Skip to content

October 6, 2025

Key Discussions

OpenAI DevDay Major Announcements

OpenAI announced significant new capabilities at DevDay 2025:

  • Apps SDK: New framework for embedding applications directly in ChatGPT with full support for different views and context passing
  • AgentKit: Platform for embedding agents directly in websites
  • GPT Realtime Mini: New model with 70% lower pricing for real-time applications
  • MCP Integration: Confirmed that the Apps SDK is built on top of Model Context Protocol (MCP), using resource endpoints to deliver UI

Technical Insights

SDK Challenges: - OpenAI Agents SDK identified as having poor documentation and limited usefulness for real-time applications - Azure OpenAI API incompatibility with the SDK due to different API specifications - Codex SDK criticism for requiring extensive configuration through $CODEX_HOME/config.toml

Pricing Analysis: - GPT Realtime API pricing analysis revealed ~$10 for 100K tokens during development testing - Acknowledgment that pricing is expensive for non-cached outputs and inputs

Project Updates

Agents Platform Development

Progress on internal Agents Platform showcasing advanced capabilities: - Successfully scaffolded authentication service API issues based on specifications - Automated integration with existing Data Mission project examples - Demonstrated automatic delegation of tasks to agents

Claude Code Behavior Observations

Notable Claude Code patterns identified: - Automatic splitting of complex requests into discrete GitHub issues - Potential use of internal planning TODO lists for issue creation - Recognition of sophisticated behavior emerging from "LLM in a while loop with tools" architecture

MCP Integration Opportunities

Discussion of potential integrations: - Thumbnail editor could be embedded directly in ChatGPT through Apps SDK - MCP-based applications already positioned for ChatGPT integration - Interest in testing private/internal app development capabilities

Community Development

Meetup Planning Progress

Continued discussions on Claude Code community meetups: - No official word from Anthropic on backing for larger events - Alternative approach suggested: smaller, informal happy hour and coworking sessions - Focus on building momentum through approachable gatherings

Strategic Considerations

  • Exploration of independent community building versus waiting for official support
  • Recognition of competitive landscape with OpenAI's aggressive expansion

Technical Highlights

Real-time Development Challenges

  • Azure OpenAI realtime model producing special tokens in text-only mode
  • Custom wrapper development to standardize OpenAI and Azure APIs
  • Decision to move away from OpenAI Agents SDK for better control

Development Automation

Claude Code demonstrated sophisticated request handling: - Automatic breakdown of complex tasks into manageable GitHub issues - Integration of planning capabilities with issue management - Evidence of advanced workflow automation

Resources Shared

  • OpenAI Live Keynote: https://openai.com/live/
  • GPT Realtime Mini Announcement: Twitter thread on 70% price reduction
  • Apps SDK Information: Details on MCP-based ChatGPT app integration

Themes & Insights

Platform Convergence

Growing integration between different AI platforms through standardized protocols like MCP, with OpenAI building on existing standards rather than creating proprietary alternatives.

SDK Quality Concerns

Recurring theme of major AI companies releasing SDKs that lack proper documentation and practical utility, suggesting opportunities for community-driven alternatives.

Community vs. Corporate Support

Tension between waiting for official backing and taking independent action for community building, with preference for organic growth through informal gatherings.

Competitive Acceleration

Recognition of rapidly evolving competitive landscape with major platforms aggressively expanding capabilities across multiple domains.