It’s almost August, and the summer is winding down. But amid the sticky popsicle wrappers, strewn-about swimwear, and blasting fans, there’s something new in play: agentic AI has LLMs now acting more and more like humans, online and offline.

One of the most interesting pioneers on AI has come out with timely advice on how to use agentic AI, just in time for the end of the summer. As we approach Labor Day and Back-to-school season, Ethan Mollick, an MIT-connected computer scientist of note, is giving us notes on how to direct new agent systems to do our tasks for us. This is all new to most people, so if you are looking to get involved in delegating work to AI agents for the first time, I would give it a read.

One of Mollick’s first points, in the new post on his One Useful Thing blog on Substack , is that for “low-level” or less mission-critical tasks, a free tool can suffice. However, he suggests that users wanting more consistently accurate and capable results might want to shell out the $20 or so for a paid subscription to either ChatGPT or Claude, both of which offer the kinds of agentic tools that Mollick is reviewing.

“If you just want a chatbot that can give you a recipe, answer a low-stakes question, or help you write a letter,” Mollick writes, “there are now tons of options that are good enough, including the default free models. They are all at least fine when the stakes are low, so pick the one you like. But there is an important caveat: if you are chatting about high-stakes issues, like getting a second opinion on a medical or legal concern, you will want the results to be better than ‘good enough’ advice. For these issues, you will want to use the most advanced models you can get access to.”

Another point that Mollick makes is that utility, in many cases, is provided by access: the more you give the tool access to, the more results you will get. He does point out that there are security issues attached to this. In an anecdote about having both GPT and Claude help him with an academic project, he notes how one model sent an actual email, because he had authorized it to do so, while the other did not, because he hadn’t given that permission.

“When you use these systems for real work, the permissions matter a lot,” he writes. “Both companies let you decide whether the AI must check with you before acting, such as before sending an email, buying something, or changing a file. Until you trust the … leave everything to ask for approval first, which is the default.”

One thing that stuck out to me about the post is how Mollick makes the subject of AI agent task delegation real, so that as the reader, you start to think about what it’s like to “give an AI a computer.” It makes intuitive sense that an AI can use a computer, because within the world of the hardware, and the visual display screen, everything is digital, and there’s no physical work needed. But on the other side (and I’m reminded of the old programmer’s adage: PEBCAK) it’s uncanny to have an AI agent be able to trigger commands the way a human would with a mouse, or a keyboard.

“You can start a long job from your phone, close the app, and check the results later,” Mollick writes. “Delegating a few hours of work while standing in line for coffee is a liberating experience.”

Reading that made the whole thing seem real to me. And there’s this:

“Any time I have a problem with my computer, Codex just fixes it, which feels like having a tiny goblin IT department hiding in my computer.”

(Again, see his cybersecurity caveat…)

“If you turn on the ‘computer use’ option in Code or Codex, the AI can literally take over your mouse, browser, and computer.”

That, I thought, was sort of eerie.

Mollick also told us to look for his new book coming out in the fall. Interestingly, in his AI guide post, he talks about using an agent to help edit the manuscript.

“It has been through rounds of professional editing and proofreading, but I gave GPT-5.6 Sol in Codex the full PDF anyway and asked it to check it all over,” Mollick writes. “The AI worked for 30 minutes, chased down 195 references, and gave me pages of notes that would have taken a team of researchers many hours.”

You can see the appeal to efficiency here. We really are in an amazing time, and while we face particular challenges, we also see opportunity on the horizon. Mollick mentions Google Omni, the only video editor of its kind, and other tools that will automate so much of traditional workflows that nothing is really ever going to look the same again.

How do we then navigate such a world?

Here’s Mollick’s ending advice to the hapless user.

“My practical advice remains pretty similar: pick Claude or ChatGPT, pay the $20, and give an agent a real task from your real life. Then look carefully at what comes back, and, rather than just accepting or rejecting the results, ask for changes, just as you would ask a real person. See if you can accomplish your goals, even if you failed at first. You will learn more about what AI means for you from that one experiment than from any guide, including this one.”

In other words, give it a try. And look for that book in October.

The world is waking up to the power of AI. Let’s see what happens through the end of 2026.