App factory or agent fleet manager: the fork is whose rate limit stops the work

Bring-your-own-credentials sets the size of the fleet, not just the invoice. runtimewire.com reports that Conductor users inherit the rate limits and concurrency ceilings of their Claude Max, Cursor or Codex plans, and that the agents stop when those quotas are spent even if Conductor itself is ready to keep going ^[12]. The same piece says Replit runs on capacity negotiated with the model providers, and calls the resulting ceiling often higher and more predictable than individual consumer or Pro quotas, without attaching a figure to it ^[13]. One side of that comparison is a number you can read off your own billing page. The other is an adjective.

What Replit does publish is a task ceiling: up to 2 concurrent tasks on Core and up to 10 on Pro, each running in an isolated copy of the project, tracked on a board that moves work from Draft to Active to Ready to Done and merges back after review with automated conflict handling ^[11]. Ten is five times two ^[16], and ten is also more open branches than a small team will review in a working day. An orchestration layer bought as a separate product earns its keep above that line, not below it.

Model choice is where the two stop being substitutes. Replit's published Max Mode list runs Claude Fable 5, Claude Opus 4.8 and 4.8 Fast, Claude Opus 5 and 5 Fast, Kimi K3 and GPT-5.6 Sol ^[10]. Five of those seven entries are Claude ^[15], and neither Cursor Agent nor Codex appears ^[17]. Since the picker and the Effort slider both sit behind the paid tiers ^[9], a team cannot run the like-for-like test against its existing Claude Max subscription without first paying for the platform it is evaluating.

Replit also has Shell and SSH access, full mobile apps, and the option to move work onto reserved private machines ^[14], and it carries the runtime, databases, authentication and hosting in the same system as the agent ^[8]. runtimewire frames the split as all-in-one app factories on one side and management software for fleets of coding agents on the other ^[18]. For a team whose repositories already sit inside the app factory, the fleet manager is a layer to add when the review queue becomes the bottleneck, rather than when the laptop lid does.