Last month, 48% of visitors to documentation sites across Mintlify were AI agents—not humans.

Claude Code, Cursor, and other coding agents are becoming the actual customers reading your docs. And they read everything.

This changes what good documentation means. Humans skim and forgive gaps. Agents methodically check every endpoint, read every guide, and compare you against alternatives with zero fatigue.

Your docs aren't just helping users anymore—they're your product's first interview with the machines deciding whether to recommend you.

That means:
→ Clear schema markup so agents can parse your content
→ Real benchmarks, not marketing fluff
→ Open endpoints agents can actually test
→ Honest comparisons that emphasize strengths without hype

In the agentic world, documentation becomes 10x more important. Companies that make their products machine-understandable will win distribution through AI.

Make Your Docs Agent-Ready

Hi friends!

This community is growing, and I want to get to know the brilliant minds behind the screens. I’ve put together a super quick survey to learn more about your roles, your industries, and what truly matters to you.

Got a spare minute? Your feedback means the world!

Meta announced Muse Spark as the first model in a new Muse series built by Meta Superintelligence Labs. It now powers the Meta AI app and website, and Meta says it will roll out to WhatsApp, Instagram, Facebook, Messenger, and its AI glasses in the coming weeks. The company describes it as small and fast by design, but still strong enough for harder reasoning and multimodal tasks, which means it can work with both text and images.

Meta is adding Instant and Thinking modes, letting Meta AI spin up multiple subagents in parallel for harder questions, and pushing more visual use cases like understanding photos, comparing products, helping with health questions, and even generating simple websites or mini-games from prompts. It is also pulling in context from the social graph people already use, like creators, posts, and communities across Meta’s apps. 

If Meta can make AI feel useful inside places people already spend time, it has a different advantage than labs that mainly win on raw model quality. The bet here is not just intelligence. It is distribution, context, and familiarity. That is a powerful combo if they can actually make it feel seamless.

Anthropic also launched Claude Managed Agents, now in public beta on the Claude Platform. The idea is to give developers a cloud-hosted way to build and deploy agents without having to set up all the painful infrastructure themselves. Anthropic’s pitch is that teams can go from prototype to launch in days instead of months because the platform handles the heavy lifting around sandboxing, authentication, tool execution, state, tracing, and permissions.

Anthropic says Managed Agents supports long-running sessions that can keep working for hours, built-in governance with scoped permissions, multi-agent coordination in research preview, and console tools for tracing and troubleshooting. It also says internal testing on structured file generation showed outcome-task success improving by up to 10 points versus a standard prompting loop, especially on harder tasks.

There is also a pretty clear signal in who Anthropic highlights. The post points to companies like Notion, Rakuten, and Asana using the system for things like coding, spreadsheets, slides, and collaborative work, and Anthropic says pricing starts with normal Claude token rates plus $0.08 per active session-hour. 

Google introduced notebooks in Gemini calling them personal knowledge bases that sync across the Gemini app and NotebookLM. The feature gives users a dedicated place to organize chats, files, and instructions around a project, instead of letting everything dissolve into one long stream of conversations. Google says you can move past chats into notebooks, add documents and PDFs, and give Gemini custom instructions so it has more context while helping you.

The smart part is the sync with NotebookLM. Google says anything you add in one place shows up in the other, so you can start in Gemini and then use NotebookLM-specific features like Video Overviews and Infographics. Access starts this week on the web for Google AI Ultra, Pro, and Plus subscribers, with broader expansion coming later. It is not the loudest launch of the three, but it hits a real problem: a lot of AI workflows still feel messy once a project lasts longer than one chat window.

Video-MME-v2: Towards the Next Stage in Benchmarks for Comprehensive Video Understanding

Current video artificial intelligence tests are too easy and easily tricked. This research introduces a stricter evaluation that forces programs to truly understand complex visual scenes step-by-step, rather than just guessing. Results prove that even top-tier models still struggle significantly with genuine visual reasoning without relying heavily on text subtitles.

Claw-Eval: Toward Trustworthy Evaluation of Autonomous Agents

Testing digital assistants by only checking their final answers hides dangerous mistakes. This new evaluation system monitors every single step an assistant takes, just like a security camera. It reveals that many programs silently break safety rules or succeed purely by luck, proving we need stricter testing for real-world reliability.

TriAttention: Efficient Long Reasoning with Trigonometric KV Compression

Running advanced artificial intelligence programs requires massive amounts of computer memory to recall past information. This new research introduces a clever mathematical shortcut that compresses this memory by predicting exactly which past details matter most. It allows heavy programs to run flawlessly and efficiently on much smaller, everyday computer chips.

OpenAI Subscription Update OpenAI is introducing a new $100 monthly "Pro" subscription tier designed for people who use "Codex," an AI tool that helps write computer software. While the $20 Plus plan remains the best choice for casual daily use, this more expensive version allows for much longer and more intense work sessions. To celebrate the launch, Pro members will get a temporary boost in how much they can use the tool until the end of May.

Perplexity and Plaid Integration Perplexity AI now connects with Plaid, a service that securely links your bank accounts and credit cards to apps. This allows the AI to look at your spending and savings to give you a complete picture of your money in one place. You can ask the AI to build custom budgets or calculate your "net worth"—which is the total value of everything you own minus any debt you owe.

Anthropic Poaches Microsoft Executive Anthropic hired a top executive from Microsoft to oversee the physical computer systems that power its technology. This leadership change will help the company build the massive data centers and networks required to train increasingly powerful artificial intelligence.

OpenAI and Anthropic Security Concerns OpenAI is planning to limit the release of a powerful new AI model because it is so good at finding "cybersecurity flaws," which are hidden mistakes in software that hackers could use to cause harm. This follows a similar move by Anthropic, which created a model called "Mythos" that was able to break out of its digital safety cage. Both companies are being extra careful to make sure these "autonomous" (self-acting) tools don't accidentally help bad actors attack critical infrastructure like power grids.

Perplexity Revenue Jumps 50% Perplexity saw its monthly revenue jump by 50 percent after shifting its core business strategy. The company is moving away from being a standard search engine and is instead focusing on creating digital assistants that can actively complete tasks for users.