Like many of us, I’ve been curious about whether AI can actually take on real work. Not just write snippets of text or answer questions, but actually do things inside SaaS tools.

So after a few nights of very bad sleep recently, I decided to put that curiosity to the test and see if AI could take over dashboard-building duties for me. If I wasn’t firing on all cylinders, maybe ChatGPT’s recently-launched Agent Mode could pick up the slack and build a dashboard for me using Geckoboard. Or so I thought.  

The experiment: Could ChatGPT take my half-baked prompt and build out a working dashboard in Geckoboard?

The benchmark dashboard

Before setting ChatGPT loose, I needed a clear target. The benchmark I chose was a simple Zendesk Support dashboard I’d built recently in this video, which took me all of about eight minutes. It didn’t contain anything particularly fancy - just the kinds of metrics most support teams want to see at a glance:

  • New tickets today
  • Open tickets by priority
  • Tickets nearing SLA breach
  • One-touch ticket rate
  • CSAT (customer satisfaction)
  • Average wait time
  • Top ticket solvers

That’s the dashboard I challenged ChatGPT to recreate.

The experiment setup

OpenAI recently launched Agent Mode, essentially giving ChatGPT the ability to take actions on your behalf through a baked-in browser. With the right prompting, there’s a lot it can do. 

For this experiment, I put very little effort into crafting a comprehensive prompt, but there was some basic guidance I provided before starting the build: 

  • A screenshot of my benchmark dashboard from the video
  • Instructions to call the new one “Zendesk dashboard built by ChatGPT.” (I have a lot of dashboards in my test account, so this was important)
  • A reminder to use only the Zendesk Support integration.
  • A nudge to pay attention to the layout of the dashboard, filters, and time periods.

Then I sat back and watched it work.

What happened when AI took over

At first, it was magical. After thinking for a short while, ChatGPT opened up a mini browser, navigated to the Geckoboard website, and found the login page, all within a few seconds.

It then asked me to take control and enter my login details, before taking back control. Within another few minutes it had found the Add widget button, and started dropping in metrics. It even adjusted filters and descriptions as it went, all the while narrating its thought process.

Things were looking promising, but soon the cracks appeared:

  • It looked for metrics that didn’t exist (like a very specific “Normal 77” open ticket count it had seen in my screenshot).
  • It got stuck on presets, missing the more flexible “Start from scratch” option.
  • It struggled with widget sizing and layout — the final dashboard looked similar but definitely not identical.

That said, it still managed to:

✅ Add open tickets and tickets nearing SLA breach.

✅ Set up top ticket solvers.

✅ Add a CSAT widget (though it showed “no data” because of test data limitations).

Where it stumbled:

❌ It couldn’t locate the one-touch ticket rate (swapping in SLA achievement instead).

❌ “New tickets today” wasn’t configured properly.

❌ Layout didn’t quite match the example.

You can watch how things unfold in the video below.

The verdict: a C+ effort

In the end, ChatGPT delivered about half the dashboard correctly. Not terrible for 26 minutes of automated work, but slower and less accurate than if I’d built it myself.

On balance, I’d give it a C+. It showed promise, but also got tripped up by UI quirks and a lack of specificity in my prompt.

What this means for dashboard builders

There’s plenty of hype around AI replacing humans. This little experiment shows a more nuanced reality:

  • Today’s AI could be a decent dashboard-building assistant — it can get you partway, suggest alternatives, or do repetitive clicks.
  • But humans still need to guide it — knowing which metrics matter, how to arrange a dashboard for clarity, and when to troubleshoot “no data” issues.
  • Prompts are everything — if I’d been more prescriptive (exact metric names, exact layout), the outcome might have been stronger. I’ll certainly be experimenting with different ways to prompt over the coming weeks to see if I can get this to work reliably in one shot.

For now, the fastest, most reliable way to build a Zendesk dashboard is still… a human using Geckoboard au naturel.

👉 Try Geckoboard’s Zendesk integration yourself - most new users can spin up a working dashboard in 15 minutes or less — no AI required.

Final thoughts

Was this experiment a failure? Not at all. Watching ChatGPT navigate Geckoboard was fascinating. It almost felt like pairing with a new joiner - keen, quick to try things, but often in need of correction.

And that’s the real takeaway: AI might not fully replace dashboard-building any time soon, but it can probably help you get bit of a head start. 

For now, though, if you’re underslept and need a Zendesk dashboard fast… Geckoboard is still your best bet.