GIDS news media, articles, insights and virtual events educate and illuminate its audiences so they can be fully prepared to deal with the new realities at work and in their professions.
Get the latest event updates, and insights from today's leading voices.
< session />
Wed, April 22ArchitectureDeepTech TechLead
As AI agents become central to software workflows, a critical issue often goes unnoticed: instruction compliance. Do agents actually follow the rules, constraints, and behavioural guidelines defined in system prompts and agent workflows? This session argues that the industry has largely framed this challenge as a context engineering problem, when in reality it is a problem of measurement and compliance.
Drawing on research published between 2025 and 2026, the talk presents a grounded analysis of how instruction drift occurs and why current practices fail to detect it. It introduces a practical framework that helps engineering teams evaluate whether agents are following defined constraints and how to improve reliability in production systems.
What You Will Learn
Why instruction compliance is a core reliability challenge for AI agents
How instruction drift occurs even when prompts and workflows appear well defined
A practical framework for measuring and improving compliance in agent systems
Who Should Attend
AI and machine learning engineers
Software developers building agent-based systems
Platform and infrastructure engineers
Software architects
Engineering leaders responsible for AI reliability and governance
< speaker_info />
Karrtik Iyer is an AI researcher and Head of the Data Science & AI Community at Thoughtworks India.
He brings over 25 years of industry experience, with nearly a decade at Thoughtworks. His expertise encompasses Large Language Models (LLMs), Generative AI, Natural Language Processing (NLP), Knowledge Graphs, and Bayesian Learning.
As part of the Thoughtworks AI Labs, he is actively engaged in research on the interpretability of LLMs, addressing challenges such as the AI trust paradox and model transparency.
Jem is a product strategist for generative AI engagements in India, leading the Thoughtworks India Applied Gen AI R&D team. He partners with clients in the early stages of problem-solving and opportunity identification, shaping enterprise AI strategies & roadmap, and integrating design and product thinking into AI-based applications. His experience spans various domains, including banking and financial services, retail, pharma, and healthcare.
With nearly 13 years of industry experience, Jem has led projects building payment solutions and OTT platforms. Before joining Thoughtworks, Jem specialized in asset management and asset servicing technology.
