Zapier transforms its support function into a code-shipping powerhouse, leveraging AI to automate bug fixes and redefine developer workflows.
The "App Erosion" Crisis at Zapier
- Zapier, with over 8,000 integrations built on third-party APIs, faces constant "app erosion"—API changes and deprecations that create reliability issues. This continuous degradation leads to a backlog crisis, impacting customer experience and increasing churn.
- Zapier's 14-year history means some integrations are equally old, requiring constant maintenance.
- API changes and deprecations generate a relentless stream of bugs.
- The support team's ticket volume outpaced their capacity, necessitating a new approach.
- Lisa Orr states, "Tickets were coming in faster than we could handle them, creating integration reliability issues, poor customer experience, even churn."
Experiment 1: Empowering Support to Ship Code
- Zapier initiated a two-year experiment to shift its support team from mere triage to active bug fixing. This move addressed a critical need and capitalized on support members' aspirations for engineering roles.
- App erosion represented a major source of bugs flowing from support to engineering.
- Many support team members unofficially maintained apps and sought engineering experience.
- Initial guardrails included focusing on four target apps and requiring engineering review for all support-generated merge requests.
- Orr notes, "Unofficially many support members were already helping to maintain our apps."
Experiment 2: AI-Driven Context and Codegen (Project Scout)
- Parallel to empowering support, Zapier launched "Project Scout" to investigate how AI-powered code generation (codegen) could accelerate app erosion solutions. The initial focus was on automating context gathering.
- Discovery revealed significant time spent by engineers and support gathering context from API docs, internal logs, and external sources.
- Zapier built APIs, some utilizing Large Language Models (LLMs) for diagnosis and context curation, others for tasks like unit test generation.
- Early challenges included low adoption of a web-based "playground" tool because it required engineers to leave their primary workflow.
- Orr explains, "One big discovery we had is how much time is spent gathering the context going to the third-party API docs, even crawling the internet looking for information about a bug."
Embedding Tools and Orchestrating AI Agents
- Zapier recognized that tool embedding was crucial for adoption. The company integrated its AI tools directly into developer workflows, leading to the creation of "Scout Agent" for end-to-end bug resolution.
- The "diagnosis" API, which curated context for support, became a "support darling" after being embedded directly into their Jira ticket creation process via a Zapier integration.
- The launch of Zapier's internal "MCP" (likely a platform for embedding tools) and external tools like Cursor (an AI-powered IDE) allowed deeper integration of Scout's APIs into engineers' workflows.
- Challenges persisted with tool runtime (diagnosis was slow) and scattered adoption across a suite of tools.
- Zapier shifted focus to "Scout Agent," an orchestration layer that combines diagnosis with codegen to produce merge requests automatically.
- Orr emphasizes, "Embedding tools is the key to usage as we find out."
Scout Agent's Impact and Support's Superpowers
- Scout Agent now directly processes support issues, categorizes them, assesses fixability, and generates merge requests. This system significantly boosts support velocity and allows engineering to focus on complex problems.
- Support submits an issue to Scout Agent, which categorizes it, assesses fixability, and generates a merge request.
- Support reviews and tests the generated merge request, requesting adjustments via chat in GitLab, which triggers another pipeline run.
- The process is orchestrated by Zapier's internal "Zaps" and utilizes GitLab CI/CD pipelines, integrating Scout MCP tools and Cursor SDK.
- Scout Agent currently generates 40% of support team app fixes, doubling individual support member velocity from 1-2 to 3-4 tickets per week.
- Support teams possess three "superpowers": proximity to customer pain (fresh context), real-time troubleshooting, and superior validation capabilities.
- Orr asserts, "40% of support team's app fixes are being generated by Scout. So we're doing more of the work on behalf of the support team."
Investor & Researcher Alpha
- AI-Driven Developer Tooling: The success of Scout Agent highlights the immediate value of AI orchestration layers that integrate existing tools (LLMs, IDEs, CI/CD) into seamless workflows. Capital will flow to platforms that reduce context switching and automate multi-step development processes.
- Shifting Talent Dynamics: The program demonstrates a viable path for upskilling non-traditional roles (e.g., support) into code contributors via AI assistance. This suggests a future where AI democratizes coding, expanding the talent pool for technical roles.
- Internal AI Agent Adoption: Zapier's journey from individual APIs to an orchestrated agent provides a blueprint for enterprises building internal AI solutions. Focus on embedding and end-to-end workflow automation drives adoption and measurable impact.
Strategic Conclusion
Zapier's Scout Agent proves that empowering support teams with AI-driven code generation dramatically improves operational efficiency and product reliability. The next step for the industry involves widespread adoption of AI agents to automate routine development tasks, enabling specialized teams to contribute code directly and focus engineering talent on complex innovation.
