AI Coding Weekly

Orosz says only Grok Bot ran his cloud workloads

One engineer's day, and an argument for never holding a single AI subscription.

Gergely Orosz says neither Codex nor Claude Code figured out how to run his workloads on a cloud machine, while Grok Bot did. That is the capability half of his post. The other half is availability. Claude, he says, was dead for him that day on the specific task he wanted it for, so he moved that task to Grok Bot.

Gergely Orosz
@GergelyOrosz
X
amusing how Grok Bot leapfrogs the two leading labs, and those two labs simply don't or cannot catch up
Nov 13, 2026 · View on X
Gergely Orosz
@GergelyOrosz
X
One reason you really cannot just have a single subscription is how unreliable each and every AI service is
Nov 13, 2026 · View on X

His conclusion from the pair of problems is a purchasing one rather than a ranking one. You cannot just have a single subscription, he argues, because of how unreliable each and every AI service is. Note the scope of that claim. He is not saying one vendor is worse at uptime, he is saying all of them are unreliable enough that a single point of failure in your tooling is a bad bet.

What this is and is not

This is one engineer's day, not a benchmark. There is no eval here, no repeated runs, no detail in the post about what the workloads were or how hard either agent tried. Treat it as a report of a failure mode, not as a measurement of which coding agent is better at cloud execution. Nobody has published a comparison on this specific capability.

Orosz also flags a gap in his own test. He says he does not know whether Dots does any of this, because it is not available in the EU, and Grok Bot is. So the comparison set is whatever he could actually sign up for and run, which is a real constraint for anyone working out of Europe and a reason to read his lineup as availability driven rather than exhaustive.

For working engineers

The part worth pricing is the second subscription. Most teams buy one agent plan per seat and treat the vendor's uptime as somebody else's problem, which works right up until the day the service you standardized on is down for the one task that is blocking you. Orosz's argument is that redundancy in agent vendors is an operating cost, in the same way you would not run production on a single provider with no fallback. Whether that is worth a second monthly bill depends on how much of your work is actually agent shaped, and his post does not try to do that math for you.

He sounds more entertained than annoyed about the ordering, noting that Grok Bot leapfrogs the two leading labs and the two leading labs do not or cannot catch up. Worth watching whether anyone reproduces the cloud machine result on a larger set of tasks.

Have the good opinion, not the loud one

One email a day. Under three minutes. What working engineers found out about coding with AI.

Get the next one by email

Coding with AI, read daily so you do not have to. The experiments, the receipts and the arguments from engineers shipping real software. Not a changelog.