Software Engineering Daily SmartBear and Multi-Agent QA
16 snips
May 5, 2026 
Fitz Nowlan, VP of AI and Architecture at SmartBear and co-founder of Reflect, brings expertise in AI-native QA and web test automation. He discusses why web UI testing is uniquely hard. He explains BearQ’s multi-agent approach to exploring apps, coordinating tester agents, and managing test data at scale. He also covers how agents infer app structure and where humans fit into the outer loop.
AI Snips
Chapters
Transcript
Episode notes
AI Has Moved The Bottleneck To QA
- AI coding tools have shifted the SDLC bottleneck from authoring code to validating and testing at much higher velocity.
- Fitz Nowlan explains BearQ was built to match AI-driven development speed by using autonomous agents to explore and test web apps continuously.
Use Separate Exploration And Test Runner Agents
- Run exploration sessions plus targeted test-run sessions using browser-attached agents that take screenshots and write results after completion.
- Fitz explains agents spin up browsers, generate valid random inputs (e.g., unique emails), and author new tests from exploration.
Multiagent Design Uses Lightweight Workers And QA Leads
- BearQ uses a pool of async worker agents that instantiate per task; QA lead agents are heavier and consulted only when test runners are stuck.
- Communication occurs via pub/sub and blob storage to avoid moving large screenshots inline between agents.
