Replit's August 21 bundle sells scheduled agents; the governance half is still a label

The pricing seam is where the design shows. Free Mode does not draw down Power Mode or Max Mode credits, Replit picks the model itself, and the allowance is tied to a user's subscription ^[4]. A Routine, by contrast, schedules repeated work inside a Conversation, runs in Power or Max, and needs a budget for every execution ^[5]. The billed surface therefore stops tracking how much someone explores and starts tracking how often a job fires ^[13]. A team that turns one useful prompt into a nightly task has bought a recurring line item.

That makes the governance items the ones worth reading closely, and they are the thinnest part of the record. Runtimewire's account of the August 21 changelog, sourced to Raouf Chebri's thread on X ^[2], describes Free Mode, Routines, Skills imports and Steer in working detail, while enterprise model policies and the three-part security test appear only as names in a summary line ^[15]. Those two are what a buyer actually evaluates: which models an agent may call, and what runs before code ships. Two of the seven releases, Free Mode and the Level 3 security scan, had also already rolled out separately ^[3], which puts roughly two sevenths of the bundle in the re-announcement column ^[12].

There is a mechanical gap underneath the scheduling as well. Steer, the default in Conversations and Projects, lets someone push new context while Agent is mid-task ^[9]. A Routine's output returns to the same thread for review and follow-up ^[6]. Those two facts do not compose: a job that runs on a schedule has nobody present to steer it, so correction moves from mid-flight to after the fact ^[14]. That is the argument for a pre-ship scan being a gate rather than a report, and it lands on the audience Haya Odeh designed the product around, people who do not identify as programmers ^[17], working in a place where hosting, databases, authentication and deployment sit in the same workspace ^[11].

The one control here a buyer can inspect today is Skills. Import them from a public repository, folder or file URL, and a project owner marks each saved Skill Required, Available, or inaccessible to members ^[7]. Because the files live in GitHub, coding standards and test practice can be versioned and reviewed where the rest of the review already happens, instead of only inside Replit ^[8]. Runtimewire frames the bundle as prompt-based app creation becoming repeatable, governed work ^[18]. The repeatable half is documented. The governed half is a label on a box.