Does AI Make You More Productive? Or Just Busier?

Cal Newport, my favorite writer on technology and productivity, has an excellent column asking “Why Hasn’t AI Made Work Easier?” He notes that the early returns on AI use suggest that many workers are not actually getting more done with AI. They are just getting busier and spending less focused, uninterrupted time, or what Newport calls “deep work.”

Whatever difference AI will ultimately make in our current jobs, we have to be constantly vigilant against the human temptation to fiddle around. We have an endless capacity for messing with stuff, instead of actually working on our most important long-term project and/or the work that is adding value for our employer.

AI poses a major threat to the integrity of all writing, from one-page middle school reports to books with major publishers, all of which are increasingly synthesized by AI instead of being created by humans. Ethically, AI is in the “Wild West” phase of general lawlessness and lack of boundaries, especially for writers.

A less obvious problem for our AI use, however, is that AI could ultimately become just one more time-waster that creates a bunch of “stuff” but few products of enduring value. (Cal Newport compares it to email, which unquestionably made us busier, but arguably made us less productive.)

As I wrote in my previous newsletter, however, there are certain functions of AI that can give a major boost to humanities research. In my work, the primary benefit I see so far is transcribing historical manuscripts.

The problem with many history topics since around 1700 is not a lack of sources, but the fact that there are massive amounts of sources functionally locked away in hand-written manuscripts that are often difficult to access and read. For major figures like Jonathan Edwards or Thomas Jefferson, this is not so much of a problem, because scholars have transcribed their papers and published them in edited collections. But even for as well-known a figure as the evangelist George Whitefield, we still don’t have a scholarly edition of his papers.

A fair number of his papers are digitized. I know this, because when I search on “George Whitefield papers” many of the initial results are papers at archives like Emory and SMU that digitized their Whitefield papers when I paid them to do so for my 2014 biography! (You’re welcome.)

But digitized is not the same thing as transcribed. Reading 18th or 19th century handwriting is laborious, even if you don’t have to travel to London or other cities to physically visit the archives. AI can already scan and transcribe historical handwriting much more efficiently than I could manually.

AI uses a ton of electricity, but it doesn’t get tired, and there is no tired for a human researcher like reading-in-an-archive tired! Optimized AI is also arguably better at identifying handwritten characters (letters) than the human eye is. And it will only get better in the coming years.

Just to remind you that this is a “developing story,” I have switched to Google Gemini from Copilot to generate transcriptions of documents for my biography of Adoniram Judson. Gemini is significantly better than Copilot, even in its free version.

When I submit a pdf file with a single document in it, Gemini can usually transcribe it with about 95% accuracy, except when the manuscript is damaged or really dark (in which case I can’t read it either). Gemini struggles to manage pdfs containing many documents, but I can usually get a good transcription of each if I spend enough time explaining which document I want (it often helps to start a new dialogue for each one).

I don’t see ethical problems with having AI generate transcriptions, because you still must read the transcriptions yourself. Now they’re just a lot easier to read. AI allows you to move quickly to the work only a human author can/should do, namely interpreting the source and using it in a book project.

With AI’s assistance, I’m reading sources that I likely wouldn’t have read in previous book projects, because the manual labor involved in reading thousands of pages of historic handwriting was so demanding and inefficient that I would just run out of energy and time at the archive.

So if we become a regular AI user at work, we have to constantly ask ourselves, is this actually saving me time? Or am I just procrastinating and generating more busyness? Worse, am I violating ethical boundaries by presenting AI-generated material as my own writing?

My initial conclusion is that in this narrowly-defined task, AI transcribing historical documents is a major improvement on previous manual processes. It will allow me to do more comprehensive research in the long run.

My latest journal article is “The Name of Father Cummings Will Be Long Remembered”: Cummings v. Missouri, Religious Liberty, and the Test Oath Controversy,” Journal of Church and State [subscription].

Pre-order my book The Death of Religion?: Nones, Others, and the Flourishing of Faith, co-authored with Byron Johnson (B&H Academic, Sept. 2026).