AI Recommended It, Your Team Built It, the Client Rejected It — Here's the Fix

The essence of good interaction design is to devise interactions that let users achieve their practical goals without violating their personal goals. — Alan Cooper, The Inmates Are Running the Asylum

I keep seeing the same problem with automation requests: owners asking to implement the wrong thing.

You might say: “I’m a smart owner. I don’t do that.”

Maybe. But have you ever delegated a task and gotten back something you didn’t actually need? Or worse, your team passed it to a client and got a complaint saying it’s all wrong.

This happens more often now that AI is in the mix:

1. A business owner has a high-level goal but doesn’t know how to reach it.
2. They ask AI for a solution. It gives them something plausible-sounding.
3. They don’t know how to implement it, so they hire a contractor to “build this.”
4. The contractor is confused — the solution seems off, but they build it anyway.
5. After a lot of back and forth, it becomes clear the original solution was a dead end.

The contractor was asked to execute the wrong thing, with no visibility into the actual problem.

A real example

This is a request I assessed recently:

We run a content agency. Editors post finished videos to ClickUp cards as attachments. Once approved, we want to push the video and caption to a scheduling app automatically. All we’d need to do is review and post. Right now we download each video manually, upload it to GHL, write the caption, and post. That’s 10–15 videos a week. We know tools like Metricool and Publer exist, but are pricey. Is there an easier way, or do we just have to pay? Currently using Zapier as our middleman.

When I passed this to AI, it recommended this workflow: ClickUp → Zapier → Google Drive → Scheduler. It said ClickUp can’t natively push video attachments to social schedulers because ClickUp attachment URLs are private and authenticated, so scheduling apps can’t pull from them directly. The proposed workaround:

• Editors upload finished videos to a shared Google Drive folder AND attach to the card
• When the card status changes to Approved, Zapier grabs the Drive file URL and sends it + the card description to your scheduler

If you’d asked someone to implement that workflow, you could have run into:

• Setting up Google Drive access for the whole team
• Building a folder structure inside Drive
• If they don’t already use Drive, paying extra for more than one user
• More steps mean more Zapier credits, more complexity, and more things that can break

There are also issues that come with AI:

• It jumps to recommend solutions before understanding the actual problem
• It hallucinates. If you don’t have the expertise to spot it, you won’t know when it’s wrong
• It can miss relevant data, or reference outdated data

In this case, the “private URL” problem wasn’t real. Zapier, when authenticated, can download the file and re-upload it without an intermediary. Adding Google Drive was unnecessary.

My approach was much simpler. I set up a two-step workflow with a ClickUp trigger (status changed on a specific list) and a step to publish the video to Instagram using the attachment in the task. Works on a free Zapier plan and uses the stack they already had.

What’s your goal?

Goals are not the same things as tasks. A goal is an end condition, whereas a task is an intermediate process needed to achieve the goal. It is very important not to confuse tasks with goals, but it is easy to mix them up.

If my goal is to laze in the hammock reading the Sunday paper, I first have to mow the lawn. My task is mowing; my goal is resting. If I could recruit someone else to mow the lawn, I could achieve my goal without having to do the mowing. — Alan Cooper, The Inmates Are Running the Asylum

In software engineering, this is known as the XY Problem: the user asks about their attempted solution (Y) instead of the actual problem (X). It wastes time, obscures the real issue, and creates new problems.

The fix is simple: describe your goal.

You’ll get the right things done if you explain what success looks like, what your pain is, how your needs would be solved, and the impact on your business.

That’s why, when a client comes to me for advice, I start by asking questions to understand the context and the expected outcome. For example: this would mean we no longer spend three hours a week downloading and uploading approved videos.

Once I have a clear picture, I translate it into specific requirements (e.g. must support up to 20 videos of 1GB each per week), look at the alternatives, and test them if needed.

If you don’t have someone to help with this, ask AI to interview you — have it ask questions to map out the situation before you delegate the task.

AI is useful for ideation and brainstorming. But if you’re not the expert, you won’t know when it’s recommending you wrong.

So before passing that task to someone, ask yourself: What are we actually trying to achieve?