Don't be a meat proxy

The Origin: A Painfully Accurate Metaphor

On his blog, Simon Willison shared a term coined by Niklas Gruhn—'meat proxy'—describing people who treat AI like a search engine or translator, copy-pasting its output verbatim and forwarding it to colleagues, clients, or bosses without adding any value. They become a useless middleman, a human proxy server.

The metaphor stings because it describes a real problem. As tools like ChatGPT and Claude become ubiquitous, more people rely on AI for emails, reports, code, and even creative work. But many treat AI as a faster typist rather than a tool that demands scrutiny and mastery.

The Breakdown: What's the Right AI Workflow?

Willison emphasizes that using AI itself is fine—what matters is what you do with its output. He proposes a simple yet powerful four-step process:

1. Prompt: Ask AI your question or request. This step requires skill, but it's not the finish line.
2. Read: Carefully read what AI generates. Don't just skim and copy—go through it word by word.
3. Understand: Make sure you truly grasp what AI is saying. If something's unclear, follow up, research, and clarify.
4. Verify: Check whether the output is accurate, reasonable, and meets your needs. AI makes mistakes, fabricates information, and gives answers that sound right but aren't.

The final step is expressing it in your own words. This isn't just to avoid plagiarism—it's because your own language, knowledge, and judgment are what actually bring the output to life.

Trend Insight: The Stronger AI Gets, the More Precious Human Judgment Becomes

This perspective reveals a deeper trend: as AI tools grow more powerful and 'user-friendly,' the biggest mistake people make isn't failing to use AI—it's over-trusting and over-depending on it.

This mirrors a concept from autonomous driving: 'automation complacency.' When systems seem to run smoothly, human operators unconsciously let their guard down—until the system suddenly fails and they can't react in time.

The same thing is happening in knowledge work. AI-generated reports look professional, code appears to run, translations seem fluent—but these appearances can be traps. If you use them without verification, you're essentially lending your credibility to AI's mistakes.

Practical Value: From 'Meat Proxy' to AI Wrangler

For IT professionals, this reminder applies to several real scenarios:

• Code Development: After generating code with AI, don't commit it directly. Understand what each line does, test edge cases, and ensure there are no security vulnerabilities.
• Technical Documentation: When AI drafts documentation, use your expertise to add details, correct terminology, and ensure technical accuracy.
• Team Collaboration: Before forwarding AI-generated content to colleagues, summarize the key points in your own words to show you've reviewed it.
• Client Communication: Never send AI's raw output directly to clients. Your professional judgment and personalized expression are what they're paying for.

Counterintuitive: Slower Is Faster

Some might think: 'AI is supposed to save me time—if I have to spend time verifying and rewriting, what's the point?'

This is a misconception. AI's real value isn't letting you skip thinking—it's giving you more time to think about what matters. If you treat AI as an accelerator rather than a thinking aid, you'll quickly become an efficient 'meat proxy'—good at forwarding, but unable to provide real insight.

True efficiency gains come from AI handling tedious drafting and information gathering, freeing up your time for higher-level judgment, creativity, and decision-making. That's the right way to collaborate with AI.