Microsoft Build 2026: The Agent Era Is Here
Microsoft kicked off its annual Build developer conference today at San Francisco's Fort Mason Center, and the message was unmistakable: AI agents are no longer a preview feature — they are the platform.
CEO Satya Nadella's keynote, which the company billed as "no fluff," delivered a stack of announcements that together paint a picture of an operating system, productivity suite, and cloud platform being rebuilt around autonomous AI agents.
Windows Agent Framework Goes Open-Source
The biggest surprise was the open-sourcing of the Windows Agent Framework, a set of APIs that let developers build AI agents capable of interacting with the Windows desktop at the OS level. Alongside it, Microsoft announced a Windows Agent Store — essentially an app store for autonomous agents — and new computer-using agent (CUA) capabilities that let AI navigate desktop applications visually, just like a human user would.
Project Polaris: Microsoft's Own Coding Model
In a move that signals a significant shift in Microsoft's relationship with OpenAI, the company unveiled Project Polaris — an in-house AI coding model that will replace GPT-4 Turbo as the default engine for GitHub Copilot starting August 2026. While Microsoft hasn't disclosed Polaris's architecture, the decision to build an internal model for one of its most popular developer tools suggests the company is hedging its bets on AI supply chains.
Office 365 Gets Agent Mode by Default
Copilot's Agent Mode is now the default experience across Word, Excel, and PowerPoint for enterprise users. As Nadella described it, Office is moving from "synchronous assistants" to "async coworkers that can execute long-running tasks across key domains." Azure AI Foundry, the consolidated platform for enterprise AI, also reached general availability with its Agent Service.
Source: ChatForest, Windows Forum, Notebookcheck
Google's Gemini 3.5 Flash: A Flash-Tier Model That Beats Pro
Google's Gemini 3.5 Flash is now generally available across the Gemini app, Google AI Studio, the Gemini API, and AI Mode in Google Search — and it's making waves for a surprising reason: it outperforms Google's own Gemini 3.1 Pro on coding and agentic benchmarks.
Announced at Google I/O on May 19, Gemini 3.5 Flash delivers frontier-level intelligence at four times the output speed of comparable models. Pricing is set at $1.50 per million input tokens and $9.00 per million output tokens, with a one-million-token context window. The model scored 76.2% on Terminal-Bench 2.1, marking the first time a Flash-tier model has surpassed a Pro-tier model on these workloads.
Google also launched Managed Agents in the Gemini API — enabling developers to build and deploy autonomous, stateful agents running in secure, Google-hosted sandbox environments. Gemini 3.5 Pro, currently in internal use at Google, is expected to follow sometime this month.
Source: Google Blog, TechCrunch
OpenAI Sunsets GPT-4.5 as GPT-5.5 Takes the Stage
OpenAI is officially closing the book on the GPT-4 generation. GPT-4.5 will be retired from ChatGPT on June 27, 2026, following a 30-day sunset period. The models remain available to paid users via model settings until then, but the writing is on the wall.
The replacement? GPT-5.5 Instant, which is now the default model in ChatGPT. According to OpenAI, GPT-5.5 delivers "clearer, more natural replies" and in-chat writing and coding blocks. Internal benchmarks show significant improvements in accuracy for legal, financial, and medical contexts — areas where hallucinations have been particularly problematic.
GPT-5.5 is also available for agent-building, with creators able to set the reasoning effort their agents use. The retirement of GPT-4.5 applies only to ChatGPT — API access remains unchanged for now.
Colorado's Landmark AI Act Gutted Before Taking Effect
In a striking turn for AI regulation in the US, Colorado's pioneering AI antidiscrimination law (SB 24-205) has been effectively killed before its June 30 enforcement date. The law, which would have required AI developers and deployers to exercise "reasonable care" to prevent algorithmic discrimination, was challenged in court by xAI with support from the US Department of Justice.
A federal magistrate judge stayed enforcement on April 27, and on May 14, Governor Jared Polis signed a replacement bill (SB 26-189) that strips out the original law's most ambitious provisions — including risk management programs, annual impact assessments, and the duty to avoid algorithmic discrimination. What remains is a narrower notice-and-transparency model.
The development is significant because Colorado's law was widely seen as a bellwether for state-level AI regulation in the US. Its replacement signals that the political appetite for comprehensive AI governance may be waning, even as the EU pushes ahead with its AI Act enforcement.
Source: Law and the Workplace, Brownstein
Quick Hits
Cognition AI raises $1B: The maker of Devin, the AI software engineer, closed a $1 billion round at a $25 billion pre-money valuation. Devin's annual recurring revenue has hit $73 million. (LLM Stats)
GitHub Copilot's token billing backlash: GitHub's switch to token-based billing for Copilot, effective June 1, has drawn criticism from developers who say it makes costs unpredictable compared to the previous flat-rate model. (LLM Stats)
SoftBank plans €75B AI data center buildout in France: SoftBank is planning up to 5GW of AI data center capacity in France, representing one of the largest single-country infrastructure commitments for AI to date. (LLM Stats)