PM automates 80% of workday with Claude: blueprint for AI-native GTM teams

The Gist

• Daniel Blum built a Claude system handling 70-80% of his PM work at Melio
• Self-improving AI manages Notion, Slack, email with 15-minute onboarding for colleagues
• System identifies knowledge gaps daily and updates context files autonomously
• Proves ARM architecture beats tool selection for AI productivity gains

• Daniel Blum, a PM at Melio, has automated 70% to 80% of his workday using Claude, a self-improving AI system.
• The architecture of an AI system matters more than the specific AI tool used, as it enables self-improvement and integration with existing tools.
• Context for AI systems requires ongoing updates to maintain accuracy and relevance, achieved through recurring refreshes of context files.
• AI systems can identify gaps in their knowledge and proactively ask targeted questions to fill those gaps, improving their effectiveness over time.
• The value of personalized AI systems builds slowly but becomes enormous once context and workflows are established.
• Self-improvement loops in AI systems learn from the differences between AI drafts and final human edits, refining future outputs.

• Focus on building a robust architecture for AI systems rather than selecting the perfect AI tool.
• Implement recurring updates to context files to keep AI systems aligned with current company knowledge.
• Develop self-improvement loops that learn from the differences between AI drafts and final human edits.
• Create personalized onboarding flows for AI systems to ensure adoption across teams.

• 70% to 80% (Percentage of Daniel Blum's workday automated by Claude)
• 15 minutes (Time it takes to onboard other Melio employees to the AI system)
• 1 day (Time Daniel Blum now takes to accomplish what previously required a week)