MAIN TRACK TALK

Trust in AI Agents: The Communication & Oracle Problems

Agentic SW development is an active discussion point now, with the promise of improving efficiency ten fold. However, replacing the whole development process with agents is not a trivial task, especially if you are a Quality Assurance advocate always looking for risks. Szilard believes that designing and building the whole SDLC with autonomous agents still has two major challenges to be solved, the communication problem, and the oracle problem.

Solving the communication problem will be one of the key activities to get the full benefit of agentic development. Understanding how to communicate with the agents, forming proper prompts about the tasks at hand and making sure enough context is defined is key. These problems will be even more of an issue when multiple AI agents are connected. Solving the oracle problem in generative AI involves several strategies, like human-in-the-loop to check sub-results before moving forward, or give access to knowledge to the agent through Retrieval Augmented Generation (RAG), Model fine tuning to enrich the model’s local context, or a fully specific LLM training.

What you’ll learn

Session details

Szilárd Széll

Szilárd Széll is a DevOps and AI Transformation Lead, Test Coach, and SAFe 6.0 SPC at Eficode. He has 25 years of experience with DevOps transformation, especially in the telco industry, and lately focusing on AI. He has also worked as an assessor, trainer, facilitator, and coach, and awarded Tester of the Year 2024 in Finland. Szilard is very involved in the testing community, runs a Meetup, active in ISTQB as a Product Owner, and is a member of the Hungarian Software Testing Board (HTB). For many years, Szilard has been working on and supporting conferences like HUSTEF, UCAAT, EuroSTAR.