
Thursday’s five-story technology recap is about controlled delegation. Today’s developments move more work into background agents, cloud-and-edge AI systems, production CI images, security partnerships, and automated performance engineering—while keeping the limits of each announcement visible.
Reporting cutoff: November 19, 2026, 5:08 p.m. EST (22:08 UTC). All five selections were published on November 19 after the prior 5 p.m. Eastern cutoff. Product capabilities, availability, benchmarks, partnerships, and outcomes below are first-party claims unless the article says otherwise.[1][2][3][4][5]
Listen to the broadcast-ready recap
This KittenTTS Bella narration was synthesized locally from the same checked facts as the article.
Tonight’s five-story rundown
- OpenAI describes an internal Performance Factory for regression work.
- Google adds background computer use to Gemini on Mac.
- NVIDIA and Microsoft expand AI deployment options from Azure to PCs.
- GitHub makes RHEL 9 and 10 larger-runner images generally available.
- Fujitsu expands its Google Cloud security and high-confidentiality work.
1. OpenAI explains its Performance Factory
OpenAI’s indexed engineering article describes a Codex-powered internal system that automates parts of investigating and fixing performance issues.[1]
The indexed article says the system combines a configurable task tracker, a dispatch loop, an API that runs Codex in the cloud, and subscriptions that wake agents when work changes. OpenAI says that, for the regression types it focused on, an alert could return a diagnosis in one run without step-by-step guidance.[1]
Practical guidance: teams adapting this pattern should keep the agent’s authority narrow, preserve human review for production changes, and require measurable recovery checks before closing a performance incident.
2. Gemini can work through Mac tasks in the background
Google says computer use in the Gemini app for Mac can handle multistep work across local files, desktop apps, and the web, including clicking, typing, menus, and forms.[2]
Google says the feature runs in an isolated background window with a live preview. Users can select which apps Gemini may access, watch its steps, pause it, and approve key actions.[2]
The rollout is in English in the United States for personal Google accounts held by adults on Apple-silicon Macs running macOS Sequoia 15.0 or later, with Google AI Plus, Pro, or Ultra access.[2]
Practical guidance: start with reversible, low-risk chores; exclude passwords, payments, customer data, and irreversible submissions until app permissions, approval prompts, logs, and stop behavior have been tested.
3. NVIDIA and Microsoft stretch AI deployment from Azure to local systems
NVIDIA says the companies are expanding access to Azure virtual machines powered by Vera Rubin NVL72 systems, adding NVIDIA open models and customization options through Microsoft Foundry, and broadening edge deployment choices.[3]
The announcement says Vera Rubin NVL72 systems are running in Azure data centers. It also describes Foundry Local support for Nemotron 3.5, planned RTX PRO 6000 Blackwell GPU support in Azure Local, and local-model demonstrations on Windows PCs powered by NVIDIA RTX Spark.[3]
Not every item has the same release state: NVIDIA labels Azure’s Vera Rubin offering a preview, says some models are coming soon, and describes a Windows 365 Direct Stream limited preview with public preview planned for early 2027.[3]
Practical guidance: procurement teams should map each named capability to its exact region, hardware, preview status, support boundary, data path, and exit plan rather than treating the announcement as one generally available bundle.
4. GitHub’s RHEL runner images reach general availability
GitHub says Red Hat Enterprise Linux 9 and 10 images for GitHub-hosted larger runners are now generally available and fully supported for production GitHub Actions workflows.[4]
GitHub says teams can use the images to run workflows on the same operating-system family they deploy to or as the foundation for custom images. Existing preview users do not need to change their current runners or workflows.[4]
The images are available for larger runners on GitHub Team and GitHub Enterprise Cloud plans.[4]
Practical guidance: test package availability, system libraries, caching, permissions, network access, and custom-image rebuilds on a nonproduction branch before changing a critical workflow.
5. Fujitsu expands its Google Cloud partnership
Fujitsu says it expanded its Google Cloud partnership around secure AI and cloud use, cybersecurity, high-security cloud environments, forward-deployed engineering, and retail.[5]
Fujitsu says it is participating as a launch partner for Google AI Threat Defense in Japan. It also says the companies will explore Google Distributed Cloud for on-premises and closed environments that require high confidentiality and data sovereignty.[5]
Practical guidance: organizations evaluating a joint offering should ask which party owns identity, logging, incident response, support, data residency, updates, and contractual responsibility for each deployed component.
What to do next
- Desktop-agent users: begin with reversible tasks and the minimum app permissions.[2]
- AI infrastructure buyers: write down which capabilities are available now, in preview, or only planned.[3]
- GitHub Actions teams: test RHEL images against the exact package and permission assumptions in production workflows.[4]
- Security leaders: assign ownership for identity, logs, support, incident response, and rollback across vendor boundaries.[5]
- Automation builders: require measurable postconditions before an agent closes a task.[1]
Need help securing, fixing, or planning your technology? Schedule The IT Guys
More from The IT Guys
Frequently asked questions
Is OpenAI’s Performance Factory a new customer product?
The checked indexed article describes OpenAI’s internal engineering system. This edition does not claim it is a generally available product.[1]
Can every Mac user try Gemini computer use today?
No. Google lists U.S. English, age, hardware, operating-system, account, and subscription requirements.[2]
Are all the NVIDIA and Microsoft capabilities generally available?
No. The announcement mixes running infrastructure, customer availability, previews, planned previews, and coming-soon models.[3]
Are GitHub’s RHEL runner images available on every plan?
No. GitHub lists larger runners on GitHub Team and GitHub Enterprise Cloud plans.[4]
Does the Fujitsu partnership prove a customer environment is secure?
No. It describes partnership scope and intended initiatives; deployment-specific security still requires evidence and testing.[5]
Sources checked for this edition
- OpenAI — How OpenAI’s Performance Factory keeps ChatGPT fast
- Google — Automate Mac tasks with Gemini computer use
- NVIDIA — NVIDIA and Microsoft Expand Accelerated Computing From Azure to Windows PCs
- GitHub — Red Hat Enterprise Linux runner images now generally available
- Fujitsu — Fujitsu expands strategic partnership with Google Cloud to accelerate AI transformation
Source state checked through November 19, 2026, 5:08 p.m. EST / 22:08 UTC. The OpenAI and Fujitsu canonical pages were automation-blocked, so their coverage is limited to preserved complete search-index observations and clearly attributed. The other three entries use direct first-party captures.
Bottom line: automation becomes useful when its authority, release state, evidence, and recovery path remain visible.
