You're not ambitious enough with Claude

I'll pull the linked post and HN thread so the paragraphs stay grounded in what's actually there—no invented versions or figures.Christine Zhu, an AI UX PM at Intuit, argues in a widely shared post that most people underuse Claude by confining it to busywork. Her largest gain in productivity and capacity came only after she stopped clearing a “dopamine backlog” of small, satisfying tasks and started assigning high-leverage product work she previously would not trust to a model. On Hacker News the item sits at 3 points with 0 comments under a Claude/Codex/Fable framing, while the original thread itself is the primary source.

The operational core is a three-layer product-work model adapted from Shreyas Doshi: optics, execution, and impact. For optics, she runs a Cowork scheduled task twice a week that pulls Slack, calendar, meeting notes, and git activity into a Google Sheet status board, then uses a skill to repackage updates by audience. For execution, a Monday context-dump skill reads the prior week across the same sources, scores patterns (one example flagged leverage below a ~70% target because alignment and ceremony took ~35–40% of calendar time), recommends priorities, and pairs with Jira planning and ticket creation; a monthly job mines support channels for themes and requests. For impact, she voice-dumps into a Claude project preloaded with product decks, roadmap, codebase summary, goals, metrics, and stakeholder maps, then spars on Fable at high or max to produce a first draft in a short break instead of days.

For engineers and builders, the claim is practical rather than philosophical: judgment is mostly pattern-matching over context, and a well-onboarded model can hold more of that context at once than a single PM routing five to ten work streams. Claude Code in a VS Code workspace becomes a feasibility sparring partner—architecture overviews, PRD debate, small PRs—while HTML artifacts and a Figma connector cut the cost of turning rough ideas into something visual before anything is shared. The system also turns weekly dumps into a rolling career coach for blind spots and review writing, which is less about automation theater and more about closing the loop between work product and self-assessment.

The competitive frame is not model marketing so much as workflow maturity. Notion’s cited survey of 6,000 professionals across 10 markets is used only as backdrop: if an org already provides Claude or ChatGPT plus connectors and MCP access, the differentiator is how far individuals push personal systems past optics into execution and impact. Fable is called out as calmer and more concise for hard thinking than earlier model behavior; Claude desktop Cowork handles scheduled automations, Claude Code owns repo-grounded work, and connectors (Slack, Google Drive, Outlook, Jira, GitHub, HeyMarvin) plus Obsidian daily notes and evergreen PM markdown folders keep context current. The stack is intentionally ordinary: the edge is ambition of the ask, not exotic tooling.

The practical takeaway is to treat setup as secondary and ambition as primary. Pick the highest-leverage task you have been delaying, load the context the model would need plus your current thinking, and ask what you are missing—then iterate skills and schedules only after that loop works. Watch whether teams keep using Claude as autopilot for status and tickets while reserving “real” strategy for humans, or whether they start treating impact-level sparring, context folders, and weekly pattern dumps as standard PM infrastructure. The failure mode Zhu flags is the next dopamine trap: polishing automations instead of raising the difficulty of what you ask the model to do.