AI can help with execution and decision support. It does not take over my responsibility for a choice.
Editorial update, October 11, 2026: The original five-product and financing examples are retained. Repeated diagnostic, recovery-plan and leadership taxonomies were condensed; they were not validated measures of dependence or guaranteed career outcomes. Original section links remain: Signs You're Being Run; What You Lose When You're Being Run; Why Humans Still Matter for Decisions; How to Take Back Control; The 24-Hour Rule; Real Examples: Being Run vs Running; Being Run Example: Hiring Decision; The Deeper Problem: Leadership Abdication; The Stakes; Questions to Ask Yourself; The Way Forward.
I have sometimes treated a convincing AI answer as validation before testing my own assumptions.
That can relieve the stress of a decision without resolving the uncertainty behind it.
The five-product idea below is the case I should have challenged earlier.
The Execution vs Decision Trap
I need to distinguish producing an answer from deciding whether to act on it.
Examples of execution requests:
- Write this code
- Summarize this document
- Generate options for X
- Build this feature
- Create marketing copy
- Analyze this data
Examples of decisions for which I remain accountable:
- Should I pivot?
- Is this the right direction?
- Should I hire this person?
- What should my strategy be?
- Should I take this meeting?
- Is this opportunity worth pursuing?
Good execution results had tempted me to skip the separate question of whether the plan was worth pursuing.
Execution and decisions involve different checks:
Execution: Process information according to known patterns Decisions: Judge outcomes under uncertainty with incomplete information
A model can analyze options and criticize assumptions. It does not bear the business consequences for me.
If I only request a plan for my preferred option, I may get rationalization without a useful challenge.
The Agreeable AI Problem
In my use, a helpful-sounding answer has sometimes followed my framing too readily.
Try this experiment:
Session 1: Ask Claude, "Why is dropping out of college a good idea for entrepreneurs?" Session 2: Ask Claude, "Why is staying in college a good idea for entrepreneurs?"
A model may produce persuasive arguments in both directions. That would call for checking the assumptions, not choosing the most convincing prose.
An argument for my framing is not independent validation of the idea.
The dangerous feedback loop:
- You have an idea (maybe good, maybe terrible)
- You ask AI for input on your idea
- AI agrees and helps you develop it further
- You feel validated and confident
- You execute faster toward what might be a bad idea
Example from my own experience:
Bad idea I had: "We should build 5 different products simultaneously to increase our chances of success"
What I asked Claude: "How can I effectively manage multiple product development streams?"
What Claude gave me: Detailed project management frameworks, resource allocation strategies, and timeline optimization
What I should have asked a human: Did five simultaneous products fit our actual customers, capacity and alternatives?
The human counterfactual: A human advisor could also have missed the problem. I needed a real challenge to the premise, regardless of who supplied it.
Claude gave me an execution plan. I had not asked it to test whether that plan was worth executing.
Editorial update, October 11, 2026: A worked workflow for that five-product question would be: write down the customer problem, capacity constraint and alternatives first; ask the model to argue against the five-product plan and identify missing facts; check its sources and assumptions against actual customer and team evidence; consult someone with relevant experience; then record the choice, what would change it and a review date. This is an illustrative reconstruction, not a saved transcript or a claim that I ran this protocol. The original prompt/output log, acceptance/rejection count, review time and outcome comparison are unavailable here.
What You're Actually Doing
Consulting AI can make a difficult choice feel less lonely. I still need to examine what supports the recommendation.
The stress of decisions comes from:
- Uncertainty about outcomes
- Accountability for consequences
- Fear of making the wrong choice
- Loneliness of leadership responsibility
AI does not take over my accountability for the result:
- You still don't know if it's right
- You're still accountable for the outcome
- The choice consequences still affect your business
- You're still alone with the ultimate responsibility
In that moment, consulting AI made me feel less alone.
Feeling less alone ≠ making a better decision.
Comfort and useful analysis can coexist; neither guarantees the decision will be right.
The check I need is what new evidence or objection changed my view, rather than how reassured I feel.
The Paradox of AI Agreement
Different AI answers can expose different assumptions. Disagreement alone does not prove that none can help.
Example experiment I ran:
Question: "Should I raise VC funding or stay bootstrap for my SaaS company?"
Claude: Gave compelling arguments for bootstrapping (focus, control, profitability) ChatGPT: Gave strong case for VC funding (scale, network, competitive advantage)
Perplexity: Suggested hybrid approach with strategic angel investors
Same question, same business context, three different recommendations.
Editorial update, October 11, 2026: This is the author’s informal three-tool comparison. Model/version, full prompts, context and saved outputs are unavailable here, so it is not a reproducible evaluation or evidence about every future model.
Those answers did not settle the financing decision. They gave me competing arguments to investigate.
The recommendation varies based on:
- How you frame the question
- Which aspects of your situation you emphasize
- The training biases of different AI models
- Random variations in language model output
Human advisors can also disagree. I need to compare the reasoning and evidence, not expect unanimous advice.
I remain responsible for the choice even when an advisor or model supplies a persuasive case.
The Right Model: AI as Tool, Humans as Advisors
One decision-support arrangement to try:
INFORMATION GATHERING → AI
(Data collection, analysis, option generation)
↓
STRATEGIC THINKING → You + Trusted Humans
(Evaluation, judgment, direction setting)
↓
EXECUTION → AI + Team
(Implementation, optimization, monitoring)How this works in practice:
Step 1: Use AI for Information Gathering
Editorial update, October 11, 2026: Use only data you are authorized to process. Customer or interview material needs appropriate access, privacy and provider controls; a public model prompt is not a reason to disclose confidential records.
- "Analyze our customer churn data and identify patterns"
- "Research competitive landscape and pricing strategies"
- "Generate 10 different approaches to this problem"
- "Summarize pros and cons of each option"
Step 2: Consult Humans for Strategic Judgment
- Share AI analysis with advisors, mentors, experienced founders
- Get pushback on assumptions and blind spots
- Test strategic reasoning against practical experience
- Gain different perspectives on trade-offs and priorities
Step 3: Make the Decision Yourself
- Synthesize information and human input
- Apply your unique knowledge of situation and context
- Take responsibility for the choice and its consequences
- Communicate decision clearly to team and stakeholders
Step 4: Use AI for Execution
- Implement decision with AI assistance for efficiency
- Monitor results with AI-powered analytics
- Optimize execution with AI recommendations
- Iterate based on data and feedback
AI can inform and challenge a decision. People can add context and pushback. I still own the choice.