We've raised $65 billion in Series H funding at a $965 billion post-money valuation, led by @AltimeterCap, Dragoneer, @Greenoaks, and @sequoia.
This investment will help us advance our research and expand our capacity to meet growing demand for Claude.
8. Spend at least 20% of your time improving the process
Continuously refine instructions, workflows, evaluation methods, and automation systems. Focus on eliminating recurring mistakes while keeping the overall process simple and efficient.
CEO of @NotionHQ hasn't written a single line of code since last summer.
The productivity advantage no longer comes from writing code.
It comes from designing systems where AI can work, execute, and deliver.
7. Remove yourself from the execution loop
Avoid manually driving every step. Agents should execute tasks, validate results, and provide evidence of completion. Your role is oversight, not execution.
6. Separate roles into specialized agents
Create dedicated agents for planning, implementation, review, testing, issue triage, and quality assurance. Build a workflow where each agent contributes its specialty.
5. Set up adversarial reviews
Use independent reviewer agents to compare completed work against requirements, identify gaps, and challenge assumptions before work is considered complete.
4. Spend your time writing planning documents
Instead of constantly monitoring the agent, invest in detailed planning documents. Good plans should be self-contained and include implementation details, interfaces, and end-to-end validation strategies.
3. Maintain a persistent task queue
Always keep a backlog of well-defined tasks ready. Each task should include objectives, validation criteria, and evidence of completion so the agent never runs out of work.
2. Keep sessions alive for days or weeks
Maintain long-running sessions so the agent can retain project context, conventions, architecture decisions, naming patterns, and coding styles over time.
1. Increase the task scope
Don't break work into tiny pieces. Give the agent larger, outcome-oriented tasks that would normally take a senior engineer days or weeks to complete.
@simonlast hasn't written a single line of code since last summer.
The productivity advantage no longer comes from writing code.
It comes from designing systems where AI can work, execute, and deliver.