Originally reported November 14, 2026.
Inkitt has built an AI-assisted workflow for small, low-risk software changes. It can take a task submitted through Slack, Linear or an API call from another agent and produce a draft pull request (PR) for testing and review. Inkitt Tech says the system is used to deliver code daily.
Chief engineer Agi Letkiewicz Tanner describes the system as an event-driven orchestrator coordinating short-lived agent containers. Separate stages handle requirements, implementation, validation, CI failure repair and merge-conflict resolution. The orchestrator stores each run’s state in a database so work can resume from a recorded point.
The system has defined limits. It rejects tasks involving database migrations, schema or shared-contract changes, CI and container-build configuration, authentication, payment code, or package additions and updates. Validation, CI repair and merge-conflict resolution are each limited to two attempts. A separate check examines the Git diff for changes to restricted paths.
According to the account, the orchestrator retains credentials for external services, while agent containers receive short-lived repository tokens and model credentials. It also deduplicates repeated webhooks and checks GitHub to confirm that all relevant CI jobs have finished, rather than treating an individual job event as completion.
A person still controls the final steps: the system creates a draft PR but cannot bypass branch protection or merge it. After CI passes and the change reaches a pre-release environment, it notifies the task requester. A person must then mark the PR ready for review and manage reviewer requests and approvals.
Source: infoq.cn (originally published 2026-11-14 10:00 UTC+8)