Welcome, AI Builders!
You’ve probably used Windows for most of your life. Now imagine it without a keyboard. Microsoft is rethinking the OS around AI agents that see, hear, and act like you. OpenAI just made its first open-weight models public in five years. Google quietly launched a developer agent that works while you're away. And Cloudian is rebuilding the pipes that keep all of it running.
📌 In today’s Generative AI Newsletter:
OpenAI releases two Apache-licensed OSS models with strong reasoning skills
Google’s Jules launches as an always-on dev agent with structured pricing
Microsoft previews an AI-first operating system for 2030
Cloudian builds storage built to feed AI agents at full speed
🗃️ OpenAI Cracks Open the Vault: OSS Models Are Here
Image Credit: OpenmAI
After five years of closed releases, OpenAI just dropped gpt-oss-120b and gpt-oss-20b, its first open-weight models since GPT-2. Both are licensed under Apache 2.0 and already top Hugging Face’s leaderboard, giving developers full access to near-frontier reasoning systems they can run locally or build on commercially.
Here’s what’s in the release:
Performance rivaling closed models. The 120B version matches o4-mini on major reasoning tasks. The 20B model runs on a laptop with 16GB RAM and still competes with o3-mini.
Built for agents. Both support tool use, web search, Python execution, and adjustable reasoning levels, making them fit for autonomous workflows.
High hallucination rates. On PersonQA, the 120B model hallucinated 49% of the time, nearly triple the rate of o1. OpenAI says this reflects tradeoffs in model size and knowledge scope.
No training data shared. While the weights are open, OpenAI withheld its datasets, citing legal risk. The models are text-only, with no image or audio capabilities.
OpenAI’s OSS models give developers something to build with, but not enough to rival its premium line. By keeping training data and full capabilities under wraps, OpenAI maintains its commercial moat while deflecting criticism and appeasing policymakers. These releases signal openness, not surrender.
🧑💻 Google’s Coding Agent Jules Graduates From Beta
Image Source: Jules
Google has officially launched Jules, its asynchronous AI coding agent, out of beta and into structured pricing. Built on Gemini 2.5 Pro, Jules can spin up a Google Cloud VM, clone your repo, and silently handle coding tasks while you step away. It's Google's straightforward solution to Cursor and DevGPT; it's a hands-off executor as well as a copilot.
Here’s what’s changing:
Free access is now capped at 15 tasks per day. Google replaced the 60-task beta limit with structured pricing. Pro users can unlock higher limits for $19.99 or $124.99 per month.
Jules works in the background. Unlike agents that wait for prompts, Jules handles tasks asynchronously. Developers can step away and return to completed code.
GitHub integration has been upgraded. Jules can now open pull requests on its own and preserve environments with reusable dependency snapshots.
Mobile use is rising fast. Forty-five percent of Jules’ 2.28 million beta visits came from mobile browsers. Google is now exploring mobile-specific features.
Jules is already being used on internal projects and will soon be embedded across more of Google’s engineering workflows. It turns the traditional dev loop into something quieter and more autonomous. Whether that’s empowering or unnerving depends on who you ask.
🖥️ Microsoft Envisions Windows By 2030
Image Credit: Microsoft
Microsoft has kicked off its “Windows 2030 Vision” campaign with a bold claim: future versions of Windows won’t need a keyboard. In a new video, the company outlined plans for an AI-first operating system that can see, hear, and act like its user, replacing manual interaction with multimodal autonomy.
Here’s what Microsoft is previewing:
Agentic AI will drive the OS. Microsoft VP David Weston says Windows will become multimodal, seeing, hearing, and speaking like its user. The goal is to replace traditional commands with more natural, conversational input.
Desktop UX could vanish. Weston compares future keyboard use to MS-DOS nostalgia. The interface could dissolve into a background layer while autonomous agents manage apps, tasks, and data flows.
Security will be ‘appliance-level.’ Microsoft wants every user, even in small businesses, to deploy fleets of AI agents for security without needing to configure or manage them.
Work becomes less mechanical. The pitch is that AI will handle the “toil,” freeing humans for creativity, vision, and judgment. The system itself becomes a thinking collaborator.
Microsoft is attempting to redraw the boundary between human and machine agency at the OS level. But agentic AI cuts both ways: the less you click, the more your OS decides. And in that future, Microsoft doesn’t just want your input. It wants your trust.
MIT-Trained Engineer's Cloudian Solves AI's Data Bottleneck
Image Source: MIT
Michael Tso, an MIT-trained engineer, co-founded Cloudian to solve a growing problem in AI infrastructure. Legacy storage systems were never designed to serve millions of parallel AI requests in real time. Cloudian’s new platform rewires that pipeline, making it possible to feed data-hungry models without friction.
Here’s what’s rolling out:
Cloudian’s unified architecture merges storage and compute on a single platform, allowing direct, high-speed transfer of large datasets to GPUs and CPUs with minimal latency.
The system relies on object storage, tagging each file with metadata and supporting massive, flat-file datasets that suit unstructured inputs like video, audio, or genomics.
A newly added vector database layer turns incoming data into embeddings on the fly, enabling AI agents to access, search, and reason over fresh data without separate preprocessing.
Tight integration with NVIDIA GPUs ensures near-zero delay in model training and inference, reducing bottlenecks in AI pipelines and improving real-time decision-making.
Modern AI agents consume data in parallel at a scale far beyond human workflows. Cloudian's platform closes the speed gap between storage and compute, turning data lakes into live wires. From self-healing robots in auto plants to AI tools for cancer genomics, the system is powering deployments that demand always-on intelligence.
🚀 Boost your business with us. Advertise where 12M+ AI leaders engage!
🌟 Sign up for the first AI Hub in the world.
