OpenAI used its June DevDay to stop selling standalone models and start selling a complete platform for building, deploying, and governing enterprise agents.
The central DevDay 2026 announcement was GPT-5.5, which consolidated the unified-router approach started with GPT-5: a single base weight set with three latency profiles (Instant, Thinking, Pro), context up to 1M tokens in the Pro tier, and a 40% price cut in the Instant tier versus GPT-5.1. Sam Altman framed the launch not as "a smarter model" but as "a cheaper model to operate at scale," reflecting that 2026's competition plays out as much on cost per task as on raw capability.
AgentKit, originally introduced in 2025, reached version 2.0 with a visual agent-flow builder (connectable nodes for tools, conditionals, and sub-agents) that competes directly with products like LangGraph Studio and Microsoft's Copilot Studio. The most relevant technical addition is native support for MCP as the default tool protocol, replacing the proprietary "functions" format OpenAI had used since 2023.
ChatKit offers production-ready chat interface components (React and Swift) with built-in streaming handling, source citations, and user feedback, cutting enterprise conversational interface development time from weeks to days according to cases shown at the event. Guardrails, the moderation and safety framework OpenAI open-sourced in 2025, now includes system-prompt extraction attempt detection and industry-specific classifiers for regulated content (healthcare, financial services).
The Apps SDK, which lets third parties build applications that run embedded inside the ChatGPT interface, passed 3,000 published apps since its 2025 launch. At DevDay, OpenAI announced a revenue-sharing monetization model for developers, in an explicit move toward turning ChatGPT into a software distribution platform, similar to a conversational App Store.
OpenAI pushed a new metric in its enterprise materials: "cost per resolved task" instead of "cost per token," presenting case studies where GPT-5.5 Instant with AgentKit resolves tier-1 support tickets at a cost of $0.03 per full interaction, including tool calls and guardrail checks.
Companies like Duolingo, Klarna, and an unnamed Latin American insurer presented production deployments: Klarna reported that its customer service agent built on AgentKit 2.0 handles 75% of inquiries without human escalation, a figure the company had partially reported in 2025 and now consolidates with the platform's new version.
For teams evaluating full adoption of the OpenAI stack, the recommendation is to start by migrating the tools layer to MCP even before adopting AgentKit as a builder, since that decouples business logic from the specific model provider and makes it easier to compare GPT-5.5 against Claude Opus 4.5 or Gemini 3 Pro on the same tooling infrastructure without rewriting integrations.
Carlos Montiel is an enterprise AI solutions architect. He implements LLMs, Agents, RAG, and orchestrators for companies across Guatemala and Latin America. Reach out for a consultation.
Contact Carlos Montiel