Szilárd Széll
DevOps and AI Adaptation Lead at Eficode, Finland
Talk
Old Problems of New Agents: Oracles and Communication
Abstract
Agentic software development promises tenfold efficiency gains. But replacing the development process with autonomous agents is far from trivial — especially from a quality assurance perspective.
Two fundamental challenges remain: the communication problem and the oracle problem. In current development, effective communication between customers, developers, testers, and operations is key to delivering value. That challenge does not disappear when AI agents enter the conversation — it intensifies. Understanding how to communicate with agents, craft effective prompts, and provide sufficient context becomes a critical skill.
The oracle problem — determining what the correct output should be when there is no clear ground truth — is equally pressing. Strategies such as human-in-the-loop validation, retrieval-augmented generation, and model fine-tuning can help, but each comes with trade-offs.
This presentation examines how communication and oracle challenges affect software development today, and how to design agentic processes that deliver high-quality outcomes.
Two fundamental challenges remain: the communication problem and the oracle problem. In current development, effective communication between customers, developers, testers, and operations is key to delivering value. That challenge does not disappear when AI agents enter the conversation — it intensifies. Understanding how to communicate with agents, craft effective prompts, and provide sufficient context becomes a critical skill.
The oracle problem — determining what the correct output should be when there is no clear ground truth — is equally pressing. Strategies such as human-in-the-loop validation, retrieval-augmented generation, and model fine-tuning can help, but each comes with trade-offs.
This presentation examines how communication and oracle challenges affect software development today, and how to design agentic processes that deliver high-quality outcomes.
About
Szilárd Széll is a DevOps and AI adaptation lead passionate about Quality with over 25 years of experience. He has played a key role in building testing communities and has been involved in the program committee of major events. Recently, he co-founded the Finnish Testing Meetup, which has grown into an active community of over 500 members. Szilard also contributes to the advancement of testing by sharing his expertise on stage and creating content as an active member of ISTQB.
← Back