Zapier

Zapier has spent 14 years accumulating 8000+ integrations, and that inventory is also a liability. Lisa Orr, an engineering leader at the company, frames the problem as "app erosion": third-party APIs change constantly, so integrations degrade continuously and support tickets arrive faster than they are resolved. The backlog crisis she describes threatens reliability, customer experience, and potentially churn.

The response was two parallel experiments started two years before her account. The first moved support staff from triage to fixing bugs directly, hedged with guardrails: four target apps, mandatory engineering review, app fixes only. The second, Scout, applied generative AI after Orr's own dog-fooding surfaced the real bottleneck, 50% of the time went to gathering context. Scout answers that with a context analyzer, a diff generator, and a test generator. Merge request success reached 97%, and resolution time fell from 72 hours to minutes.

Zapier recurs across the sources as an adoption datapoint rather than a product. Anthropic's 2026 agentic coding trends report cites 89% AI adoption there alongside «800+ agents internes». Justin Reock of DX cites the company's onboarding bots, which made new engineers productive in 2 weeks instead of 90 days, as an example of an organization aiming AI at an actual constraint rather than at coding time.

What the sources leave open is where the boundary between support and engineering settles once it has been deliberately blurred.