Nikunj Handa works on product for the OpenAI API. His public work explains tools, models and interfaces for developers building agents, including ways to make long-running interactions more responsive.[1][3]
Start with Why Dwarkesh is Wrong about Computer Use + How OpenAI shipped its Jev competitor in 1 Week →
X · nikunjhanda Career
Contributions
Keep a tool call from stopping the interaction
His API updates explain asynchronous tool calls and mid-turn steering. A model can continue while a tool runs, and new information can steer the work already underway. The product question is how an agent exchanges information with its surrounding system while a task is still in progress.[1]
Manage the context of a long-running agent
In the DevDay conversation, Handa discusses caching and compaction as separate concerns: reuse context efficiently, while retaining a useful representation of a long interaction. These API choices affect what an application must manage in its own harness.[2]
Tools, writing & talks
Writing, research & talks
Start here
Why Dwarkesh is Wrong about Computer Use + How OpenAI shipped its Jev competitor in 1 Weeklatent.spaceA discussion of agent APIs and developer infrastructure.
Start at 19:14 for API interaction, then 32:23 for caching and compaction. Familiarity with APIs helps.
Interview participant, API segment[2]Original source