Filestage reports 50% more merged PRs with AI agents, credits CI guardrails

Filestage's Elio Capella says the team went from about 200 to 300 merged PRs a month with no rise in bug reports, crediting both better agents and the team's CI guardrails.

In a blog post published October 2, Elio Capella says his team at Filestage went from around 200 to 300 merged PRs a month "without an increase in bug reports." He puts the cost at a $20 monthly AI subscription per developer, and says each developer picks their own tool: Codex, Claude Code or Cursor. The headline is about giving up hand-written code. The useful part for builders is what he credits alongside better agents. CI guardrails give feedback every time an agent makes a change: linting, type checking, duplication checks, 100% test coverage and end-to-end tests.

Treat it as one self-reported data point. The post doesn't say how bug reports were counted or over what period. It doesn't give team size or say whether PRs got smaller. The only other coverage found, El Ecosistema Startup, repeats the blog's numbers.

Capella also says the extra throughput exposed CI bottlenecks. The team hit Cloudflare's free tunnel limits because end-to-end test instances need to receive third-party webhooks, and it had to give the shared development database more resources. He now considers merging without an AI code review irresponsible. He hasn't settled whether passing checks plus AI approval could be enough to merge.

Larger datasets look less tidy. Faros AI's 2026 report is based on telemetry from 22,000 developers. Comparing each organization's lowest- and highest-AI-adoption periods, it reports PR merge rate per developer up 16.2% and incidents per PR up 242.7%. A summary of DORA's 2026 ROI report argues that automated CI testing is what separates gains from instability. If you want to copy Filestage, copy the guardrails. Also track change failure rate and incidents per PR alongside throughput, as Ideatives recommends.

Sources