AI Coworkers Could Be Great, But Only If They Get Serious

On Sunday morning, a lot of us woke up to Grok Bot, the always-on AI assistant, decidedly off. “100% used” read my quota meter. It’d been 45% late the night before. It was an error, but of the Freudian slip sort, and that’s a problem if tools like this are to be taken seriously.

My favorite use of Grok Bot since I first wrote about it has been for handing off jobs like posting my church’s live stream link to the church social media accounts. I can focus on getting the stream up on YouTube, Facebook and X, while my Bot does the promotional grunt: posting links for our three different streams to each social and doing so intelligently, linking to the streaming service best fit to the audience and noting, for example if I put something special in a given service’s YouTube description.

It’s great, because as a pastor, the last thing I need to be doing was spending the precious minutes before a service posting the necessary slight variations to a bunch of different places. Yet, members of the church counting on the livestream (not to mention online visitors) need those links to join in.

Yes, there are multi-posting tools, but this is better than anything I’ve tried before. Often those just meant having a single page to still create the different manually tuned variants. Negotiating the menagerie was still on me. Grok Bot can use its own web browser to take care of things for me even when an API doesn’t exist and it usually gets new projects figured out quickly and repeats run even faster.

You surely have a whole bunch of similar chores — repetitive tasks you do with little thought. Sure, there might be an API or other means to automate the work, but few of us have the time to get a reliable automation going and give up the first time it hits an error. Wouldn’t it be nice to just say a few words in plain English and know it is taken care of?

Of course it would. SpaceX promotes its tool as being equivalent to adding coworkers and, as the only full time staff member at my church, when it works, it feels like it is one. (No, before someone gets outraged, I’m not stealing a good wage from someone. You can’t squeeze an extra salary out of a budget that barely is able to provide for a single one. It just isn’t happening.)

I already had an X Premium+ account to cover my usage of the very nice Claude Code competitor, Grok Build. Bots appeared as an inviting, “free” cherry on top. So, over the last couple of months, I’ve kept trying to spot those drudgery-type items and see if Grok Bot could handle them. It consistently has, but I’ve still had good reason to distrust it despite the appeal.

Like most AI products, the usage quota for it is notoriously opaque. Grok, Claude or ChatGPT, it makes no difference: ask how much work you can get done on a given plan and, at best, you’ll be told it is a certain amount more or less than the company’s own lesser or more expensive plan.

Want to compare usage quota from Claude to those of ChatGPT or Grok or Gemini? Good luck.

That’s a problem if a plan is promoted as giving you a new coworker. Imagine hiring an employee who set his own pay scale and wouldn’t tell you how many hours he’d work for that, only “I’ll work four times more than I would under a different plan.” No one would hire someone on such an ill-defined basis.

At least, one would hope, though, those comparison numbers would work within a single company’s plans, right? Even then, it’s complicated.

After seeing about what I could expect Grok Bot to do in a week, I developed a feel for a pace that would keep getting things done without risking exhausting the whole quota too early. I’ve been able to guess about how much each percentage point was worth in the Bot system for my particular tasks, keeping an eye for a margin that helps when the still-too-impressionist usage pool goes a tad faster than expected. A last-minute “blitz” each week became part of my routine, devouring the last big chunk on jobs I set aside right before it expires.

Then came the X Subscription. SpaceX not only has ingested the former xAI and X, it also bought programming tool maker Cursor. (Cursor’s infrastructure, in fact, is behind Grok Bot.) The X Subscription was supposed to solve the company’s growing variety of subscriptions with one plan that covers everything.

Great, except it changed how they calculated usage. According to SpaceX, the new slate of plans provides more usage per dollar than the plans they replaced. Fair enough, but the plans didn’t all keep the same price tag, so the amount of usage in a new plan compared to its most analogous old plan is murky.

The original X Premium plan offered a vaguely defined amount of access to Grok (but no Grok Bot) for $8 and was replaced with a “Plus” plan of the same price that seems to have more features, but potentially a tighter watch on usage. Not great, but also not unexpected in a plan where AI usage was always more of a bonus than a focus. The real question for me was what happened to the most affordable “real AI work” plan, the former $40/month Premium+, now replaced with the cheaper $30/month Premium plan — the sweet spot for most people leveraging AI for real use.

