[#9] OpenAI's Coding Revolution, Microsoft's Agentic Web, and NVIDIA's AI-First PCs

Hey there,

It's been a wild month in AI. OpenAI just dropped their first full-fledged AI coding agent, Microsoft unveiled 50+ tools to build what they're calling the "agentic web," and NVIDIA launched personal AI supercomputers that pack up to 20 petaflops of performance.

Let's dive into what's happening and why it matters.

OpenAI Codex: The First True AI Coding Agent

OpenAI just launched Codex, their first full-fledged AI coding agent, and it's a game-changer. Unlike previous coding assistants that just suggest snippets, Codex runs in a sandboxed virtual computer and can handle entire software engineering workflows.

What makes Codex different:

• Takes 1-30 minutes to complete tasks (showing its work as it goes)

• Connects with GitHub to preload your code repositories

• Handles multiple tasks simultaneously

• Can write features, fix bugs, answer questions about codebases, run tests, scaffold new features, and draft documentation

The most interesting part? You can include an "AGENTS.md" file in your repos with custom instructions for how Codex should work with your specific codebase.

This release puts serious pressure on Anthropic's Claude Code (released February 2025) and Google's Gemini Code Assist (updated with agentic abilities in April). The coding AI market is heating up fast – Cursor reportedly hit ~$300M annualized revenue in April, and OpenAI apparently acquired Windsurf (AI coding platform) for $3B.

Microsoft's 50+ Tools for the "Agentic Web"

At Build 2025 yesterday, Microsoft announced over 50 AI tools focused on building what they're calling the "open agentic web" – autonomous systems that can make decisions and complete tasks with minimal human intervention.

The most significant announcements:

GitHub Copilot coding agent: Goes beyond suggesting code to autonomously refactoring code, improving test coverage, fixing defects, and implementing new features. They're also open-sourcing GitHub Copilot Chat in Visual Studio Code. Pricing: $10/month for individuals, $19/user/month for businesses

Azure AI Foundry Agent Service: Now generally available, it lets developers build enterprise-grade AI agents with support for multi-agent workflows and open protocols like Agent2Agent (A2A) and Model Context Protocol (MCP). Pricing: Pay-as-you-go with compute and API call charges, enterprise plans available

Windows AI Foundry: A unified platform for local AI development on Windows, including Windows ML (built-in AI inferencing runtime) and tools for preparing and optimizing models for on-device deployment. Pricing: Free with Windows 11

Microsoft Entra Agent ID: Assigns unique identities to agents in an Entra directory, helping enterprises manage agents and avoid "agent sprawl." Pricing: Included with Microsoft Entra ID plans (starting at $6/user/month)

Microsoft's CTO Kevin Scott emphasized that memory is "conspicuously missing" in current agents, and they're introducing several memory-related technologies to address this, including structured RAG (Retrieval-Augmented Generation).

NVIDIA's AI-First Personal Computing Systems

NVIDIA just announced DGX Spark and DGX Station – personal AI supercomputers built by major manufacturers including Acer, ASUS, Dell, GIGABYTE, HP, Lenovo, and MSI.

DGX Spark:

• NVIDIA GB10 Grace Blackwell Superchip and fifth-generation Tensor Cores

• Up to 1 petaflop of AI compute and 128GB of unified memory

• Available starting July (reservations open now)

• Pricing: Starting at $14,999 for base models

DGX Station:

• NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip

• Up to 20 petaflops of AI performance and 784GB of unified memory

• NVIDIA ConnectX-8 SuperNIC supporting networking speeds up to 800Gb/s

• Can be partitioned into seven instances with Multi-Instance GPU technology

• Available later this year

• Pricing: Expected to start around $149,000

Jensen Huang called these systems "direct descendants of the DGX-1 system that ignited the AI revolution," designed to "power the next generation of AI research and development."

The Best AI Models Right Now (May 2025)

Major Model Releases This Month:

• OpenAI o3 & o4-mini (May 8): First AI models that can "think with images" – integrating visual information directly into reasoning chains. Users can upload whiteboards and sketches for AI analysis, and the models can rotate, zoom, and use other image-editing tools.

• Alibaba's Qwen3 (May 5): Significantly narrows the gap with leading US firms with advancements in cost efficiency and multilingual capabilities.

• Google Gemma 3 (Early May): Latest family of open AI models optimized for developer flexibility and performance across tasks.

• Google's AMIE (May 1): AI-powered medical imaging assistant capable of interpreting X-rays and MRIs, automatically highlighting abnormalities to help clinicians.

Major Funding & Industry Moves

• Harvey AI: Legal tech startup in talks to raise $250M+ at a $5B valuation (led by Kleiner Perkins and Coatue). They recorded $75M in annualized run-rate revenue in April and recently started using models from Anthropic and Google in addition to OpenAI.

• Shield AI: Autonomous drone startup raising $200M at a $5B valuation from Airbus, L3Harris, Lockheed Martin, and others. They make Hivemind, an AI pilot for autonomous drone swarms that can operate without GPS or communications.

• Innovaccer: Healthcare AI startup raised $275M Series F to expand AI and cloud capabilities, including adding multiple copilots and agents to its offerings.

• Mistral AI: French AI startup confirmed plans to go public instead of being acquired. CEO Arthur Mensch: "We are not for sale." They recently opened an office in Singapore to target Asia-Pacific.

• IndiaAI/Sarvam AI: The Indian Government's IndiaAI initiative provided 200 crore worth of GPUs to Sarvam AI, significantly boosting the company's AI research and development capabilities.

Quick Hits:

• API.market hit 4,500 users! We're offering one-time license fees for AI models with a 10X price reduction compared to alternatives. Some customers are using their text-to-video generation for 50K videos per day, saving over $200K USD.

• Amazon launched "Enhance My Listing," a new generative AI tool to transform and optimize product listings (May 8).

• FDA announced agency-wide AI implementation by June 2025, following a successful pilot that improved review efficiency (May 8).

• Microsoft will offer rival AI models from its own data centers, including those from Elon Musk's xAI, Meta, Mistral, and Black Forest Labs.

That's all for this week. As always, if you have questions or want to chat about any of these developments, just hit reply.

Until next time,
Shashank Agarwal
Founder/CEO API.market/Noveum.ai