Mike Krieger doesn’t think Claude AI, Cowork, and Claude Code will continue to exist as three separate products. He told me at Anthropic’s developer conference last week that Anthropic is “shipping our harness strategy” rather than a product, and that the current separation between the products is “a broken abstraction we need to fix.”
Mike joins this week’s episode of ACCESS, the podcast I co-host with Ellis Hamburger. You can watch the conversation above or listen to it in your podcast player of choice by clicking this link.
After joining Anthropic in May 2024 as CPO, when the product and engineering org was only about 30 people, the Instagram co-founder moved about four months ago into Labs, the internal moonshot group that produced Claude Code, MCP, Agent Skills, and Computer Use. He described Labs as doing two things: “closing the gap” between what Claude can theoretically do and how it shows up in daily life, and acting as “advanced scouts” for where the models need to go next.
Last fall, the Instagram co-founder told Anthropic co-founder and president Danielle Amodei that he wanted to leave Anthropic to start a new company. The product org had grown to several hundred by then, and he’d drifted out of what his coach calls his “zone of competence.” He ended up staying to run a new incarnation of the Labs team, which now runs on two-week sprints with mostly former founders. The internal mission, in his words, is “find the next Claude Code.”
In practice, that means an “agent plus canvas” pattern for productivity software (Claude Design is the first instance) and long-horizon agents, the Labs project Krieger was leading, whose learnings became Managed Agents, and the screenshot-verification step Boris demoed onstage. He’s also pushing on MCP as consumer distribution: a nonprofit search tool that briefly surged to the top of Anthropic’s Connector list is, in his view, the early shape of how consumer AI breakouts will happen.
On Mythos, Krieger pushed back on the idea that Anthropic is sandbagging the model for safety reasons: “The internal belief is we should be able to release a Mythos-class model safely. And the fact that we haven’t yet is bad.”
Two more things from the convo with Mike that stood out:
Krieger abruptly stepped off Figma’s board after Anthropic launched Claude Design, a move that tanked Figma’s stock and was read as Anthropic coming for every vertical. His framing is that Anthropic ships first-party products when they “say something new,” and Claude Design (agent-first mockups for non-designers like him) is a different use case from Figma’s more polished production and collaboration work, not a replacement. The goal isn’t to be “the only product” in any space, which he said, “would be a tragedy.”
Health is the consumer white space that Krieger keeps coming back to as a place where Anthropic should do more. He said the Opus 4.7 jump has finally made genetic and lab-data analysis with AI “actually useful versus cute” for the first time, and pointed to Anthropic’s January Function Health partnership as the starting point for what could become a much bigger first-party health push. The other product directions he kept coming back to as natural expansion areas for Anthropic: AI sourcing representative groups for civic debate (”keep the humans in there but make sure we’re listening to the right voices”).
Listen or watch wherever you get podcasts.
Ellis Hamburger and I had a blast last night at the ACCESS live show and launch party at Notion HQ. Thanks to everyone who came out and to our special guest, Substack CEO Chris Best. Our conversation with Chris will be published in the feed next week.
Sources is a newsletter by Alex Heath about the AI race, featuring scoops, unique analysis, and exclusive interviews. Every week, Sources is read by thousands of decision makers in tech, finance, policy, and media. Click here to learn more or inquire about a sponsorship.