Why Can’t A.I. Manage My E-Mails?
Chatbots can pass the Turing test—but they can’t yet handle an office worker’s inbox.
For this week’s Open Questions column, Cal Newport is filling in for Joshua Rothman.
One morning last month, I decided to try artificial intelligence on a dire problem: my inbox. In the past twenty years, the e-mail address I use for writing projects has been discovered by a staggering number of P.R. firms, scammers, and strangers with eccentric requests. On this particular day, I had eight hundred and twenty-nine messages. (Some knowledge workers might consider that pretty good, but for me it caused plenty of stress.) Of the fifty most recent e-mails, the majority were dreck, but about eight were of actual interest, suggesting a hit rate of sixteen per cent—just enough that I had to worry about missing something important.
Cora is one of many web-based applications that interact directly with users’ Gmail accounts—reading, tagging, and archiving messages on the user’s behalf. “Give Cora your inbox,” the app’s website says. “Take back your life.” Cora aims to use A.I. to protect users from any messages that don’t genuinely require a response. The rest are archived and summarized in a nicely formatted, twice-daily briefing. According to Cora’s creators, ninety per cent of our e-mails “don’t require a response. So then why do we have to read them one by one in the order they came in?”
During the setup process, Cora read my two hundred latest e-mails to learn who I am, which will help it identify messages that matter to me. It deduced that I work for Georgetown University and am an author (both correct), and that my work focusses on “digital minimalism & productivity research.” (I’m also a technology critic and a digital ethicist, which it didn’t pick up on.) I entered my credit-card information; the service costs twenty-five dollars a month. “Cora is working on building your next brief,” the app told me. “It will send you an email when it’s ready.” To help Cora start with a fresh slate, I archived my eight hundred and twenty-nine unanswered messages—to anyone who never heard back from me, I am sorry—and resolved to check back the next morning.
My experiment was about more than exorcising the demons of my inbox. In my years studying and writing about technology and work, I’ve come to believe that the seemingly humble task of checking e-mail—that unremarkable, quotidian backbeat to which digital office culture marches—is something more profound. In 1950, Alan Turing argued in a seminal paper that the question “Can machines think?” can be answered with a so-called imitation game, in which a computer tries to trick an interrogator into believing it’s human. If the machine succeeds, Turing argued, we can consider it to be truly intelligent. Seventy-five years later, the fluency of chatbots makes the original imitation game seem less formidable. Yet no machine has yet conquered the inbox game. When you look closer at what actually goes into this Sisyphean chore, an intriguing thought emerges: What if solving e-mail is the Turing test we need now?
Typically, engaging with your e-mail involves sorting messages into different layers of complexity and importance. The shallowest layer includes spam, promotional e-mails, and long-forgotten newsletter subscriptions that you can confidently delete. The next layer contains messages that require your attention but which can be satisfied with a simple reply: “Got it!” “Thanks.” “It’s at 4:00. See you then!” These e-mails can produce a pleasurable sense of productivity with minimal mental effort. But, until they’re answered, they can also create a creeping anxiety, as if a crowd of correspondents is impatiently awaiting your attention.
The deepest layer is made up of messages that are quick to read but which require significant thought. Consider this hypothetical e-mail:
“Hi Cal! I’m John Doe’s brother. I’ve been reading your books for years, and as such was so excited to find out just recently that he knows you! Anyway, I’m working on a new tech startup that uses principles from your book “Deep Work” to rethink your digital calendar. I’d love to grab coffee with you when I’m in town next week. Any particular days work best?”
Before I can respond, I need to assess the social and practical implications of the request. Is John Doe sufficiently important to me that I need to do his brother a favor? Is there a chance his startup will be interesting to me—the sort of thing I’ll be glad to have helped shape? If I decide to meet him, when and where should I suggest? If John’s brother has a higher status than I do—maybe he’s a well-known entrepreneur—I may need to defer to his busy schedule, but if he’s a young man seeking mentorship, he can work around mine.
In the end, the response I send back may consist of just a handful of words, but it will be the result of a nuanced sequence of rapid-fire discernments and decisions. This single activity arguably bundles many of the cognitive skills—filtering, decoding, planning, analysis—required to succeed in virtually any kind of knowledge work.
The morning after I activated Cora, I logged into my Gmail account with some trepidation. Typically, I might see a chaotic jumble of thirty or forty messages, but now I found only five requiring my attention, one of which was Cora’s briefing. Boring as the moment might seem, I’m not sure that A.I. has ever made me more excited than I was then.
The briefing revealed that the app had archived twenty-nine e-mails on my behalf, and its decisions seemed pretty good—a quick scan suggested that all but two were indeed deletable. (I could read and reply to the wrongly filtered messages—both of which were notes from my readers—directly from the briefing web page.) Among the handful of messages Cora left in my inbox, the app identified several as layer two and offered a rough draft of possible replies. In response to a reader who wanted feedback on her new website, Cora proposed, “Thanks for reaching out, and I appreciate the kind words about my work. