We Interviewed Our Own Team Using AI. Here’s What we Learned.

Mahesh Kulkarni | 15 December 2025

At Tekdi, one of our core beliefs is "Enjoy the Magic of Creation." But as we pushed to become a fully AI-enabled organization, we hit a very practical wall: How do we actually measure our team's AI maturity?

To build an AI-native future, we needed a baseline. We needed to know how our developers, managers, and designers were currently using tools, what blocked them, and where they saw potential.

Usually, you have two bad options for this:

  1. A Google Form: Which creates shallow data because people hate typing out long answers.

  2. 1:1 Interviews: Which yield great data but take weeks of scheduling time.

We wanted the depth of the interview with the speed of the form. So, we decided to use the very technology we were trying to adopt. We let an AI run the research.

The Logistics Nightmare

Our target audience was 40+ colleagues ranging from junior devs to our CXOs. To do this manually, we would need 45 minutes per session, plus the buffer time for rescheduling and calendar coordination. That is weeks of dedicated management bandwidth we simply didn't have.

We needed a way to conduct deep discovery without being bound by human calendars.

Solution: A Conversation, Not a Chatbot

We built a conversational voice agent to do the job.

For this specific experiment, we utilized ElevenLabs to create a voice interface that felt natural and fluid. (Note: The tech landscape is moving fast—today, we could achieve this same conversational depth and latency using Gemini Live, which is excellent for this kind of dynamic interaction).

We avoided text-based chat. We wanted people to talk because people speak faster than they type, and they share much more context when they aren't staring at a blinking cursor.

How we set it up:

  • The Context (RAG): We didn't use a generic bot. We fed the agent specific context about internal roles. It knew the difference between a Project Manager's deadlines and a Tech Lead's code quality concerns.

  • The Voice: We chose a profile that felt professional but approachable to reduce the weirdness of talking to a machine.

  • The Analysis: We treated the conversation transcripts as the raw data, feeding them into a secondary workflow to extract sentiment and key themes.

Pleasant surprise when one of our CXO used it 

We expected the agent to handle basic questions fine. We were worried about how it would handle senior leadership.

To our surprise, the agent held its ground. It didn't just read a script; it asked follow-up questions based on the answers provided. When a CXO gave a high-level answer, the AI dug deeper into the strategic nuances. It was a serious conversation, though the novelty of "talking to the code" definitely made the process more engaging.

The Reality Check (of course, it wasn't perfect) 

  • The Win: The quality of information was far superior to a written survey. We got stories, not just checkboxes.

  • The Friction: Because the agent was incredibly thorough, a few users actually lost patience with the detailed follow-up questions. It was a good lesson for us: even with AI, brevity is key.

How did it help us 

The efficiency gains were certainly there.

  • Speed: We collected high-quality interview data from our initial batch of key stakeholders in just 3 to 4 days.

  • Analysis: Manually, this depth of collection and synthesis would have taken a human team over two weeks.

  • Outcome: We mapped out our enablement strategy immediately based on real feedback, not assumptions.

Interesting use case for the "Bazaar"

We realized this wasn't just an internal HR hack. It solved a universal industry problem: Knowledge Transfer.

We recently took this architecture and demoed a Proof of Concept (POC) for a client in the manufacturing industry. They are facing a "brain drain" risk as long-serving experts approach retirement. They need to capture decades of tribal knowledge before it walks out the door.

Using the same logic we tested on ourselves, we showed them how an empathic AI agent could interview retiring staff, capturing their technical know-how in a way a manual form never could.

Conclusion

AI isn't just about generating code or images; it's about scaling human connection. Whether you use tools like ElevenLabs or Gemini Live, the goal is the same: to remove the bottleneck of logistics so you can focus on the value of the conversation.

We tested it on ourselves, it worked, and now we’re ready to build it for others.

View Similar