Lisa Orr is a senior product manager at Zapier whose work on support-led software development enables customer-support specialists to fix broken integrations with AI-generated code and engineering oversight. She leads product development for Scout Agent, which turns customer issues into proposed, tested software changes while keeping the people who understand those issues involved in the solution.
From behavioral research to product discovery
Orr earned a PhD from the University of St Andrews and studied bonobo gestural communication at Milwaukee County Zoo in 2011 and 2012. She subsequently joined Urban Airship, now Airship, as a data scientist, building predictive products for mobile marketing. Her work on predictive send-time optimization combined individual engagement histories, local time zones, and broader audience behavior to determine when users were most likely to open an app.
At Airship, she moved into product management and helped establish an internal community for women. Her product discovery work explored how marketers could coordinate customer interactions across channels and time, and how predictive systems could earn users’ trust. Her background also includes work in health technology and government or nonprofit organizations.
At Zapier, Orr introduced AI-powered Custom Actions, which help users add previously unavailable actions to existing integrations without writing the underlying code. Her work on Scout addresses a related organizational challenge: third-party API changes continually break integrations, and support teams encounter those failures before engineers do.
How Scout turns customer context into software fixes
Orr shipped integration fixes herself and shadowed engineers and support specialists to identify the actual bottleneck: gathering external API documentation, internal logs, bug reports, and existing tests. Her approach to AI-assisted integration repair rests on four practical decisions:
Context before code generation: Scout assembles diagnostic evidence before proposing changes, uses language models for synthesis and unit-test generation, and relies on ordinary search when retrieving existing test cases.
Tools embedded in existing workflows: A standalone Autocode playground gained little traction; adoption improved when diagnosis appeared inside Jira workflows and Model Context Protocol integrations brought tools directly into Cursor.
Human-reviewed agent orchestration: Scout classifies incoming issues, assesses whether they are fixable, and runs a GitLab pipeline that plans, executes, and validates changes. Support specialists review the resulting merge requests, request revisions, and submit them for engineering approval.
Evaluation against operational outcomes: Orr tracks classification accuracy, fixability assessments, code quality, and team throughput. During the reported experiment, Scout generated approximately 40 percent of the participating support team’s integration fixes.
For Orr, support specialists contribute fresh customer context and informed validation; engineers remain accountable for reviewing and approving the code.
