The Battle for the AI Stack

What happened 
OpenAI is reportedly building a code‑hosting platform to rival GitHub. The Information said engineers decided to create their own repository after service disruptions made GitHub unreliable. The project remains in its early stages and may be offered for purchase to OpenAI’s enterprise customers. If launched, it would compete directly with Microsoft — which holds a significant stake in OpenAI — and signals the AI lab’s ambition to control more of the software development stack.

Why it matters
GitHub underpins modern software development. An alternative from OpenAI could reshape developer workflows by integrating AI‑assisted coding and hosting in a single environment. It also underscores tension between OpenAI’s independence and Microsoft’s influence. Any new platform would be entering a crowded market, but it could appeal to organizations seeking greater reliability and tighter integration with AI models.

What’s next 
The repository is months away from completion. Observers will watch whether OpenAI can attract a community of developers and differentiate the product beyond AI‑assisted code generation. Microsoft’s reaction will also be telling: the tech giant must balance its investment in OpenAI with the success of its own GitHub service.

What happened
A report from the National Association of State Chief Information Officers (NASCIO) finds that while most U.S. states now use generative AI tools, a handful are experimenting with agentic AI — digital agents that carry out multi‑step tasks with limited human oversight. NASCIO’s survey counted eight states using agentic AI (names not disclosed) and pointed to early projects like Alaska’s plan to add agents to its mobile app for proactive notifications, dynamic form‑filling and workflow orchestration. Virginia’s former governor piloted an agentic system to scan and streamline regulations, dubbed the “Virginia Model”. The association outlines a five‑phase maturity model, from assistive generative AI to AI‑initiated work.

Why it matters
The report suggests that agentic AI is creeping into government, albeit cautiously. States see potential in automating low‑risk tasks like notifications and eligibility checks, but legacy systems and risk aversion slow adoption. Policymakers are watching private‑sector pilots but face accountability pressures that make fully autonomous agents a hard sell. As public‑sector demand grows for efficient service delivery, agentic AI could help bridge staffing gaps — provided agencies can ensure transparency and control.

What’s next
Expect incremental experimentation. NASCIO predicts states will discuss early deployments at its mid‑year conference in April. Adoption will likely focus on discrete, low‑risk tasks before expanding to stateful, multi‑step workflows. Vendors building agent frameworks may find a growing public‑sector market, but must address compatibility and governance concerns.

What happened
Sunnyvale‑based Noble Machines emerged from stealth with a general‑purpose robot designed for hazardous industrial tasks. Founded in 2024 by veterans from Apple, SpaceX, NASA and Caltech, the startup shipped its first units to a Fortune Global 500 customer within 18 months. Noble’s system pairs AI‑driven whole‑body control with cost‑effective hardware so robots can learn new skills in hours via language instructions, demonstrations and gestures. The company works with partners like ADLINK, Schaeffler and Solomon to deploy robots in manufacturing, construction, logistics, energy and semiconductor operations.

Why it matters
General‑purpose robots capable of rapid learning represent a shift from narrowly‑programmed industrial arms to adaptable physical AI. Noble’s approach integrates hardware and autonomy so robots can perform varied, physically demanding tasks, potentially reducing injuries and labour shortages. Deployments with a top‑tier customer suggest industry demand for versatile robots that work alongside people and can be iterated quickly.

What’s next
Noble Machines will demonstrate its robots at Nvidia’s GTC conference and expand collaborations with industrial partners. The key metrics will be uptime, safety and return on investment. Success could accelerate adoption across heavy industry, raising questions about workforce impact and the need for new safety and training standards.

What happened
After President Trump’s February 27 order to phase out Anthropic, three U.S. cabinet agencies — State, Treasury and Health and Human Services — began dropping Anthropic’s Claude platform[15]. Officials directed staff to switch to OpenAI’s GPT‑4.1 or Google’s Gemini, giving agencies six months to comply. The move extends the military’s boycott and threatens to make Anthropic a national‑security outcast. OpenAI announced that its defence deal would bar intentional domestic surveillance.

Why it matters
The contract dispute underscores a power struggle over AI usage rights. Anthropic insisted its models not be used for autonomous weapons or domestic spying, while the Trump administration demanded unrestricted access. Washington’s pivot to rival platforms shows how quickly political decisions can reshape the AI supplier landscape. Federal agencies moving en masse from one vendor to another is rare and could influence private‑sector procurement.

What’s next
Anthropic intends to challenge the national‑security designation in court, but its government business may suffer in the interim. Competitors will need to balance ethical guardrails with customer demands for flexibility. Expect ongoing debates over whether AI vendors can impose usage restrictions — and whether governments will respect them.

The battle for the AI stack is expanding across software platforms, government workflows, physical robotics, and national policy. The companies that control these layers won’t just build AI — they’ll shape how it’s deployed across the economy.