Vibeleaderboard
← All Intel
Intel / video

We Built an AI Support Agent That Resolves 80% of Tickets — AssemblyAI

Source
youtube.com
Author
AI Engineer
Date
From the Daily Brief

AssemblyAI's talk is useful because it attaches numbers to a support agent. Joey runs on the Claude Agent SDK and works from a checked-out copy of the company's documentation, combining embeddings with agentic search and a single CLAUDE.md file that holds its instructions. AssemblyAI says it resolves 80% of support tickets for about $700 a month. Two operational details stand out. Escalations that the agent cannot handle are turned into roadmap items, so failures feed product work rather than disappearing into a queue. And the team built a deploy loop that ships fixes to the agent in about 30 seconds, which keeps the cost of correcting a bad answer low. The agent later gained a voice channel through a websocket. These are the company's own figures from a conference talk, and the resolution rate depends on how a ticket counts as resolved. Still, it is a concrete reference architecture for teams pricing a similar build.

Read the 2026-10-05 Brief →

Context

AssemblyAI reports that its support Joey resolves 80% of conversations end to end, compared with 10% for the previous off-the-shelf bot, for about $700 a month. These are company-reported results for its own support workload, presented in this conference talk.

Joey combines Anthropic's Claude Agent SDK with locally stored documentation, -based retrieval and agent-directed file search. Cases that still require human help become items on the agent's improvement roadmap. The example connects access to product knowledge with a feedback process built from unresolved customer questions.

Key takeaways · AI-distilled
  • AssemblyAI built Joey because it gets about 1,000 API signups a day with one onboarding engineer, and the off-the-shelf support bot it replaced resolved only 10% of conversations end to end.
  • The voice version uses AssemblyAI's Voice Agent API over a single websocket that handles speech-to-text, the step and text-to-speech.
  • Matt Lawler's broader argument is that forward deployed engineers should automate themselves out of a job and dogfood the product their customers build.
Terms in this piece · Glossary
  • AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
  • embedding — A list of numbers representing a piece of text's meaning, so that similar meanings end up numerically close and can be searched.
  • LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
Read the source www.youtube.com
More from AI Engineer
Recommended reads
Comments

Checking sign-in…

Loading comments…