--- schema: 1 type: video title: We Built an AI Support Agent That Resolves 80% of Tickets — AssemblyAI description: >- 1,000 API signups a day and one onboarding engineer. So AssemblyAI built Joey. Matt Lawler, forward deployed engineer lead at AssemblyAI, shares how his team replaced an off-the-shelf support bot that resolved 10% of conversations with their own agent, Joey, which resolves 80% end to end for about $700 a month. Joey is built on the Claude Agent SDK with docs checked out as local markdown, embeddings for retrieval, agentic file search and a large CLAUDE.md. It's deployed on Railway so fixes ship in about 30 seconds. Then Matt gives Joey a voice with AssemblyAI's Voice Agent API and demos it live, making the case that FDEs should automate themselves out of a job and dogfood the product their customers build. In this talk: • Why the FDE model doesn't scale to 1,000 signups a day • Joey's architecture: local markdown docs, embeddings, agentic search and a single CLAUDE.md • Turning every escalation into a roadmap item for the agent • Adding real-time voice with one websocket: speech-to-text, LLM and text-to-speech SPEAKER Matt Lawler, Forward Deployed Engineer Lead, AssemblyAI LINKS AssemblyAI: https://www.assemblyai.com Voice Agent API docs: https://www.assemblyai.com/docs/voice-agents/voice-agent-api Introducing the Voice Agent API: https://www.assemblyai.com/blog/introducing-our-voice-agent-api AssemblyAI on X: https://x.com/assemblyai CHAPTERS 0:00 Intro 0:33 Meet Matt 1:13 What AssemblyAI does 2:08 1,000 API signups a day 2:38 What a forward deployed engineer does 3:28 Why the FDE model doesn't scale 3:57 Automate yourself out of a job 4:12 The off-the-shelf bot: 10% resolution 5:17 Building Joey 6:12 Architecture: docs, embeddings, agentic search 7:22 Deploying on Railway 8:02 Results: 10% → 80% for $700 a month 9:01 Escalations become the roadmap 9:31 Giving Joey a voice 10:56 Code walkthrough 12:11 Live demo: talking to Joey 14:24 Dogfood your own product 15:08 Go build Recorded at the AI Engineer World's Fair 2026 in San Francisco. Subscribe for more talks from the engineers building with AI. AI Engineer: https://ai.engineer YouTube: https://www.youtube.com/@aiDotEngineer X: https://x.com/aiDotEngineer LinkedIn: https://www.linkedin.com/company/aidotengineer/ #VoiceAI #AIAgents #AIEngineer canonical: 'https://liquidgold.cc/watch/pyvRID_CZZU' markdown: 'https://liquidgold.cc/watch/pyvRID_CZZU.md' llms: 'https://liquidgold.cc/watch/pyvRID_CZZU/llms.txt' parent: 'https://liquidgold.cc/channel/aidotengineer.md' channel: aidotengineer updated: '2026-10-23T22:20:01.014Z' language: null children: [] videoId: pyvRID_CZZU durationSeconds: 979 publishedAt: '2026-10-04T15:00:15.000Z' searchChunks: - ordinal: 0 text: null startSeconds: 93 endSeconds: 139 watch: 'https://liquidgold.cc/watch/pyvRID_CZZU?t=93' heldReason: rights - ordinal: 1 text: null startSeconds: 93 endSeconds: 139 watch: 'https://liquidgold.cc/watch/pyvRID_CZZU?t=93' heldReason: rights - ordinal: 2 text: null startSeconds: 160 endSeconds: 170 watch: 'https://liquidgold.cc/watch/pyvRID_CZZU?t=160' heldReason: rights - ordinal: 3 text: null startSeconds: 224 endSeconds: 245 watch: 'https://liquidgold.cc/watch/pyvRID_CZZU?t=224' heldReason: rights - ordinal: 4 text: null startSeconds: 224 endSeconds: 245 watch: 'https://liquidgold.cc/watch/pyvRID_CZZU?t=224' heldReason: rights - ordinal: 5 text: null startSeconds: 267 endSeconds: 322 watch: 'https://liquidgold.cc/watch/pyvRID_CZZU?t=267' heldReason: rights - ordinal: 6 text: null startSeconds: 267 endSeconds: 322 watch: 'https://liquidgold.cc/watch/pyvRID_CZZU?t=267' heldReason: rights - ordinal: 7 text: null startSeconds: 333 endSeconds: 376 watch: 'https://liquidgold.cc/watch/pyvRID_CZZU?t=333' heldReason: rights - ordinal: 8 text: null startSeconds: 376 endSeconds: 387 watch: 'https://liquidgold.cc/watch/pyvRID_CZZU?t=376' heldReason: