Uber's Genie answers on-call questions in Slack and saved ~13,000 engineering hours

Uber built Genie, a retrieval-based copilot that answers engineers' questions in internal Slack support channels using wiki pages, internal Stack Overflow, and docs.

Organization: Uber
Industry: Ride-hailing & Delivery
When: Oct 2024
Evidence: Company reported
Agents and tools: OpenAI embeddings RAG Slack

Reported outcome

70,000+ questions answered across 154 Slack channels, with an estimated 13,000 engineering hours saved.

70,000+
Questions answered
154
Slack channels
~13,000 (estimate)
Hours saved
48.9%
Rated helpful

Caveats

Stated by the company that uses the agent (blog, earnings call, interview). Not independently audited.

Hours saved is Uber's own estimate.
Helpfulness (48.9%) is rated by users, so about half of answers were not marked helpful.

What the agent does

When an engineer asks a question in a support channel, Genie searches internal docs and posts a draft answer with links. The channel's on-call engineer can then confirm, correct, or skip.

Why this works

Uber's support channels got about 45,000 questions a month. Many had answers already written somewhere. The problem was finding them. Retrieval solves "the answer exists but nobody can find it".

What to copy

Start with one busy channel. Index only the docs that channel's team trusts.
Always show source links, so people can check the answer.
Collect a 👍/👎 on every answer and review the 👎 ones weekly.

Starter prompt

Written by us, not by Uber.

You answer questions in the #{channel} support channel.
Use only the retrieved documents below. Quote the relevant line and
link the source for every claim.

If the documents do not answer the question, reply exactly:
"I could not find this in the docs. Tagging on-call." and stop.

Keep answers under 120 words.

Documents:
{retrieved_chunks}

Question:
{question}