United StatesTechnology and telecom2026 – ongoing

Deepgram

Engineering processes rebuilt around Claude Code

Deepgram's regular Claude users shipped 4-10x more durable code than non-users

WorkingVendor case studyVendor ↗

Problem

Deepgram, a maker of speech models, saw the cost of writing code fall and expected competitors to ship faster. It wanted its engineering teams to keep up.

Approach

After a bake-off of coding agents, Deepgram rebuilt its engineering workflow around Claude, an effort it calls DG2. Claude Code connects through MCP to Slack, GitHub and Grafana, and reusable fixes are saved as shared skills. Agents with log and metric access check their own diagnoses during incident triage, and Claude reviews code against rubrics for each stage of the lifecycle. In mid-2026 the whole company moved to Claude Enterprise.

Results

4-10xmore durable code from regular and power Claude users than from non-users, in an internal cohort studyVendor case study[1]

about 95%share of code written by Claude on its most productive teamVendor case study[1]

40+substantial pull requests shipped by one new hire in six weeksVendor case study[1]

Takeaway

The gains came from the tools wired into the workflow as much as from the model. Deepgram also saw about 9x higher code churn among its heaviest users.

Sources

Deepgram ships 4–10x more durable code with Claude
Anthropic22 Aug 2026Vendor publication

Source summaries

[1]Anthropic22 Aug 2026

Anthropic's customer story describes how Deepgram rebuilt its engineering workflow around Claude Code and moved the company to Claude Enterprise. It reports 4-10x more durable code from regular Claude users in an internal study, about 95% Claude-written code on the most productive team, and incident triage cut to minutes. The page carries no date; the Internet Archive first saved it on August 22, 2026.

Full details are in the original sources.

Added 4 Nov 2026 · Updated 4 Nov 2026