Intercom

Overview

Intercom is a customer service software company that has surfaced in AI product and engineering discussions for how it applies AI internally to improve software delivery. In the newsletter coverage here, Intercom stands out less for its customer-facing product and more for its operating model: it used Claude Code workflows, internal instrumentation, and open-source model training to drive measurable gains in engineering productivity.

For AI Product Managers, Intercom is notable because it provides a concrete example of moving beyond vague “AI adoption” claims into operational metrics and workflow design. Its reported 2× increase in merged pull request throughput, paired with logging of skill invocations and session data, offers a useful case study in how to set AI goals, instrument usage, and connect model-driven workflows to business-relevant outcomes.

Key Developments

• 2026-03-27: Intercom was highlighted alongside Pinterest, Airbnb, Notion, and Cursor as a company training open-source models in-house, with the claim that these models were cheaper, faster, and more effective than API-based approaches for some use cases.
• 2026-04-21: Intercom reported a 2× increase in merged pull request throughput within nine months after going all-in on Claude Code. The company built and instrumented specific workflows such as PR description quality enforcement and an autonomous flaky-specs fixer, while logging every skill invocation to Honeycomb and session data to S3. The effort was tied to a clear goal set by CTO Darra to double throughput per R&D head.

Relevance to AI PMs

• Use outcome metrics, not just adoption metrics. Intercom’s example shows the value of anchoring AI efforts to a hard operational KPI like merged PR throughput per R&D head rather than softer metrics like number of active users or prompt volume.
• Instrument AI workflows at the task level. Logging each skill invocation to Honeycomb and session data to S3 illustrates a practical measurement pattern: track workflow usage, latency, quality, and downstream outcomes so you can understand which automations actually create leverage.
• Treat AI as workflow design, not just model selection. Intercom’s gains came from building concrete workflows—like PR-quality enforcement and flaky-test fixing—not merely adopting a coding assistant. For AI PMs, this is a reminder to define repeatable jobs-to-be-done and engineer around them.

Related

• Claude Code: Central to Intercom’s reported productivity gains, serving as the foundation for internal coding workflows and automations.
• Brian Scanlan: Associated with the discussion of how Intercom scaled engineering velocity with Claude Code.
• Darra: Intercom’s CTO, who reportedly set the goal of doubling throughput per R&D head.
• Honeycomb: Used by Intercom to log and analyze skill invocations, making AI workflow instrumentation observable.
• S3: Used to store session data from Claude Code workflows for later analysis.
• Snowflake: Relevant as part of the broader analytics and data infrastructure ecosystem connected to AI measurement and operational reporting.
• GitHub CLI: Related to developer workflow automation, which complements the kind of engineering instrumentation and tooling Intercom implemented.
• Pinterest, Airbnb, Notion, Cursor: Peer companies mentioned alongside Intercom in the context of training open-source models in-house.