Time: 9:45 AM
Speaker Bio: Lead Product Manager at Zapier. Former senior PM at Human API and Airship.
Speaker Profile: Full Speaker Profile
Company: Zapier automates workflows across apps. Recently launched Custom Actions that use AI to enhance automation capabilities.
Focus: How support teams can leverage AI and automation to ship code. This is about enabling non-engineers to build.
Slides
Slide: 09-55
Key Point: Demonstrates an automated workflow where a “Scout” system categorizes support issues, assesses feasibility, generates code fixes, and integrates with human support for testing and review, creating a semi-automated bug fixing pipeline.
Literal Content:
- Title: “Automating app fixes with Scout + Support”
- Flowchart showing two paths:
- Left path (support submits issue): Scout categorizes issue → Post Assessment (if not feasible/wrong category) → Scout assesses feasibility → (if feasible) → Scout generates code fix
- Right path: support submits feedback → support tests fix → support submits fix for review
Notes
- Parallels between Grand Canyon and Zapier: erosion
- Good story about how AI agents are in the system
- True value comes with tiering the tools together
- Combine them to orchestrate them
- Key point is that support is doing the “coding” directory
- Ton of dogfooding at Zapier
- Diagnose
- Figure out if its fixable
- Then
- Plan, execute, validate
- Scout MCP and Cursor MCP
- Couple of evals
- ~70% for accuracy
- ~fixability
- Impact app erosion
- 40% of support teams videos done by scout - doubling support velocity, 3-4 fix/week with scout
- Puts potential fixable solutions on the backlog - the analysis is helpful in triage
- Support + code gen = magic
- They are closest to customer pain
- Real-time troubleshooting
- Best at validation
- Support team members who are part of this experiment are moving into the engineering team