TLDR 2026-02-20

Every B2B company hits the same inflection point — enterprise customers show up and they need SSO, directory sync, audit logs, and role-based access before they'll move forward. Most teams lose months building that infrastructure. It doesn't have to be that way.With WorkOS you get all of it. One platform for auth, identity, and security. Infrastructure for teams that ship fast and stay fast.OpenAI, Anthropic, Cursor, and Perplexity already chose WorkOS over building it themselves.Build faster with WorkOS →

Google's new Gemini 3.1 Pro is now rolling out in preview. The model's benchmark results show mostly modest improvements. The updated model is now available in AI Studio and the Antigravity IDE. Enterprise users will see the model in Vertex AI and Gemini Enterprise, and regular users can use it in both the Gemini app and NotebookLM. The API costs for developers have not changed, nor has the context window.

Amazon is now the biggest global company by revenue. Walmart was previously the leader for more than a decade. Jeff Bezos had carefully studied Walmart's founder, Sam Walton, and embedded many of his business strategies while building Amazon. Amazon's revenue increased at almost 10 times the pace of Walmart's over the past decade. Amazon is the biggest online retailer, attracting 2.7 billion visits to its websites and mobile apps each month, while Walmart is the biggest physical retailer in the world, with more than 10,000 stores and shopping clubs globally.

Scaling laws have defined tech in the last five years. Large language models have demonstrated consistent returns to scaling. However, these 'laws' apply to very few other fields. Little work has been done to systematically understand, in theory or in general, what causes scaling laws or determines their slope and robustness. This article looks at what causes scaling laws, what their preconditions are, when they stop, why they have worked in some fields but not others, and more.

Almost everyone agrees that AI policy should be evidence-based. However, science is often messy and full of caveats, which doesn't fit well with policymakers' demands for clean answers. Many AI benchmarks have saturated and can no longer give a clear 'no' on risks. Experiments show how much we still have to learn. We need to spend less time proving that today's AIs are safe and more time figuring out how to tell if future AIs are dangerous.

Traffic skyrocketing? If you're on Microsoft Azure, there's no reason to panic. When user demand spikes, Azure adjusts capacity in the background. Performance stays steady with no manual intervention, no late-night firefighting, and no scrambling to explain why the site went down during the big launch. Build without limits with Azure

The winning formula in early 2026 is structured context, tiered rigor, smaller teams with higher leverage, and relentless measurement of downstream effects. The teams that adapt AI tools without the underlying taste and discipline are discovering that AI just makes their existing problems bigger. Teams with strong engineering cultures, robust CI/CD, clear architectural standards, and effective review processes see AI compound their advantages. This is widening the gap between elite and average teams.

AI should be treated as an amplifier of human capacity rather than a replacement. AI can still act autonomously with specific tasks, but in a way that is an extension of human decision-making and context. This allows humans to do dramatically more work, more sustainably, with less injury and fatigue. The AI industry's dream of having a fully autonomous AI employee is seductive, but it will likely set us up for disappointment.

OpenAI's technology isn't unique. While it has a big userbase, there is limited engagement and stickiness, and no network effects. Its competitors have matched its technology and are leveraging their product distribution. Executing better than everyone else is certainly an aspiration, but it isn't a strategy.

ByteDance is hiring nearly 100 employees for its AI division in the US. The company's AI team, called Seed, was established in 2023 and has labs across the US, Singapore, and China. The open roles include various job responsibilities such as producing international data, advancing ByteDance's text, image, and video generation tools, AI research, and building models for drug discovery and design. The company's recent AI launches have thrust it into the spotlight in the US. There has been some controversy over the data used to train its models.

Tech stack complexity is out of control, and the last thing you need is a bigger ‘stack'. Microsoft Azure gives you a unified view of applications, cloud resources, and AI—so you can ship more software with less open tabs. Simplify with Azure

Software that can be generated in a weekend is disposable inventory, not something you can build a startup around.

Nine observations from building with AI agents over the last year.

Meta is separating its Quest VR platform from its Worlds platform, shifting the focus of Worlds to be almost exclusively mobile.

The most value has always historically accrued to the companies that did the hardest things.

There were around 1,500 federal trade secrets cases filed in the US last year, up 20% from the previous year.

The productivity gains from AI are real, but it won't be catastrophic, as it will just devalue mechanical skills while elevating judgment skills.