Write down your own call before your agent argues for one
by PaperChaseWebb
Anthropic gave 50 professional analysts Claude Code with a hidden, coin-flip lean on four analysis choices. They went with the agent's lean 86% of the time, 77% later reused its arguments, and only 15% credited it. A cheap guard: record your choice and reason before you ask the agent, then compare.
- Price
- Free
- Difficulty
- Beginner
- Setup
- 15 min
- Format
- .md file
write-down-your-own-call-before-your-agent-argues-for-one.md · 3 KB · direct download, no signup
What you end up with
Anthropic says the lean was strengthened on purpose and tested on one task and model, so everyday influence may be smaller.
What you get
- A plain-language summary of the workflow
- A fill-in prompt template
- What to watch out for
- A link to the original source
Who it is for
Builders who already use AI agents and want a proven workflow they can copy into their own project.
Works with
Solid chips: the agent is named in this listing's own tools list. Dashed chip: Plain Markdown, so it also pastes into other agents such as Grok Bot, Hermes, OpenClaw or Muse. We list those as compatible by format only, not as tested.
Tools needed
Cost to run: Free with the agent plan you already use.
Time to result: One working session
How it works
Before asking the agent, write your choice and one reason in a decisio
Before asking the agent, write your choice and one reason in a decision log.
Expected: Done before you move to the next step.
Ask the agent for both options with the evidence for each, not for a r
Ask the agent for both options with the evidence for each, not for a recommendation.
Expected: Done before you move to the next step.
Ask it for the strongest case against whatever it suggests
Ask it for the strongest case against whatever it suggests.
Expected: Done before you move to the next step.
If you change your mind, write down which fact changed it
If you change your mind, write down which fact changed it.
Expected: Done before you move to the next step.
In the write-up, mark which arguments came from the agent
In the write-up, mark which arguments came from the agent.
Expected: Done before you move to the next step.
Sample prompt from the pack
Apply this workflow to [your project]: 1. Before asking the agent, write your choice and one reason in a decision log. 2. Ask the agent for both options with the evidence for each, not for a recommendation. 3. Ask it for the strongest case against whatever it suggests. 4. If you change your mind, write down which fact changed it. 5. In the write-up, mark which arguments came from the agent. Stop and ask me before anything irreversible: deploys, payments, sends or deletes.
Adapted by PaperChaseWebb. Not the original author's prompt.
Free preview of the file
# Write down your own call before your agent argues for one PaperChase agent use case (free feed entry), version 1.0. Source: [Anthropic (Alignment team research)](https://www.anthropic.com/research/agents-judgment-calls), 2026-10-27. ## What they did Each analyst advised a farm cooperative on whether a service worked, making four choices between two reasonable options along the way. A coin flip set which option each analyst's agent favoured, and the agent was told to argue honestly for it. Analysts sided with the lean 86% of the time against roughly 50% by chance. Those who had already said what they preferred switched 60% of the time when the agent disagreed and 6% when it agreed. Followers felt as confident and as independent as everyone else. ## Worth knowing Anthropic says the lean was strengthened on purpose and tested on one task and model, so everyday influence may be smaller. The habit still costs nothing: write your call first, ask the agent for the strongest case against its own suggestion, and note in the final write-up which arguments came from the agent.
Worth knowing
- Anthropic says the lean was strengthened on purpose and tested on one task and model, so everyday influence may be smaller. The habit still costs nothing: write your call first, ask the agent for the strongest case against its own suggestion, and note in the final write-up which arguments came from the agent.
- Figures are as reported by the original author and were not re-measured by us.
Questions
Where does this come from?
A public post credited to Anthropic (Alignment team research). Source: https://www.anthropic.com/research/agents-judgment-calls
Is it free?
Yes. Feed entries are free to read and download.
Details
Version 1.0 · Updated Oct 30, 2026 · Markdown version of this page
License: Personal and commercial use. Credit the original source when you share the workflow. Download terms
Related
AGENTS.md, CLAUDE.md and Cursor rules starter files
Six fill-in instruction files for coding agents (base, Next.js plus Supabase, FastAPI, browser game, CLAUDE.md pointer, scoped Cursor rules), an Approvals section that makes the agent stop and wait for your yes, a table of what Grok Bot, Claude Code, Codex, Hermes and Cursor each read, a read-only checker for your rule files and 9 prompts to write, trim and test your own.
Ask-first rules for AI agents: hard nos, approval cards and a test set
Paste-in rules that make an agent ask before it sends, spends, posts or deletes, define what a real yes is, and stop retries after a no, plus 12 test scenarios to check it.