Product Engineer

WorkOS launches atlas, AI agent for Slack team workflows. 57m ago Startups Tl;dr. WorkOS released Atlas, an AI agent platform for Slack that learns org context, supports custom agents with scoped permissions, and integrates with 150+ tools. Key points. * Custom agents with isolated memory, scoped integrations, and server-side secrets; no standing superuser access * Connects to 150+ tools out of the box plus any API; respects existing organizational permissions * Reusable skills across agents; agents learn from corrections within Slack conversations * Available now at atlas.workos.com with audit logs and platform-enforced safety guardrails Why it matters. For DevOps and engineering teams, Atlas provides a permission-aware AI assistant that integrates directly into existing workflows without creating shadow data copies or privilege escalation risks. The scoped agent model and audit logging address real security concerns in enterprise AI tooling, making it practical for regulated environments.

Homer Wang on building TinyFish and the future of AI agents. WorkOS CEO Michael Grinich interviews Homer Wang of TinyFish at HumanX 2026 about building AI agents and the evolving startup landscape. April 15, 2026 At HumanX 2026 in San Francisco, WorkOS CEO Michael Grinich sat down with Homer Wang, founder of TinyFish, for a candid conversation about web infrastructure, AI agents, and what it means to help machines work with the open web. The web as infrastructure for AI. The premise behind TinyFish is deceptively simple: the web is the largest database on the planet, but it was built for humans - not machines. "If you think about actually getting things done on the web, it needs a bit more than retrieval," Wang explained. "You need to act like a human: click on things, scroll, analyze, extract, reason on top of it, and then automate that entire flow. That's what TinyFish does - we give anyone a complete platform and infrastructure to automate anything on the web." Web scraping isn't a new idea, of course. Crawlers and bots have been around since the late '90s. But Wang is quick to distinguish what TinyFish does from those earlier approaches. "Developers have tried launching their own browser, bolting an LLM on top of it, trying to automate flows - but the problem doesn't go away no matter how much AI you throw at it or how much scaffolding you add," he said. "The fundamental principle of the web is that it was built in a way that protects its data. It's complex, it's hostile - and in one word, it's just simply hard and inaccessible for AI, machines, and agents to tap into." That's where the company name comes in. "If everyone else is busy building ships and submarines to explore the ocean - which is the web - we're busy getting you the tiny fish. We get you the native creature of the ocean itself, to work with it and do things on your behalf." What people are building on TinyFish. The platform is already powering a wide range of use cases. Wang described customers using TinyFish for live deep research, large-scale collection of unstructured web data, and continuous monitoring of how websites change over time. Enterprises like Google, DoorDash, and ClassPass are already on board. "You can apply that across every vertical and industry: healthcare, financial services, auto, and more," Wang said. "Anything that touches the web - that's TinyFish." Building for agents, not users. One of the most provocative ideas Wang raised was a challenge to how founders typically think about product metrics. "The old metrics - daily active users, people using your UI, coming to your platform - those days are numbered," he said. "You have to build your product for agents. It doesn't matter if you're building agents for agents: your product has to serve the machines, the other agents accessing you as a service." The shift Wang is describing is subtle but significant. Products are increasingly becoming tools and skills for general-purpose, user-facing agents - enabling them to automate specific categories of work. "That's what WorkOS has done, finding a strong vertical and being the best in it," he said. "That's exactly what we're trying to do with TinyFish for web automation." Closing the gap for the long tail. When Grinich asked how companies can make their products more agent-ready, Wang offered a nuanced take. Yes, some sites will adopt new standards like /markdown endpoints or emerging protocols that allow automated systems to browse, extract, and act on web content. But many won't - and that gap matters. "There are sites that won't add new protocols unless someone does it for them," he said. "A small salon, a small restaurant, a family business - they don't have the resources or technical awareness to do these things, but they still need to be accessible to future agents." Wang pointed to a concrete example: small hotels in Japan that even Google had trouble accessing for live pricing and availability. "Those small businesses can't adapt to new standards on their own. So what we do is send our web agents in, acting like humans: launch a page, click around, fetch live information, and bring it back." The result was better visibility and business outcomes for those hotels - without them having to change a thing. Grinich summarized it well: "In addition to people making their own sites agent-ready, you're democratizing access for agents to all these things that might otherwise be left behind as autonomous agents take on more tasks." Wang's response: "You summarized it much better than I did." This interview was recorded at HumanX 2026 in San Francisco.