How we accidentally discovered our AI agent’s true purpose

While we were building Jotform’s AI Agents, the goal was obvious: Help people fill out forms faster.

But when we launched our Beta, something unexpected happened. People weren’t just using our AI agents to fill out forms — they were using them to run their entire businesses. And we had to pivot to keep up.

Here’s what we learned about AI agents in the process:

The wake-up call: Watching users “hack” the AI

A few weeks into Beta, we started noticing some unusual activity.

• One user had integrated an AI agent into their website — not to guide customers through a form, but to answer support questions.

• Another had built an AI assistant for qualifying leads and scheduling sales calls.

• And another was using it for HR onboarding and PTO requests.

My first reaction? “Wait… that’s not what we built this for.”

But the more we observed, the more it made sense. Instead of staying focused on forms — our bread and butter — we needed to focus on pain points.

People didn’t need an AI agent that simply helped fill out forms. They needed an agent that could replace manual, repetitive conversations altogether.

And just like that, our product vision changed.

What I would have done differently

1. Launch earlier — even if it’s not ready

I’ve always lived by the “Launch early and often” motto, but I wish we had released this even sooner.

Every extra week spent refining the wrong assumptions was a week’s worth of feedback that we missed.

2. Design AI agents that are adaptable, not prescriptive

We built AI agents with one purpose in mind. Users immediately found five others.

If I had to do it again, I’d:

• Build AI modularly so users could adapt it to different needs.

• Expect unexpected use cases and make pivoting easy.

• Let users customize AI behavior instead of building for a single workflow.

3. Build AI agents to take action

It became clear early on that, contrary to our assumptions, users didn’t just want AI that suggested answers — they wanted AI that’s a workforce.

Looking back, I would have started with:

• Automating entire workflows, not just responses.

• Built-in customization for what AI should handle vs. what needs human input.

4. Stop asking users what they want — watch what they do instead

Our beta testers gave us plenty of feature requests. But our biggest insight didn’t come from their asks. It came from how they ignored what we built and repurposed it for something else.

If I had a do-over, I’d pay a lot more attention to behavior than feedback.

5. Design for gradual AI adoption

No one goes all-in on AI overnight. They start with small wins — automating FAQs, routing leads, etc. — then expand.

I’d build the product to support that natural progression from the start.

6. Stop thinking of AI agents as a standalone tools

We assumed Jotform AI Agents would be a single feature inside Jotform.

Users saw it differently.

They wanted AI to connect across their entire tech stack — CRM, email, Slack, payment systems. They wanted it to become part of their existing workflows; not another tool to manage.

If I were starting over, I’d do what we’re doing now:

• Prioritize deep integrations with the tools businesses already use.

• Make AI embeddable across different platforms — not just in forms or chat.

• Design AI to work within systems, not just alongside them.

The best AI is invisible

A big lesson we took away is that nobody was excited about AI itself. What they loved was what it let them stop doing.

So, always remember that the best AI products don’t call attention to themselves. They just work, making tedious tasks disappear.