Key Moments
DevDay 2025: Apps SDK, Agent Kit, MCP, Codex and why Prompting is More Important than Ever
Latent Space PodcastKey Moments
OpenAI DevDay introduces Agent Kit, Apps SDK, and MCP, emphasizing agent building, app integration within chat, and enhanced developer tools. Prompting remains crucial.
Key Insights
OpenAI is enhancing developer tools with Agent Kit for streamlined agent creation and deployment, and the Apps SDK for integrating applications directly within chat interfaces.
The MCP protocol, adopted by OpenAI, is highlighted as a valuable open standard for tool integration, fostering interoperability and simplifying agent development.
Agent Builder provides a visual, user-friendly interface for creating and iterating on AI agents, supporting complex workflows and human-in-the-loop interactions.
Prompt optimization is more critical than ever, with ongoing research and development focusing on automated prompt tuning and embedding these capabilities into the platform.
OpenAI is prioritizing reliability and transparency with the new Service Health Dashboard, aimed at giving developers real-time insights into their API integration performance.
The company is exploring ways to make inference costs more manageable for developers, with 'bring your own key' options being a discussed future possibility.
EVOLUTION OF DEVELOPER ENGAGEMENT AND PLATFORM GROWTH
OpenAI's DevDay 2025 saw a notable increase in developer engagement, with 4 million registered developers, up from the previous year. This growth reflects the company's ongoing commitment to empowering third-party developers to bring AGI benefits to the world. The event's improved organization, including dedicated podcast studios, signifies a maturing approach to developer outreach, building upon feedback from previous iterations like GPTs and plugins. The focus remains on iteratively releasing tools that foster developer innovation and expand the reach of AI technologies.
APPS SDK AND THE INVERTED PARADIGM OF CHAT INTEGRATION
The Apps SDK represents a significant shift by embedding applications directly within ChatGPT, inverting the traditional model of separate chatbots on websites. This allows for a more integrated and context-aware user experience, as demonstrated with integrations like Canva. This approach, learned from previous efforts like plugins, emphasizes giving developers more control over their brand and user experience, making the integration feel seamless and native within the ChatGPT environment.
AGENT KIT: COMPREHENSIVE TOOLS FOR BUILDING AI AGENTS
Agent Kit is introduced as a full suite of solutions designed to simplify the process of building, deploying, and optimizing AI agents. It addresses the complexities builders face with prompt engineering, optimization, and evaluation. Key components include the Agent SDK for programmatic control, the Agent Builder for visual workflow creation, a Connector Registry for tool integration (leveraging MCP), Trackit for tracing and debugging, and Evals for performance assessment, offering an end-to-end system for agent development.
MCP PROTOCOL AND INTEROPERABILITY AMONG TOOLS
OpenAI's adoption of the Message Passing Protocol (MCP) highlights its commitment to open standards for tool integration within agents. Developed by Anthropic and treated as an open protocol by the community, MCP simplifies connecting various tools and services. This interoperability is crucial for building sophisticated agents, allowing developers to leverage existing ecosystems and promoting a standardized approach to tool access, thereby reducing development friction and encouraging broader adoption.
AGENT BUILDER: VISUAL WORKFLOWS AND FLEXIBLE DEPLOYMENT
The Agent Builder offers a visual, canvas-like interface for designing AI agent workflows, making complex processes more accessible. It supports features like user approval nodes and state management, enabling near Turing-complete functionality. Developers can use it as a playground for iteration and prompt optimization, then export their work to the Agent SDK, or leverage OpenAI's deployment services. The builder also benefits from pre-built templates for common use cases like customer support and data enrichment.
ENHANCING EVALUATION AND PROMPT OPTIMIZATION
The Evals product has been expanded to support agent evaluations, allowing for the assessment of longer, more complex agent traces. This includes breaking down traces for granular analysis and incorporating human-in-the-loop feedback. Automated prompt optimization is a key area of investment, aiming to tie directly into evaluations to continuously improve agent performance. This is particularly important as prompting remains a critical skill, despite predictions of its obsolescence, with ongoing research like GDPA influencing future developments.
