Summary  
This chapter explains how to craft AI decision prompts by supplying real-world constraints, requiring balanced argumentation, and explicitly flagging unknowns to obtain context-specific, actionable recommendations.

General domain of usage  
Collaboration platform evaluation

Your manager's Slack message is the task. What you do with it next determines whether you get back an opinion or something worth actually bringing to a meeting.



## What a Decision Prompt Needs

* **The real constraints**: team size, current cost, what's already been tried, any deadline;
* **Both sides argued, not just one**: otherwise Claude just builds the case for whatever you implied you already lean toward;
* **What it doesn't know, flagged explicitly**: a confident answer that hides its own gaps is worse than one that admits them.



## Step 1 — Ask It Straight

> <b>Should we switch from Slack to Teams?

You'll get a generic pros/cons list that could apply to literally any company sizing up Slack vs. Teams — nothing about your actual team, your actual cost, or the fact that you already tried this once and it stalled out.



## Step 2 — Give It the Real Situation

> <b>We're a 40-person team currently on Slack, paid plan, $12.50/user/month. We already have Teams included in our Microsoft 365 license but haven't used it seriously. Sales wants to switch because they're tired of checking two apps, but engineering has built a lot of custom Slack integrations, and a partial migration attempt last year stalled after two months. Give me a straight recommendation: switch, stay, or something else. Argue the strongest case against your own recommendation, and tell me what you'd need to know that I haven't told you.



> ## Task
> Run both steps and compare them. Then check: 
> 
> - Did Claude actually argue against itself, or just add a token "however" and move on? 
> - Did it flag any real unknowns, or invent confidence it shouldn't have? 
> 
> If either answer feels too easy, push back and ask it to try again.

Master the skills needed to build reliable AI decision systems with Claude, tackling real business calls like switching tools, approving a hire, and renewing a vendor contract. Progress from prompt-only reasoning to tool-grounded analysis with live data, then package your approach into a reusable Claude agent with structured output and a no-invented-facts constraint. By the end, you'll write a full agent system prompt from scratch.

Discover how to build an AI decision system for a real tool-switch call. Start with prompt-only reasoning, ground it in live pricing and migration data via web search, then build a Tool Switch Advisor agent that routes tasks and cites its sources.

Explore the essentials of contextual prompting for hiring decisions. Compare a bare prompt to a context-rich one, analyze real team capacity data to test the case for a hire, then build a Hiring Decision Advisor agent around that data.

Dive into the fundamentals of building a Claude agent for a vendor renewal decision. Combine past notes, usage data, and market research into one recommendation, then write your own agent prompt with tool routing and a no-invented-facts rule.