Back to feed
News Story
InfoQ
1 sources

Calendly Reveals Agentic Engineering Practice: Agents Generate 64 PRs in a Week and a Half, Completing 30% of a Major Refactor

Calendly has publicly shared its internal Agentic Engineering practice, where agents are embedded throughout the development workflow. In the IAM team, agents generated 64 pull requests in just a week and a half, completing about 30% of a large decoupling project. The article highlights that as execution becomes automated, the bottleneck shifts to 'judgment,' and human involvement remains crucial.

SynthePulse Insight · AI deep readingMembers

When Agents Enter the Full R&D Workflow of a Mature SaaS: Calendly's Practice and the Bottleneck Shift

Version 1 · 1 source

Calendly has publicly shared its internal Agentic Engineering practice: agents autonomously plan, implement, and test within systems like Jira, GitHub, and CI, while humans handle final review and merge. In the IAM team, agents generated 64 PRs in a week and a half, completing about 30% of a project refactor. Calendly believes the bottleneck has shifted from writing code to judgment.

  • Calendly publicly shares its internal Agentic Engineering practice, with agents embedded in Jira, GitHub, CI, etc., while humans retain final review and merge.
  • In the IAM team, agents generated 64 PRs in a week and a half, 37 of which were merged, completing about 30% of a decoupling project.
  • Calendly believes the bottleneck has shifted from writing code to judgment, making requirement quality, scope quality, and review quality more important.
Open section navigationThe Agentic Engineering Loop: From Jira to Production

The Agentic Engineering Loop: From Jira to Production

Calendly's Agentic Engineering practice is not theoretical; it is already running in its engineering organization. The loop typically starts with a one-pager or Jira ticket. Agents first review the ticket for ambiguities and missing details, then planning agents break the work into PR-sized tasks. Implementation agents execute in parallel in isolated worktrees or self-hosted workers. QA or reviewer agents validate the output, and finally, humans perform the final review, approval, and merge.

The system has persistent state: Jira holds requirements, GitHub holds code review and change history, CI holds validation results, and repository instructions and shared rules constrain agent behavior. Calendly emphasizes this is not a clever prompt but a closed-loop system with persistent memory.

Calendly clearly distinguishes what Agentic Engineering is not: it is not AI writing all the code, not unrestricted autonomous operation, and not replacing engineering responsibility with AI. Any engineer whose name appears on agent-assisted work must still be responsible for it.

Free for now

Read the full analysis

4 more sections of analysis, plus the full takeaway

Loading

Credibility boundary

This article is based on an InfoQ compilation of Calendly's official blog, representing a company's self-disclosed practice sharing. Data (such as 64 PRs, 30% refactor) are Calendly's unilateral claims, not independently verified by third parties. The independent research (CodeRabbit) is cited as a paraphrase, and specific research details are not provided.

Sources for this version

  1. 1周半写64个PR、完成30%项目重构：一家成熟SaaS公司把Agent塞进研发全流程后InfoQ

Primary report

InfoQ

Primary source