Postman

Agents are making a big impact at Postman. However, I wasn’t surprised about the recent MIT study mentioning that 95% of generative AI pilots are failing to deliver real value. Building real-world value requires data, business context, and solid engineering to deliver results. Without context, AI tools generate a lot of garbage. Specifically for product work, I started using ChatGPT to analyze the web or do deep research about how we can improve the product. While the answers were confident, the hallucinations were ridiculous - almost to a point, that using the output would be actively harmful. ChatGPT was confident in telling me about awesome Postman features that didn’t exist, or imagined capabilities that are not needed. Other AI tools weren’t better either. They mostly cap out at ~50 sources and mostly present incomplete information. Building new capabilities for us means understanding related ideas in requests, workarounds, suggestions across multiple threads. We’ll never get done if we solve issues individually. Correspondingly, understanding thousands of issues and comments that have been filed along with keeping up with new demands takes a lot of time for product managers, designers, and engineers. We’d be mostly reading - but not doing! Frustrated, a few weeks back, I pinged Numaan Ashraf, our head of forward deployed engineering to “hire an agent” using our Agent Builder platform. This agent - called Aira - should analyze feature requests, issues, and comments that our community reports on our official tracker and help us make sense of it. The team built a quick prototype to analyze just one issue. The moment I tried it, it blew my mind. Starting with the ability of analyzing one issue, the agent can help stitch together issues and comments across the “entire” graph of reported issues. The agent can identify core problems, create workarounds, and even suggest ways to solve the problem right now. The agent is integrated with Slack so anyone can get started with @Aira and type in their query (See the attached demo). Now, every team at Postman has immediate access to this agent and can use it for their work. The agent has citations built in and it can link to the source material. You can check out the agent in action below. We have also published a how-to guide on our blog post and the stack we used [Link in the comment!]. This has been just the starting point. We are now incorporating ALL areas of feedback into this agent while also “hiring” more agents that I’ll publish next.