THE CONTINUED IMPORTANCE OF PROMPTING AND KODEX BENEFITS
Contrary to early predictions, prompting has become more entrenched and important in AI development. OpenAI recognizes this by investing in prompt optimization tools. Meanwhile, Codex continues to accelerate development, with users increasingly trusting it for larger coding tasks and feature development. Its capabilities extend to code reviews, improving developer productivity by providing starting points and aiding in code navigation, thereby reducing context-switching time and facilitating faster onboarding into codebases.
RELIABILITY AND DEVELOPER TRANSPARENCY WITH SERVICE HEALTH DASHBOARD
A significant, though not stage-highlighted, launch is the Service Health Dashboard. This tool provides developers with real-time, organization-scoped insights into their API integration performance, including token velocity, throughput, and response codes. This initiative stems from a deep focus on reliability, particularly after past outages, aiming to offer developers transparency and peace of mind as OpenAI works towards achieving higher levels of uptime, such as five nines.
THE EXPANDING ECOSYSTEM OF CONNECTORS AND WIDGETS
The Connector Registry and the ChatKit widgets are central to building a robust AI application ecosystem. While OpenAI offers first-party connectors, the MCP protocol encourages third-party development, creating a diverse range of tool integrations. Similarly, ChatKit's embeddable iframe and widget builder aim to become a drop-in solution for common chat experiences, offering a polished, consumer-grade UI that potentially reduces the need for developers to reinvent these fundamental components.
FUTURE DIRECTIONS: MODALITIES, EXTERNALIZATION, AND COST MANAGEMENT
Future developments for Agent Kit include expanded modalities (like voice), more logical nodes for diverse workflows, and enabling users to run agent workflows independently of ChatKit. The platform is also evolving to address developer concerns around inference costs, with 'bring your own key' being a potential future feature. This aligns with the vision of making ChatGPT a personal assistant by integrating deeply with user workflows and identity, as seen in partnerships with Apple and Kakao.
Mentioned in This Episode
●Software & Apps
●Companies
●Organizations
●Concepts
Common Questions
The key takeaway is OpenAI's continued commitment to empowering developers by opening up their technology. New tools like the Apps SDK and Agent Kit aim to make it easier for developers to build and deploy sophisticated AI applications, leveraging ChatGPT's distribution and advanced agent capabilities.
Topics
Mentioned in this video
A set of solutions launched by OpenAI to help developers build, deploy, and optimize AI agents.
A component of Agent Kit, described as an embeddable iframe for tracking agent activity.
A component of Agent Kit that handles the chat user experience and provides widgets.
A software development kit for building agents that can interact with tools and APIs.
An earlier feature allowing ChatGPT to interact with external tools and services.
Customizable versions of ChatGPT that users can create for specific purposes.
A component of Agent Kit that allows developers to visually model, iterate, and deploy agents.
A tool for evaluating the performance of AI models and agents, integrated into Agent Kit.
Referred to as a 'cheeky one' by the interviewer, likely a colloquial or internal name for a tool/API.
A part of the Agent Kit framework for managing connections to various tools and services.
OpenAI's API for fine-tuning models, mentioned as being grouped outside the Agent Kit umbrella.
More from Latent Space
View all 106 summaries
86 minNVIDIA's AI Engineers: Brev, Dynamo and Agent Inference at Planetary Scale and "Speed of Light"
72 minCursor's Third Era: Cloud Agents — ft. Sam Whitmore, Jonas Nelle, Cursor
77 minWhy Every Agent Needs a Box — Aaron Levie, Box
42 min⚡️ Polsia: Solo Founder Tiny Team from 0 to 1m ARR in 1 month & the future of Self-Running Companies
Found this useful? Build your knowledge library
Get AI-powered summaries of any YouTube video, podcast, or article in seconds. Save them to your personal pods and access them anytime.
Try Summify free