Key Takeaways

  • Anthropic runs multiple parallel product tracks instead of forcing teams to pick a single theoretical product roadmap upfront.
  • Each lab project has a designated Bet Lead who acts as the sole directly responsible individual (DRI) with authority to double down, add head count, or shut the experiment down.
  • Early experimentation fractured internal architecture: Anthropic discovered their Chat and Co-work products ran on separate memory stores, different file systems, and divergent MCP implementations.
  • To solve technical fragmentation without slowing down experiments, Anthropic separated frontier exploration from a dedicated foundations team that unifies memory and file primitives.
  • Managing parallel bets requires leadership to openly acknowledge team ambiguity, setting explicit expectations that overlapping teams will not all survive to production.

Why Forcing Early Product Alignment Kills Exploration

Most software leaders try to eliminate duplication early. In frontier AI, that instinct backfires. When model capabilities shift every quarter, nobody knows the final shape of an agentic workflow or an interface before users touch it.

Ami Vora argues that premature focus kills good ideas before they meet reality. As Vora puts it: “I would much rather have like five overlapping products that all work in different ways than like one overly constrained one that we never even gave a chance to actually like meet the market.”

Running overlapping tracks feels chaotic to traditional product managers. It creates competition between internal teams and burns engineering cycles on dead ends. But in fast-moving fields, testing five working variations beats debating one hypothetical winner in committee rooms for six months.

Assigning Bet Leads and Decoupling Infrastructure

Parallel exploration fails if ownership is fuzzy. To keep five simultaneous projects from collapsing into confusion, Anthropic assigns a single Bet Lead to every lab initiative.

Mike Krieger explains the scope of this role: “The bet lead is a really important role and they are the DRI for saying like I think we should double down on this or like actually we should wind this down. This team needs more people, this team needs less.”

The Bet Lead has full ownership over the hypothesis. But rapid product iterations can silently break your platform underneath if each team invents its own stack.

Anthropic saw this problem emerge firsthand between product surfaces. Krieger noted: “There was a point where like chat and co-work had different memory systems and different MCP and different ways of storing files. And it was like, okay, like for anybody to experiment on top of the surface, they're going to have a really hard time because if they built a third thing that then doesn't share memory with any of those, for example, it's going to feel disconnected.”

Their solution was an organizational split. Anthropic created a distinct foundations team. While Bet Leads race to test experimental interfaces, the foundations team builds shared primitives for memory, file storage, and the Model Context Protocol (MCP). As Krieger notes, solving memory across all surfaces is complex for eight different reasons, which is why it belongs on an infrastructure team rather than slowing down frontier labs.

What to Do With This

Audit your active product experiments by listing each project alongside its named Bet Lead and its core infrastructure dependencies. If two teams are building conflicting memory, file storage, or API connectors, assign one shared owner to standardize those components this Friday while leaving the product loops decoupled.