Your AI is only as good as what you feed it. Feed it a transcript with your product names, acronyms, and numbers spelled wrong, and you spend the afternoon hand-fixing the output.

Wispr Flow Notetaker uses your dictionary and calendar, so product names, acronyms, numbers, and uncommon names come out spelled right. It pops up when your meeting starts and captures Zoom, Google Meet, Teams, and Slack huddles with one click, no bot joining the call.

Then your meetings show up inside Claude or ChatGPT through the built-in connector. No copy paste. Try Notetaker free on Mac and Windows.

Try Notetaker Free

MIT's Eric So has written a book about the "collision" between AI, the human brain, and society. Meanwhile, McKinsey now has job applicants interview with a simulated AI client. Both ask the same question from different directions: when a machine does the thinking-adjacent work, which part of the thinking is still yours?

The collision, and what we quietly hand over

MIT News has published a Q&A with Eric So, an MIT professor, about his book on how AI, our brains, and our institutions are colliding. Professor So has chosen the word  "collision" quite deliberately. A tool that you pick up and put down doesn't collide with anything, but a ubiquitous force, which reshapes how you work, learn, and decide, does.  I also think it’s worth noting that AI has recalibrated what any single person might accomplish alone.  Professor So's book is one of the more complete attempts yet to describe what leaning on AI does to our thinking as humans.

Some adjacent evidence shows why the question is urgent. In 2025, researchers at the MIT Media Lab led by Nataliya Kosmyna ran an EEG study with 54 participants writing essays over several months. They were split into three groups: one used an LLM, one used a search engine, and one used nothing but their own mind. The LLM group showed the weakest neural connectivity, and its members had more trouble quoting their own essays afterward. The sample was small and the work was released as a preprint, so it’s an indication rather than a confirmed verdict. Also in 2025, Microsoft and Carnegie Mellon researchers surveyed 319 knowledge workers and found that greater confidence in AI was associated with lower critical thinking effort, while greater confidence in oneself went with more critical thinking effort.

Taken together, these findings sharpen So's question. Calling it "offloading" makes it sound harmless, like handing arithmetic to a calculator. But AI is being used much more extensively in our daily lives than a calculator.  AI is being used as the first draft, the summary, the outline, the list of options.  Those tasks look clerical, but one could argue that the blank page is where much of our generative thought begins.  Brainstorming, outlining, and writing a first draft is where you sketch out and stress-test your thinking; summarizing is how you learn what matters. If you skip these steps, you still get the output, but you may not get the cognitive “friction” and understanding that used to come with it. 

Professor So calls this offloading “cognitive atrophy;” I believe that as we become more reliant on AI (especially on AI agents, who become our digital co-workers), the effects will become more and more pronounced.

Some suggestions:

Keep the parts that form your judgment; like deciding what the question is, forming a first hypothesis, and choosing what counts as “good.” Hand off the parts that only reflect judgments you’ve already made, like reformatting, searching, and boilerplate.  For example: You might jot down your own best guess before you ask the AI, then check whether you can explain the final answer without it. If you can't, the tool has given you a conclusion without understanding what’s going on “under the hood.”  The brain remains the original generative machine, and So's book argues that it needs exercise to stay that way.

Also on the radar

McKinsey interviews you alongside a fake client.

The Verge reports that McKinsey is having job applicants spar with a simulated AI client as part of a new interview process. Whatever the scoring rubric looks like, the format rewards candidates who can steer a conversation with a machine while still treating it as a stand-in for a person with real stakes.   This goes beyond simply producing a polished answer.

Epic Games ships a launcher that it calls “vibe-coded.”

The Verge reports that Epic Games' new launcher was "vibe-coded," with the team cheerfully conceding that “programming is hard.” When intent replaces implementation, the person who wrote the prompt still owns every bug, so review and testing become the job rather than an afterthought.  On a recent episode of Lenny’s Podcast, guest Josh Woodward of Google Labs, shared that he believes that ‘gaming’ is one of the most exciting areas for opportunity in AI right now. The preponderance of vibe-coded software is not extraordinary now, with many leaders of frontier labs sharing that they rarely code “by hand” any more. However, I still believe craft and verification will continue to be crucial.

YouTube labels AI-generated songs as such.

According to The Verge, YouTube Music is adding "Made with AI" labels to AI-generated songs, . A label settles provenance but says nothing about quality or about how much human intention went into forming the track, and listeners have long asked those questions about ghostwriters, samples, and session players.  AI is reshaping the creative practice along with every other human endeavor, with fiction, music and filmmaking particularly implicated.  For all the artists who are vehemently opposed, there are others, like Ben Affleck, whose startup, InterPositive, seeks to leverage AI for filmmaking.  Affleck has said that InterPositive uses AI to automate the expensive, repetitive, and burdensome tasks of post-production.

worth your time.

You might check out "The Extended Mind," the 1998 paper by philosophers Andy Clark and David Chalmers, which is short and easy to find online. It imagines Otto, a man with memory impairment who relies on a notebook, and Inga, who relies on her biological memory, and argues that Otto's notebook is a part of his mind. The paper was written about two decades before the current AI moment, yet it gives you a precise test for what you can hand to a tool and still call your own thinking: whether you reliably trust, consult, and integrate what the tool holds.  Clark and Chalmers discuss the idea of an “active externalism,” whereby objects outside of your physical body act as extensions of your mind.  In their paper, they suggest that such objects could start to be considered part of a complete cognitive system, with the outside items as manifestations or extensions of the your own thoughts.   Give Neuralink enough time, and I think we’ll get there. Whether that’s something we actually want is a question for another time.

The human mind is the original generative machine.