Inspiration

AI agents are becoming increasingly powerful, but an important question remains: Do they actually follow the instructions given to them? We were inspired to explore this problem and understand how reliably AI agents follow user instructions, especially when they face confusing, conflicting, or misleading prompts.

What it does

DriftLens AI is a project designed to evaluate instruction-following behaviour in AI agents. It focuses on identifying situations where an AI agent misunderstands instructions, ignores important rules, or produces responses that do not match the given requirements.

How we approached it

Our approach is to test AI agents using different instructions and scenarios, compare their responses with the expected behaviour, and identify possible deviations. The goal is to make AI systems more reliable, consistent, and trustworthy.

Challenges

One of our main challenges was understanding how to evaluate whether an AI agent had followed an instruction correctly. Defining test scenarios and identifying differences between expected and actual responses also required careful thinking.

What we learned

Through this project, we explored AI agent behaviour, instruction-following, testing methodologies, and the importance of reliability in artificial intelligence.

Future Scope

We aim to improve DriftLens AI with more testing scenarios, better evaluation methods, and detailed reports that help developers understand and reduce instruction-following failures.

Accomplishments that we're proud of

We are proud of choosing a real problem in today's AI world and developing an idea to address it. We wanted to understand how AI agents respond to instructions and where they can go wrong. This project gave us an opportunity to think practically, explore new AI concepts, and work together as a team.

What we learned

While working on DriftLens AI, we learned that giving instructions to an AI agent is not enough. It is equally important to check whether the agent understands and follows them correctly. We also gained a better understanding of AI agents, testing methods, problem-solving, and teamwork.

What's next for DriftLens AI

In the future, we want to make DriftLens AI more advanced by adding different testing scenarios and improving its evaluation process. We also plan to develop a simple dashboard where users can easily view test results, identify mistakes, and understand how reliably an AI agent follows instructions.