Framing: AI Integration Challenges — the first of the two theses Sadra organises his work around; the second is Open Source Philosophy, framed on his site as Open World Development. Core question: what is the part of a task that AI cannot do, and how can AI help the human do that part better? Stage: third year of the CS PhD, as of July 2026. Labs: the Adaptive Computing Experience (ACE) Lab, Souti Chattopadhyay’s lab at GCS, and CUTE LAB NAME, Jonathan May’s lab at ISI.

Summary

The work is about how AI fits into the way people already work, rather than about the models themselves. One question keeps coming back: what is the part of a task that AI cannot do, and how can AI help the human do that part better? Sadra chases it in domains where LLMs arrived quickly and fit awkwardly — places where the technology is already in use and the integration is still unresolved. The stated ambition is broader than any one study: keep adding domains until the set is large enough to support a general account of what goes wrong when AI is dropped into human work.

Domains

Four so far, each one a setting where LLMs arrived fast and sat badly:

Where It Is Going

The plan for the remaining years of the PhD is to keep adding domains until two things can be put down. First, a taxonomy of these integration challenges — the recurring shapes that the awkward fit takes, across settings. Second, a framework for where AI belongs in a workflow and how it should behave once it is there: not only the integration point, but the interaction pattern expected of the system after it has been integrated.

Sadra is open to hearing from people working on adjacent problems.

USC — Adaptive Computing Experience (ACE) Lab — Souti Chattopadhyay’s lab, primary affiliation. USC ISI — CUTE LAB NAME — Jonathan May’s lab at ISI, secondary affiliation. Cognitive Biases in LLM-Assisted Software Development — the developers-with-code-agents domain, on what goes wrong in the interaction itself. Trust Dynamics in AI-Assisted Development: Definitions, Factors, and Implications — the same domain, on what developers mean when they trust a suggestion. Exploring the Challenges and Opportunities of AI-assisted Codebase Generation — the same domain, on building software by prompting rather than writing. ELI-Why: Evaluating the Pedagogical Utility of Language Model Explanations — the same-question-different-background domain, measured as a benchmark. Beyond the Page: Enriching Academic Paper Reading with Social Media Discussions — an integration point studied by building the interface rather than surveying it.