Your AI Doesn’t Know What It Can Do. Make It Find Out.

An AI Essentials Roundtable member used Grok to plan a trip to Niagara Falls. It did the strategy piece well. Alternate towns away from the fray. Dog-friendly stays. What the area actually feels like if you do the Falls as a day trip.

Then it stopped. Booking the hotel still meant opening Kayak and Trivago and Hotels.com and running the comparison yourself. Strategy, not the hard work.

That is a real gap. It is also sometimes a capability gap in the tool, not in the job. Grok has browsers. It can click through sites. I have used it to pull actual flights. Whether it can finish the booking work depends on the plan you are on, and on whether the site will let an AI through. It also depends on whether the model even knows it can try.

It will tell you no. That is not always the answer.

Grok, more than Claude, is often unaware of its own capabilities. It will say it cannot update a Google Sheet or access a site. I have been relatively certain there are multiple ways it can. Yet, it still said no.

So I often tell it to search X and see what people say about its own capabilities. It goes and looks, and then says: oh, they are saying I can do it this way. Then you can get the thing done that it said can’t be done!

That has happened a lot. Keep it in mind. Claude is much more aware of what it can do. Grok, in my opinion, has more raw capability. It can run programs in its own environment. It just does not always know that.

The push that actually works

When it shrugs, do not accept the first no. Tell it you think it can find a way. Ask it to dig. Ask it to search the X community and see what people say it can do. Then make it use what it just learned.

Same instinct as asking for twenty ideas instead of the first best answer. The model will hand you the obvious limit. Your job is to refuse the short version.

Related: Ask for Twenty Ideas, Not the Best One.