Last week was about using AI to move faster. This week is about something I use it for just as often: thinking something through before I have to defend it in a room.
In my own work, I use AI as a thought partner more than I use it as an accelerator, probably by a real margin. That side of it does not get talked about much, mostly because it is hard to put a number on the ROI the way you can with hours saved on a task.
In the free section: where thought partner use actually shows up for a CFO, and the four ways it quietly breaks down if you are not watching for it. In the paid section: six ready-to-use prompts, the actual Devil's Advocate skill file I run before anything reaches a board or a client, and a five-point checklist to run before you trust what AI just told you.
Events and updates
Tomorrow, August 26: The FP&A Series, Building Financial Intelligence. Free event with Carolina Lago, Paul Barnhurst, Mario Vasquez, and me. My session is The Human Layer of Owning AI Output. Free to join.
September 23, in Houston: TXCPA Houston, CFO Controllers Peer Group. Implementing AI with Confidence: A CFO's Playbook for Scaling AI Adoption. Register here.
The Claude in Action cohort is full. The next one runs in November.
Want to work with me directly? Reply to this email for a 1:1 session or a custom program for your organization.
Using AI as a Thought Partner
The tell is in the question you ask. An accelerator question sounds like "draft this" or "summarize this." A thought partner question sounds like "help me think this through." One asks for a deliverable. The other asks for pressure.
Where this shows up most for a CFO:
Negotiation prep. Find the weak point in your own position before the other side does.
Board presentation prep. Stress test the numbers and assumptions that are going to draw pushback before you are standing in front of the room defending them.
Blind spots. Someone, or something, naming the gaps you cannot see yourself, whether you are new in the role or ten years in.
Business case stress test. A real pressure test before it goes upstairs, instead of a friendly read from someone who already agrees with you.
Budget or forecast check. A number you have been sitting with long enough that you have stopped questioning whether it still holds.
Hard conversation prep. Board pushback, a difficult performance review, or an investor question you already know is coming.
Strategic pivot test. Whether the logic actually holds up once someone outside your own head is pushing on it.
Major decision stress test. An acquisition, a vendor switch, a key hire, when the cost of being wrong is high enough to justify the extra step.
Where this breaks down, in ways worth watching for:
Agreement bias. Ask "is this a good idea" about five ideas that flatly contradict each other, and you can get five yeses in a row.
Confirmation bias in research. Ask it to research a position you already hold, and it can hand back a summary that supports what you believed walking in, without ever surfacing the counter-evidence that exists.
Overconfidence. It can state something with more certainty than the situation actually warrants, no hedge, no "it depends," just a clean, confident answer.
Anchoring. It tends to build on the frame you give it instead of questioning the frame itself. Present a decision as already made, and it will help you defend it rather than test it.
I have run into agreement bias myself more than once, when I was testing five variations of the same idea and getting five confident yeses that could not all be right. It is an easy trap because the output sounds reasonable each time.
None of this means the tool is broken. The fix sits with you.
Ask differently. A vague "what do you think" invites agreement. Ask Claude to build the strongest case against your position, to name what a skeptical board member would push back on, or to argue the other side before it argues yours.
Judge what comes back. A sharper prompt gets you sharper pushback, not a verdict. Catching where the model's confidence runs ahead of its actual certainty is still your job, every time you use it this way.
Close the loop on recurring work. If you are using AI as a thought partner on something ongoing, a board narrative, a client relationship, a hiring pipeline, feed back what actually happened afterward. Tell it how the room reacted, which argument landed, and what you got wrong.
Know the ceiling. AI can stress-test your logic, but it cannot read a room, sense what someone is holding back, or catch when the real issue is the relationship and not the argument.
Call in a human. Some moments need judgment a model cannot supply: a CPA on a technical accounting question, a lawyer on anything with real legal exposure, a coach or an advisor on how the room actually felt. Knowing when you have hit that line is a skill in itself.
We've covered where this shows up, where it breaks down, and how to ask, judge, close the loop, and know when to step back.
In the subscriber section: the actual prompts, the Devil's Advocate skill file, and a checklist for knowing when to make that call, not just that you should.
Closing Thoughts
The rules here are much more fluid than they were in last week's automation edition. There, once you found the right prompt, it kept working. Here, the limits keep shifting as the model changes, so the real skill is testing where those limits sit right now and keeping your own judgment in the loop the whole time.
Thanks for reading this week.
We Want Your Feedback!
This newsletter is for you, and we want to make it as valuable as possible. Please reply to this email with your questions, comments, or topics you'd like to see covered in future issues. Your input shapes our content!
Want to dive deeper into balanced AI adoption for your finance team? Or do you want to hire an AI-powered CFO? Book a consultation!
Did you find this newsletter helpful? Forward it to a colleague who might benefit!
Until next Tuesday, keep balancing!
Anna Tiomina
AI-Powered CFO