The $10/month savings in the transition might be the most expensive “savings” you’ll ever find. The old plan included separate allowances for both Grok/Grok Build and Grok Bot. The new plan combines those into a single, unified allotment. On the surface that would be great: no wanting to use Grok Bot only to find all the remaining quota is limited to regular Grok, for example. But that’s only great if the unified quota is actually equivalent to both old quotas.

They promised no such thing and given it’s 25% cheaper per month, it was immediately obvious there was wiggle room here. The overall allowance could be 24.9% smaller and still meet “cheaper per dollar,” as advertised. And that’s a problem.

As my finger hovered over the “upgrade” button to the new, cheaper plan, I wanted to know: was that a hypothetical worry from my cynicism or a real concern? Multiple attempts to get the company to explain the relationship of actual usage in the old and new plans were unfruitful. Can you imagine even a provider as unlikable as your cable or cell phone company saying, “I can’t tell you if your new plan has as much usage as your old one”?

Hitting that wall, I did the only thing I could to pin it down: set up a second, parallel account and test the same tasks in the old Premium+ and the new Premium.

The result? The new plans are definitively worse than the old. Given any particular Grok Bot task I assigned, the Grok Bot on both plans drained approximately the same amount of quota. The same when using Grok Build on both accounts. This held out through numerous and varied tests. That’s actually a very bad thing. Remember: the new plan has a single unified quota instead of the old one’s two, distinct usage pools. If you only use one tool, no problem; but if you try to maximize the functionality of your account, my testing suggests the new plan offers roughly half the work allowance the old one did.

True, if you sign up and only use a small portion of your allotment, the cost savings might be the better part of that change. But, if you actually put that new coworker to the kind of serious work Musk is always showing off, getting that much less usage is a major knee-capping.

Ah, but can’t you just pay per use or grab a bigger plan?

When I failed to pace myself well and my routines ran out a day early, I attempted the à la carte usage route that seems like the obvious solution. A single day of timid usage burned an extra $20. To put that in perspective: buying an extra 1/30th of my monthly plan’s usage cost me 50% of the original plan price. Anyone trying to gain back the usage lost in the $40-to-$30 plan change will spend far more than the $10 saved just to get parity.

Takes the sheen off that ten buck savings, doesn’t it?

And that is aside from the 100% usage glitch I encountered on Sunday, which did prove to be more than a mere hiccup. As people scrambled over their accounts suddenly grinding to a halt, Musk responded about SpaceX’s need to bring usage allotments in line with the costs of running the service.

It looks like Musk (or others at the company) grew nervous about being victims of their own Grok Bot success and simply tried to reconfigure the numbers between Saturday evening and Sunday morning.

There was a bug, but it wasn’t the reduced quota; it was the oversight that changing it midweek would take many users from ample room in their quotas to far beyond exhaustion in a blink. Dramatically reducing people’s usage of a tool meant for serious work, and not just hobbyist amusement, is never going to go over well, even if it had been done with ample warning and care in how it was applied.

The quotas did get restored on Monday, but the routines were not there to help me (or many others) with our Sunday tasks. And, the willingness to attempt that change unannounced should give us pause on trusting the quota and pause because this likely only forestalled the inevitable.

It’d be easy to blame Musk’s mercurial approach for this, but SpaceX is hardly alone here. Anthropic has moved my quota reset day around at a whim in hard-to-manage ways, too. When a recent billing issue came up, I was told it just was what it was by the company’s none-too-helpful chatbot that refused me the ability to appeal to a human.

OpenAI recently played with its usage quotas too in ways that were both opaque and less than user-beneficial.

In other words, these companies are, by any other industry’s standards, deeply unserious. As I said, even the most hated companies we deal with heretofore wouldn’t have imagined they could get away with the pitch, “We’ll give you a vaguely specified amount of our product, and we might change how much we give you and it might even change regularly throughout the year, but we can’t really tell you precisely how much you have or will have or ever had.”

You can’t ask people to depend on a tool to do things when they can’t even know if an amount of the resource they budgeted for (and was adequate to the task today) will be enough to do the job tomorrow. Tools that one constantly worries about running out mid-task are not particularly helpful even if they functionally are quite capable.

As Grok Bot, Meta Muse, OpenAI Dots and other similar offerings go head to head, the first company that makes usage predictable and transparent will, by definition, be more helpful than the rest. And that will be the one that wins.