Cloudflare Agents Week with Sunil Pai

Podcast
You've Been a Bad Agent
Published
Apr 6, 2026
Duration seconds
4131
Processing state
processed
Canonical source
https://share.transistor.fm/s/f7c554ba
Audio
https://media.transistor.fm/f7c554ba/65b5243b.mp3
JSON
/v1/public/podcasts/you-ve-been-a-bad-agent-7334728/episodes/pwning-your-friends-agents-is-good-manners-make-something-agents-want-openai-buys-tbpn-mcp-goes-stateless-cloudflare-agents-week-with-sunil-pai
Markdown
/podcast/you-ve-been-a-bad-agent-7334728/pwning-your-friends-agents-is-good-manners-make-something-agents-want-openai-buys-tbpn-mcp-goes-stateless-cloudflare-agents-week-with-sunil-pai.md

Actions

  • POST https://stenobird.com/v1/public/podcasts/you-ve-been-a-bad-agent-7334728/episodes/pwning-your-friends-agents-is-good-manners-make-something-agents-want-openai-buys-tbpn-mcp-goes-stateless-cloudflare-agents-week-with-sunil-pai/transcription-requests
    Idempotently request low-priority transcript generation for this episode.
  • GET https://stenobird.com/podcast/you-ve-been-a-bad-agent-7334728/pwning-your-friends-agents-is-good-manners-make-something-agents-want-openai-buys-tbpn-mcp-goes-stateless-cloudflare-agents-week-with-sunil-pai.md
    Read the agent-friendly Markdown representation of this episode resource.

Summary

The future of software lies in malleable, generative interfaces rather than static, predetermined product experiences. This episode explores the shift toward stateless Model Context Protocol (MCP) and the rise of agents that build services for other agents.

Topics

  • Generative UI
  • Model Context Protocol
  • LLM Agents
  • Malleable Software
  • Cloudflare Workers
  • Stateless Computing
  • AI Engineering
  • Tool Use

Highlights

  • Main idea: Generative UI should move beyond simple chat to 'malleable software' where the model dynamically creates functional canvases and tools
  • Practical takeaway: Developers should build services and APIs specifically for agent consumption (e.g., friends.fyi) rather than just human interfaces
  • Technical shift: The Model Context Protocol (MCP) is moving toward a stateless model using multi-turn tool calls to handle missing arguments via elicitation errors
  • Failure mode: Stateless tool execution introduces risks where code mutations might execute twice if the developer doesn't account for re-running logic during elicitation
  • Vision: The next frontier of AI engineering is 'executable oracles'—constraining LLM degrees of freedom to produce reliable, structured outputs

Chapters

  1. 1:00 The Social Feed of AI: A brief discussion on the convergence of content platforms and the current state of social media trends.
  2. 11:00 Generative UI and Malleable Software: Exploring the concept of 'malleable software' where LLMs use canvases to create emergent, interactive experiences like dynamic tic-tac-toe boards.
  3. 22:00 The OpenAI Acquisition Rumors: Analyzing the media narrative surrounding the potential acquisition of TBPN by OpenAI and its implications for the industry.
  4. 32:00 Building for Agents, Not Humans: Discussing the shift toward creating services, knowledge bases, and APIs that are designed to be interrogated by autonomous agents.
  5. 53:00 Personal Agent Productivity: A look at using custom agents to manage local environments, browser automation via CDP, and personal administrative tasks.
  6. 1:03:00 The Future of Stateless MCP: A deep dive into the technical evolution of MCP, focusing on statelessness, multi-turn tool calls, and the challenges of elicitation errors.