▶ Watch ↗AI Engineer World's Fair 202631:24
Conference affiliation: Member of Technical Staff · Anthropic · 2026
Isabella He is a member of technical staff on Anthropic’s Applied AI team, working on Claude Managed Agents. Her engineering and technical writing address two connected problems: what an agent should preserve during long-running work, and what its surrounding software should change when a newer model makes an old workaround counterproductive.
Isabella began building organizations while still in high school. Volunteering with Friends of Children with Special Needs introduced her to autistic children and their families; an internship at an autism intervention program in Chengdu exposed the difficulty of sustaining support after a short treatment program ended. She co-founded AUesome with Anshul Gupta, Andrew Kim, and Arnav Gurudatt to develop at-home therapy kits, and founded SN Inclusion to provide career-technical education for neurodivergent people.
Her account of those ventures explains how the AUesome team revised its initial ideas. They sent kits to parents and therapists, incorporated criticism, and worked with parent ambassadors to reach more families. An unfavorable response from a professor prompted further product revisions. SN Inclusion addressed vocational preparation through an e-learning platform and a network of chapters. In 2021, Isabella was its president, AUesome’s co-president, and an intern with the Stanford Neurodiversity Project.
Her Stanford work extended into software. From December 2020 to October 2022, she worked as a research assistant in the university’s psychiatry and behavioral sciences department. She worked on an AI-driven Neurodiversity Job Matcher for the Neurodiversity at Work clinical trial. The engine used occupational competency information and entity resolution to compare candidates’ experience with job requirements. A résumé and a job description can express related abilities in different terms; the matching process needed to account for those relationships rather than rely entirely on identical keywords.
Alongside her Stanford studies, Isabella explored engineering, product management, and finance. At SGNL, she progressed from a software engineering internship in 2022 through product and engineering work into technical product management, which she held from December 2023 to March 2025. Her work there included a synthetic dataset generator, Okta and AWS integrations, and later integrations with GitHub and CrowdStrike. Her career history also records achieving Continuous Access Evaluation Protocol interoperability.
During a product-management internship at Intuit, she worked on the AI and automation team. Her other internships included software engineering at Mooch, where she worked on preventing brute-force attacks; analysis at Bank of America Private Bank; and quantitative portfolio strategy at The D. E. Shaw Group, where she analyzed competitor factor attribution using Fama–French models. These roles overlapped with her education rather than forming a sequence of full-time jobs.
Isabella joined Anthropic in December 2025. Her announcement of the move described choosing her first full-time role after exploring several industries. Reading Dario Amodei’s Machines of Loving Grace had inspired her interest in Anthropic’s mission. She was also attracted to Applied AI’s combination of product, research, and customer work, particularly Claude Code’s potential to change how teams build software.
Isabella’s context-engineering writing distinguishes problems that can otherwise look like the same problem of running out of space. Her March 2026 cookbook uses a research agent that reads documents, takes notes, and continues across sessions. Tool-result clearing removes older, retrievable outputs; compaction summarizes accumulated conversation history; persistent memory stores selected information for use across sessions. Each makes a different choice about what to discard or preserve. Clearing can invalidate cached prompt prefixes, while compaction can lose details that later turn out to matter.
With Gagan Bhat, Isabella co-authored an article on building with Claude Managed Agents, connecting the evolution of agent interfaces with the infrastructure needed to move from prototypes to production. Their division of responsibility leaves application teams supplying domain knowledge, context, and appropriate tools while Anthropic maintains the underlying harness and execution infrastructure.
In her sections of their joint talk on agent infrastructure, Isabella develops three concrete engineering concerns:
▶ Watch ↗AI Engineer World's Fair 202631:24
Loading more talks and highlights…