rights - ordinal: 9 text: null startSeconds: 387 endSeconds: 451 watch: 'https://liquidgold.cc/watch/pyvRID_CZZU?t=387' heldReason: rights - ordinal: 10 text: null startSeconds: 451 endSeconds: 462 watch: 'https://liquidgold.cc/watch/pyvRID_CZZU?t=451' heldReason: rights - ordinal: 11 text: null startSeconds: 484 endSeconds: 495 watch: 'https://liquidgold.cc/watch/pyvRID_CZZU?t=484' heldReason: rights - ordinal: 12 text: null startSeconds: 495 endSeconds: 548 watch: 'https://liquidgold.cc/watch/pyvRID_CZZU?t=495' heldReason: rights - ordinal: 13 text: null startSeconds: 559 endSeconds: 581 watch: 'https://liquidgold.cc/watch/pyvRID_CZZU?t=559' heldReason: rights - ordinal: 14 text: null startSeconds: 591 endSeconds: 656 watch: 'https://liquidgold.cc/watch/pyvRID_CZZU?t=591' heldReason: rights - ordinal: 15 text: null startSeconds: 591 endSeconds: 656 watch: 'https://liquidgold.cc/watch/pyvRID_CZZU?t=591' heldReason: rights - ordinal: 16 text: null startSeconds: 667 endSeconds: 732 watch: 'https://liquidgold.cc/watch/pyvRID_CZZU?t=667' heldReason: rights - ordinal: 17 text: null startSeconds: 667 endSeconds: 732 watch: 'https://liquidgold.cc/watch/pyvRID_CZZU?t=667' heldReason: rights - ordinal: 18 text: null startSeconds: 868 endSeconds: 878 watch: 'https://liquidgold.cc/watch/pyvRID_CZZU?t=868' heldReason: rights --- # We Built an AI Support Agent That Resolves 80% of Tickets — AssemblyAI 1,000 API signups a day and one onboarding engineer. So AssemblyAI built Joey. Matt Lawler, forward deployed engineer lead at AssemblyAI, shares how his team replaced an off-the-shelf support bot that resolved 10% of conversations with their own agent, Joey, which resolves 80% end to end for about $700 a month. Joey is built on the Claude Agent SDK with docs checked out as local markdown, embeddings for retrieval, agentic file search and a large CLAUDE.md. It's deployed on Railway so fixes ship in about 30 seconds. Then Matt gives Joey a voice with AssemblyAI's Voice Agent API and demos it live, making the case that FDEs should automate themselves out of a job and dogfood the product their customers build. In this talk: • Why the FDE model doesn't scale to 1,000 signups a day • Joey's architecture: local markdown docs, embeddings, agentic search and a single CLAUDE.md • Turning every escalation into a roadmap item for the agent • Adding real-time voice with one websocket: speech-to-text, LLM and text-to-speech SPEAKER Matt Lawler, Forward Deployed Engineer Lead, AssemblyAI LINKS AssemblyAI: https://www.assemblyai.com Voice Agent API docs: https://www.assemblyai.com/docs/voice-agents/voice-agent-api Introducing the Voice Agent API: https://www.assemblyai.com/blog/introducing-our-voice-agent-api AssemblyAI on X: https://x.com/assemblyai CHAPTERS 0:00 Intro 0:33 Meet Matt 1:13 What AssemblyAI does 2:08 1,000 API signups a day 2:38 What a forward deployed engineer does 3:28 Why the FDE model doesn't scale 3:57 Automate yourself out of a job 4:12 The off-the-shelf bot: 10% resolution 5:17 Building Joey 6:12 Architecture: docs, embeddings, agentic search 7:22 Deploying on Railway 8:02 Results: 10% → 80% for $700 a month 9:01 Escalations become the roadmap 9:31 Giving Joey a voice 10:56 Code walkthrough 12:11 Live demo: talking to Joey 14:24 Dogfood your own product 15:08 Go build Recorded at the AI Engineer World's Fair 2026 in San Francisco. Subscribe for more talks from the engineers building with AI. AI Engineer: https://ai.engineer YouTube: https://www.youtube.com/@aiDotEngineer X: https://x.com/aiDotEngineer LinkedIn: https://www.linkedin.com/company/aidotengineer/ #VoiceAI #AIAgents #AIEngineer Canonical: https://liquidgold.cc/watch/pyvRID_CZZU · Channel: [AI Engineer](https://liquidgold.cc/channel/aidotengineer.md) > Search chunk text is held until public-source provenance is verified. ## Document chunks > Missing: no stored document chunks are available.