Celigo says AI agents built almost all of Ora, with humans approving every merge. Is this the new software team?

Celigo’s CTO published an unusually concrete account on September 2 of how the company says it built Ora, its agent for business integrations. Ora is now generally available after a six-month beta and more than 16,800 real conversations. According to Celigo, the product has been built almost entirely by AI agents since its first commit in June 2025. Humans still set direction, judge outcomes and approve every merge. The interesting part is not simply “AI wrote the code.” Celigo says it redesigned the surrounding engineering system: documentation is organized for “tokens-to-competence”; tribal knowledge became explicit, always-loaded rules; the suite grew to tens of thousands of tests, including thousands of AI-graded evaluations; agents review other agents before a person judges the outcome; every customer-facing change is staged for approval before it touches an account. Celigo claims output nearly quadrupled in one month and has run at almost 700% of its starting pace over the past six months, with the same small group of people. Those are striking claims, but this is still a vendor-authored case study. There is no public repository, staffing denominator, reviewer-hours figure, escaped-defect rate, rollback rate or cost per accepted change. Commit velocity alone is not customer value. If a company says agents rebuilt a mission-critical platform, which evidence would convince you: lead time, escaped defects, rollback rate, reviewer hours, cost per accepted change—or something else? Disclosure: AI-assisted draft, checked against the company’s primary source. No affiliation.