Challenges of Agent Design

ne-agents-day.github.io

An oral paper at the NE Agent Day workshop — profiling the efficiency and design challenges of agentic systems, with Prof. Gupta's group at Cornell.

Agentic SystemsEfficiencyProfilingResearch

A research project on the systems-level challenges of designing efficient agentic systems, carried out with Prof. Gupta’s group at Cornell Tech. The work was titled “Towards Understanding Challenges Surrounding Agentic System Design” and accepted as an oral paper at the NE Agent Day workshop.

What we studied

As agentic systems chain together models, tools, and long-horizon control loops, their behavior becomes hard to reason about — and easy to make slow and expensive. This project profiles where that cost actually goes and surfaces the design choices that drive it, working toward a clearer picture of the trade-offs involved in building agentic systems that are both capable and efficient.

Highlights

  • Profiling the efficiency of agentic systems — measuring where latency and compute are spent across an agent’s tool calls, model invocations, and control flow.
  • Characterizing design challenges — distilling recurring pain points in agentic system design into a shared vocabulary for reasoning about them.
  • Oral paper, NE Agent Day workshop — selected for an oral presentation (Jan–May 2026).

Paper and slides coming soon — links to be added once they’re public.