Automate the Task. Own the Decision.

Someone on a GTM team walks into the boardroom and tells the head of another function they’re doing their job wrong. Not because of the pipeline. Not because of any evidence. Because ChatGPT said so.

That one I’ve watched happen. Decades of experience around that table, budgets to match, and the thing that ended the discussion came out of a chat window.

I got into it with Mark Stouse of Proof CausalAI and Paula Skaper of 33Dolphins Growth Strategy, back for a second round after we dug into what actually goes wrong when AI enters go-to-market

The discussion started to draw a clean line: what you can hand to a tool, what you can’t. We never finished drawing it. What we ran into instead was the thing that shows up in your boardroom whether or not that line ever gets drawn.

Takeaways

  • Whoever made the call still owns the call. No job description changed when AI arrived.
  • Teams saying “the AI told me” are being careful, in the way people get careful when being wrong costs more than being useless.
  • Leaders push work at AI to avoid deciding. Staff quote AI back to avoid blame.
  • Automating a task inside a broken model gets you the wrong answer faster.
  • Delegation has never been abdication.

Fear with a workflow around it

You might think people are being lazy. It’s actually worse.

The upside of being right and being seen to be right is perceived as being not nearly as valuable as distancing yourself entirely and making it AI’s fault.
 
Mark Stouse

Run that as arithmetic. Upside for being right, downside for being wrong, and in most companies the downside is bigger by a wide margin. So the rational play is to make sure the answer has someone else’s name on it. Hence, a tool with no career and no mortgage (and no performance review coming in March) makes an excellent “someone else”.

You can’t fix that with a prompt library.

Managers want the cover too

It’s also happening from the other end of the same pipe. Managers sending work back with “did you run it by AI? Don’t bring it to me until you have.” So now you have the manager side coupled with the staff side, work arriving with “this is what Claude told me” attached to it. Both sides are “covered”, nobody deciding anything, and nobody checking the output either.

The other thing I’m seeing is people who don’t like what they hear from a subject matter expert, going to their AI, getting their AI to agree with them and what they wanted to hear and that the subject matter expert is wrong.
 
Paula Skaper

That’s a real-world example Paula recently witnessed. A small business owner didn’t care for the sales playbook they’d been handed by a fractional VP of sales with thirty-five years of high-ticket industrial B2B behind him. So they took it to ChatGPT, which agreed the playbook was built for a retail business. That was the end of it. Thirty-five years, overruled by the first thing that was framed to agree with them. 

Which of those two would your own team rather hear from this quarter? And which one have you been rewarding?

You can’t own an answer you can’t check

GTM has quietly settled the effectiveness question in its own favour, so the only thing left to argue about is doing the same work faster and cheaper. CAC says otherwise. So do plenty of the other GTM numbers. Automate a task inside that assumption and you don’t get efficient, you get wrong at speed and scale.

Which raises the obvious problem. Everybody says “Critical thinking!” Almost nobody can define it.

Critical thinking is all about the extent to which you factor in capital R ‘Reality’. That’s it. How well does this, whatever that is, whatever is being asserted, stack up against Reality?
 
Mark Stouse

That test has to run before ownership means anything. If nobody on the team can hold an AI answer up against what is actually happening in the market, then whoever signs their name to it is signing for something they never checked.

The exposure lands on a person

So I put it to both of them: who owns the call when the answer turns out to be wrong? Mark answered in two words. The decider.

When I had to write CQ on the bottom of my article for Newsweek, saying I checked it myself, and I was effectively guaranteeing that it was 100% correct, I owned it, not my researcher. Not the AI. Me!
 
Mark Stouse

Mark built that habit decades before any of this. These days he runs what he writes past five different AI tools and lets them tear each other up. His name still goes on it at the end, and that’s the whole point of the exercise.

None of this is only a values argument. The Delaware rulings on officer oversight makes a broader accountability point: putting AI in the decision chain doesn’t remove the human responsibility around it. The law didn’t change because AI showed up. What changed is how many decisions now have a machine somewhere in the middle.

I made the cost version of this argument back when it was about headcount. Replacing people with AI doesn’t spread accountability out, it concentrates it. Automating a task works the same way, and the tool absorbs none of the exposure.

The captain was asleep when the ship ran aground

I’ve been saying this since long before there was a bot to delegate to. When you delegate anything, to a person or to a machine, you haven’t handed over the ownership. You’ve handed over the work.

The captain of the ship has to delegate everything. He could be asleep, or she can be asleep in their cabin, and if the ship runs aground? Doesn’t matter. That’s a career-ending situation.
 
Mark Stouse

It’s an extreme case, sure, but extreme for a reason. The arrangement only works because someone is answerable regardless of whose hands were on the wheel. Take that away and you don’t get a flatter organization, you get a ship nobody is steering.

The line we didn’t draw

The conversation stopped short of one thing: What has a checkable answer and can therefore be automated? What doesn’t and can’t? I asked Paula for it straight out. She took the question somewhere else, somewhere better as it turned out, and then we ran out of clock.

So I’m not going to pretend that we landed it. We didn’t. That’s the next one.

What we did land is the part that doesn’t depend on the test. Whatever the tool told you, yours is the name on the org chart.

Think of it this way: The car you drive was built by robots. But it was inspected, verified, and checked by human beings before final delivery. 

Two things for next week

  • Pull the last five decisions your team made with AI somewhere in the loop. Write a name beside each one, an actual person, not a function. Any blank line is the gap.
  • Change what the question sounds like. “Did you run it by AI?” invites abdication. “What do you think? What did you check? How did you frame it?” doesn’t. Same meeting, different outcome.

Missed the session? Watch it here.

Paula Skaper is the founder of 33Dolphins Growth Strategy and author of Rethink, Realign, Reinvent.


If you like this content, here are some more ways I can help:

  • Follow me on LinkedIn. I post weekly on who actually owns GTM calls, and what it costs when nobody does. Connect with me.
  • Work with me. If AI is going into a go-to-market model nobody outside Marketing can independently check, that’s what the GTM Reality Check is for. Start a GTM Reality Check.
  • Subscribe. Paula’s back for the next episode, and we pick up the automate/don’t-automate line there. (Enter your email below.)

Cheers!

This article is AC-A (“Authenticity Commons, Assisted”): AI helped produce it, with me in the lead. Here’s what that means.

Achim’s Razor is also published as a LinkedIn newsletter.