How to use Apps SDK and Agent Kit at DevDay 2025

Transcript

Hey everyone, welcome to the late in space podcast. This is Allesio, founder of Colonel Labs and I'm joined by Swix, editor of Late in Space. Hello. Hello. And we are here in the OpenAI dev day studio with Sherwin and Christina from the Open AI platform team. Welcome. Thank you for having us. Yeah, it's always to be here. Yeah. Um, it's so it's suc... Read More

Key Insights

• Apps SDK and Agent Kit extend OpenAI’s developer platform by enabling modular tool usage and agent orchestration.
• MCP is presented as a general open protocol for connecting tools, making integration easier for third parties.
• Visual agent builders are emphasized to simplify complex workflows and preserve brand identity within chat experiences.
• Internal dogfooding demonstrates practical use cases, such as OpenAI support powered by AgentKit.
• Prompting remains a critical skill, with emphasis on prompt optimization and human oversight in production.
• The platform aims to scale agents by improving tool connections, evaluation, and governance.
• OpenAI values developer feedback and iterates on dev day experiences based on sponsor and attendee input.
• The evolution from plugins to GPTs to Agents highlights a gradual shift toward production-ready agent platforms.

Questions & Answers

Q: A real search query question about the video (e.g. 'How to...', 'What is...', 'Why does...', 'When should...'). Question 1

What is AgentKit and how does it relate to the Apps SDK and MCP protocol, and what problem does it solve for developers building AI agents? AgentKit provides a comprehensive toolset for building, deploying, and optimizing AI agents, while MCP offers a universal way to connect tools. The combination helps developers create production-ready agents with better tool integration, governance, and consistency across platforms.

Q: Question 2

How does the live 8 minute agent build demonstrated by Christina illustrate the capabilities of the visual Agent Builder, and what benefits does it offer for developers new to agent development? The live demo showcases rapid assembly of an agent using a visual interface, reducing coding friction, enabling rapid testing, and enabling teams to experiment with prompts, tool calls, and flows. It demonstrates how non-experts can start building effective agents quickly.

Q: Question 3

Why is the inversion of the website chat paradigm significant, with apps embedded inside ChatGPT rather than a separate chatbot, and what advantages does this bring to users and developers? Embedding apps inside ChatGPT creates a more seamless experience where conversations control and access to tools is contextual and fluid, enabling richer interactions, branding consistency, and easier access to functions without switching contexts or leaving the chat flow.

Q: Question 4

What role does human in the loop play in the AgentKit workflow, and how does it support responsible and scalable agent deployment? Human in the loop enables oversight during tool use, approval flows, and safety checks, improving reliability and control when agents operate in production, while still leveraging automated prompts and optimization pipelines for scaling.

Q: Question 5

How does MCP contribute to tool connectivity and why did OpenAI adopt it for Apps SDK, according to the team on DevDay? MCP provides a general, open protocol for tool integration that is easy to adopt and widely usable. OpenAI adopted MCP to ensure developers can connect tools in a consistent way, fostering interoperability and easier scaling across different systems.

Q: Question 6

What is the relationship between plugins, GPTs, and the new agent platform in the OpenAI roadmap, and what lessons were learned along the way? The evolution shows a progression from plugins to GPTs to agents, with each step teaching better integration, user control, branding, and production readiness. The team emphasizes learning from past feedback to improve developer experiences and tooling.

Q: Question 7

What challenges are mentioned regarding serving custom fine-tuned models at scale, and how does AgentKit aim to address them? Challenges include maintaining performance, governance, and consistency across deployments. AgentKit addresses these by offering structured tool workflows, evaluation pipelines, and scalable prompt optimization to manage complex agent ecosystems.

Q: Question 8

How does OpenAI plan to involve developers after DevDay and what feedback mechanisms are highlighted for ongoing iteration? The team notes they read feedback emails and act on them, continuing to refine API surfaces, visual builders, and tool integrations based on community input, ensuring the platform evolves to meet developer needs and production requirements.

Summary & Key Takeaways

• OpenAI presents Apps SDK and Agent Kit with a focus on MCP as the tool-connect protocol and visual workflows for agents. The hosts explain how the new tools integrate with ChatGPT and emphasize branding and developer experience. The show includes a live demo and reflections on past plugin and GPT developments.

• The discussion covers production considerations for agents, including evaluation pipelines, prompting optimization, and human in the loop. Visual builders are highlighted as a means to simplify complex agent systems and maintain control over tool use and brand adaptation.

• The conversation concludes with reflections on internal deployment and the persistence of prompting as a core skill, noting challenges of scaling custom models and the evolution from plugins to GPTs to agent platforms and the value of an open protocol consensus.