This was a watershed week for OpenAI’s Codex ecosystem. Codex shed its “coding assistant” label and emerged as a universal work agent with 90+ integrations. GPT-5.5 Instant became the new default model, dramatically reducing hallucinations. And OpenAI announced DeployCo, a $4B deployment services company that signals a new level of enterprise commitment. For IT leaders evaluating AI agent platforms, the strategic picture is coming into focus.
Codex Becomes a Universal Work Agent
On May 14, OpenAI officially expanded Codex beyond its coding roots into a general-purpose automation platform. The updated macOS and Windows app now supports 90+ plugins and integrations including Jira, GitLab Issues, Microsoft Suite, Slack, Google Drive, email, and calendar. New capabilities include computer use, in-app browsing, image generation, memory, and plugins.
Sam Altman posted: “Big upgrade for codex today! Try it for non-coding computer work.” And it’s not just desktop — TechCrunch reported that Codex is being integrated into the ChatGPT mobile app, extending agent capabilities to phone users.
Why this matters: This changes the procurement conversation entirely. If Codex can handle both developer workflows and knowledge-worker automation, organizations can standardize on a single agent platform instead of buying separate tools for engineering and business operations. The Chrome extension launched May 7 adds browser automation on top — three in one.
GPT-5.5 Instant: New Default, Sharper Output
OpenAI shipped GPT-5.5 Instant as the default ChatGPT model with impressive quality improvements across the board:
- 52.5% fewer hallucinated claims compared to GPT-5.3 Instant in high-stakes domains
- 37.3% reduction on historically problematic conversation categories
- 30.2% verbosity reduction — shorter, tighter responses
- 29.2% fewer lines on average
- New memory source controls and context-aware personalization
Why this matters: The hallucination reduction alone is a headline for regulated industries. Combined with verbosity cuts, this is a meaningful step toward the reliability bar that enterprise compliance teams require. GPT-5.5 Instant is available in ChatGPT today and also powers Codex on NVIDIA infrastructure.
OpenAI DeployCo: $4B for Enterprise Deployment
On May 11, OpenAI announced a dedicated deployment services company — internally referred to as “DeployCo” — with a $4B initial investment and a ~$10B pre-investment valuation. The company will deploy Forward Deployed Engineers (FDEs) who work directly with enterprises on integration, customization, governance frameworks, and change management.
Why this matters: This is OpenAI’s response to the perennial enterprise complaint that “we have the models but don’t know how to deploy them at scale.” DeployCo is effectively a system integrator with OpenAI-level access to roadmap and engineering. For consulting firms like Big Hat Group, this creates both a partner and a competitor — and underscores that enterprise AI deployment is a service, not just a product purchase.
The Next Evolution of the Agents SDK
OpenAI published the evolved Agents SDK with a model-native harness that incorporates patterns from Codex-style agents. Key additions:
- Native sandbox support with 7 officially integrated hosted sandbox providers: Blaxel, Cloudflare, Daytona, E2B, Modal, Runloop, Vercel
- Manifest abstraction for portable workspace descriptions across sandbox providers
- Configurable memory, sandbox-aware orchestration, and Codex-like filesystem tools
- MCP integration and progressive disclosure via skills
- Shell tools and apply-patch tools for code editing
The SDK launches first in Python; TypeScript support is planned. OpenAI also published an agent improvement loop cookbook demonstrating a feedback flywheel: trace → human feedback → automated evaluation → harness improvement.
Why this matters: The sandbox abstraction layer is the SDK’s killer feature. Teams can develop against one sandbox provider and deploy to another — or use different providers for dev, staging, and production. For enterprises with existing relationships with Cloudflare or Vercel, this reduces the friction of adopting OpenAI’s agent framework.
Codex CLI v0.130.0
Codex CLI 0.130.0 shipped May 12 with:
- Plugin sharing metadata exposure for better discoverability
codex remote-controlcommand for headless startup- Thread pagination for large conversation histories
- Enhanced Bedrock authentication supporting AWS console login credentials from standard AWS profiles
- Subagent workflows for parallel execution of specialized agents (exploration, file analysis, testing, log triage)
Enterprise & Compliance
A flood of enterprise updates this week:
Workspace agents with EKM — Available to eligible Enterprise workspaces with Enterprise Key Management enabled. Admins can create, publish, schedule, and deploy agents across Slack channels with RBAC. Free preview through June 2, 2026.
Compliance Platform expansion — Grown to 13 eDiscovery/DLP providers including CrowdStrike, Global Relay, Teleskope, and Zenity. CrowdStrike’s Falcon Shield integration provides security teams visibility into all GPT agents deployed within workspaces. Legacy stateful route deprecated June 5, 2026.
ChatGPT for Intune — OpenAI released a Microsoft Intune-managed ChatGPT iOS app, enabling enterprise MDM deployment.
Data residency expansion — Customers can now store and process data in the US or Europe for ChatGPT Enterprise, Edu, and the API Platform.
Analytics dashboard — ChatGPT Business workspace owners gained unified views of agent adoption, usage patterns, and performance metrics.
OpenAI–Microsoft: The Exclusivity Era Ends
The April–May 2026 amendment to the OpenAI–Microsoft partnership dissolved Microsoft’s cloud exclusivity rights, enabling true multi-cloud AI procurement. This follows the earlier Bedrock partnership with AWS, confirming that OpenAI is pursuing a multi-cloud strategy.
Why this matters: Enterprise architecture teams that were locked into Azure for OpenAI workloads now have options. The freedom to run Codex and OpenAI models across AWS, Azure, Google Cloud, or on-prem infrastructure simplifies compliance, reduces vendor risk, and enables cost optimization through competitive pricing. Read our full Codex platform comparison: AWS Bedrock vs Azure OpenAI for a deeper architectural analysis.
Voice Intelligence Models Join the API
OpenAI released three new real-time audio models in the API:
- GPT-Realtime-2 — Conversational task execution
- GPT-Realtime-Translate — Multilingual translation across 70+ languages
- GPT-Realtime-Whisper — Streaming speech-to-text
What to Watch
- Codex on mobile — The ChatGPT mobile integration could be the catalyst that pushes Codex from 4 million to 40 million users. Watch for the public beta timeline.
- DeployCo’s first enterprise engagements — How DeployCo positions itself relative to existing system integrators and consulting firms will define the enterprise AI services landscape.
- Workspace agents free preview deadline — June 2 window gives a narrow, no-risk opportunity for enterprises to evaluate workspace agents before credit-based pricing kicks in.
- Multi-cloud Codex — With Bedrock support already live and the Microsoft exclusivity clause dissolved, watch for Codex availability on Google Cloud and other providers.
That’s Codex Weekly for May 15. OpenAI transformed Codex into a universal work agent, shipped a smarter default model, founded a $4B deployment company, and opened the doors to multi-cloud AI. The pace of change is relentless — and the enterprise AI platform landscape looks fundamentally different than it did a month ago.
Check back next week for the next edition. Explore how Big Hat Group can help your team evaluate and deploy Codex — from strategy and architecture to hands-on implementation. We specialize in AI agent systems on Azure, AWS, and Windows 365.
— Kevin Kaminski, Principal Architect at Big Hat Group. We help enterprises design and deploy AI agent systems on Azure, AWS, and Windows 365.
Note: Three of four planned research briefs (Codex Core & CLI, ChatGPT/API, and Company Ecosystem) were unavailable this week. This edition is compiled from the Agents & Enterprise research file only. All claims are verified against linked sources.