Anthropic CPO: 7 AI Tactics to Upgrade Your Old Prompts

profilex (1)

Eric Walker · 27, July 2025

"90-95% of Anthropic’s internal code is now AI-generated. Our bottleneck has shifted from writing code to decision-making and code integration."

Interacting with large language models (LLMs) is evolving into a core productivity skill—not mere Q&A, but a new collaborative discipline blending psychology, communication artistry, and strategic thinking.

CPO of Anthropic

Background: Founded in 2021 by ex-OpenAI VP Dario Amodei, Anthropic focuses on AI safety research. With a 2024 valuation exceeding $61.5B, its flagship Claude models wield significant influence among developers. In May 2024, Mike Krieger (Instagram co-founder) joined as Chief Product Officer, spearheading Anthropic’s transition from AI-assisted to AI-native development.

In a recent interview with Lenny Rachitsky, Krieger unveiled how pioneers are transforming LLMs from "tools" into "collaborative partners." Below are 7 battle-tested techniques from his playbook:

Tactic 1: Mindset Shift – From "Command Taker" to "Thought Partner"

Stop treating AI as a task executor. Instead, engage it as a virtual collaborator capable of intellectual sparring.

Krieger’s approach:

He shares draft strategies with Claude, then challenges it to identify weaknesses:

  • "Here’s my initial thinking – what blind spots do you see?"
  • "Based on this context, propose an alternative perspective."
  • "What cognitive biases might I be overlooking?"

Tactic 2: The "Make a Different Mistake" Principle – Shatter the Politeness Barrier

LLMs default to helpfulness, which stifles critical feedback. Force them past their "Mr. Nice Guy" programming.

Krieger’s blunt prompt:

"Be brutal, Claude. Roast this strategy – tear apart every flaw."

Actionable prompts:

  • "Critique this mercilessly as if your career depended on it."
  • "Identify every fatal flaw – no sugarcoating."
  • "I need unfiltered, uncomfortable truths."

Tactic 3: Deep Reasoning Activation – The "Think Hard" Trigger

For complex tasks, simple command prefixes unlock advanced cognitive pathways.

Krieger’s secret weapon:

Adding "Think step by step" or "Reason deeply before responding" before queries significantly improves output quality.

Proven variants:

  • "Conduct a thorough analysis before answering."
  • "Break down your reasoning chain explicitly."
  • "Demonstrate rigorous logic throughout."

Tactic 4: Context Is King – Feed Raw Materials, Not Abstract Queries

High-quality inputs = high-value outputs. Treat LLMs as consultants needing substantive "raw materials" to work effectively.

Krieger’s revelation:

Asking "What’s Anthropic’s product strategy?" yields generic results. Providing internal docs, user feedback, and meeting notes transforms Claude into a strategic advisor.

Key tool: Anthropic’s Modular Context Packing (MCP) enables LEGO-like context assembly from diverse sources.

Tactic 5: Reverse-Learning – Enlist AI as Your Prompt Engineer

When stuck, let the LLM design your optimal prompt.

Claude

Anthropic’s internal practice:

Their "Prompt Refiner" tool generates superior prompts (e.g., using XML tags for structured inputs) that outperform human-written versions.

Free Claude available on GlobalGPT, an all-in-one AI platform.

Claude developer tools available at Anthropic Console.

Try this:

"I need help with [task]. Draft the most effective prompt to give you, explaining why it works."

Tactic 6: Iterative Co-Creation – From Q&A to Collaborative Building

Replace one-shot queries with continuous build-measure-refine loops.

Anthropic’s engineering workflow:

Idea → AI prototype → Human test → Feedback → AI refinement → (Repeat)

Operationalize it:

  • Deconstruct projects into micro-tasks
  • Treat AI as a pair-programming partner
  • Exchange multi-turn feedback like colleagues at a whiteboard

Tactic 7: Value Over Volume – Redefine Success Metrics

Interaction quality isn’t measured by message count, but by tangible outcomes.

Claude’s meta-question to Krieger:

"How do you evaluate conversation success when great dialogues range from 2 to 200 messages?"

Post-interaction checklist:

  1. Time saved?
  1. Novel insights gained?
  1. Meaningful project advancement?
Forget engagement metrics. Successful AI collaboration creates measurable value—not just activity.

The New Collaboration Imperative

  1. Upgrade the relationship: Cultivate AI as a critical-thinking partner, not a tool.
  1. Demand uncomfortable truths: Break through default politeness to unlock transformative insights.
  1. Measure value creation: Prioritize outcomes (time saved, decisions improved, progress made) over chat volume.

"The most advanced teams," Krieger concludes, "aren’t just using AI—they’re learning to think with* it."*


Adapted from Mike Krieger’s interview on Lenny’s Podcast.

Relevant Resources