Why Traditional PM Assumptions Are Dead: Inside Anthropic’s Product Management

Anthropic’s product manager Cat Wu shows that rapid AI model advances force a shift from long‑term roadmaps to short‑cycle planning, prototype‑first development, continual re‑evaluation of features, and minimal viable solutions, demanding PMs to abandon perfectionism and focus on decisive, judgment‑driven work.

AI Engineering
AI Engineering
AI Engineering
Why Traditional PM Assumptions Are Dead: Inside Anthropic’s Product Management

Data Shows 41× Growth

METR measurements indicate Sonnet 3.5 completed a human‑level task in 21 minutes, while Opus 4.6 can handle a task that would take a human nearly 12 hours—a 41‑fold increase over 16 months.

New Product Management Workflow

Claude Code product management workflow
Claude Code product management workflow

Claude Code and tools like Cowork blur traditional role boundaries across the product lifecycle.

Cat Wu organizes work around three products—Claude .ai (chat collaborator), Claude Code (coding tool), and Cowork (knowledge‑work tool)—each serving distinct stages from ideation to prototyping to daily tasks.

Facing Exponential AI Growth

METR task completion time chart
METR task completion time chart

According to METR (2026, March), Opus 4.6 can complete software tasks that would require a human almost 12 hours, whereas the initial Claude Code built on Sonnet 3.5 handled only 21 minutes, confirming the 41× improvement.

Four Fundamental Shifts

1. Short‑cycle planning replaces long‑term roadmaps

Instead of a fixed roadmap followed by a PRD and months of development, the team encourages “side‑track tasks”—short, self‑directed experiments undertaken by engineers, PMs, and designers.

Features such as the Claude Code desktop app, AskUserQuestion tool, and Todo list emerged from this approach.

2. Prototypes and evaluation over documentation

Stand‑up meetings become demo sessions for new ideas; only features that receive real user engagement after internal trials are refined and widely shared, because an afternoon‑long prototype carries minimal risk.

3. Re‑examining every existing feature with each new model

Each model release prompts a review of current capabilities. For example, when users built web apps with Claude Code and manually switched to Chrome, the team recognized the need for native Chrome integration.

4. Build the simplest viable solution

When a workaround circumvents a model limitation, the next model may natively support the behavior, rendering the workaround obsolete. Claude Code’s early system prompt that reminded agents to update their Todo list was later dropped as Opus 4.6 added native support, reducing system prompts by 20 %.

New Product Management Rhythm

PMs must relinquish perfectionism, identify a few truly non‑negotiable items, and release everything else. As Cat Wu puts it, “When a product manager can turn an idea into a working prototype in an afternoon, the gap between ‘what if…’ and ‘try this’ virtually disappears.”

In the AI era, execution can be infinitely supplied; judgment becomes the scarce resource.

AIRapid Prototypingproduct managementAnthropicClaude Code
AI Engineering
Written by

AI Engineering

Focused on cutting‑edge product and technology information and practical experience sharing in the AI field (large models, MLOps/LLMOps, AI application development, AI infrastructure).

0 followers
Reader feedback

How this landed with the community

Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.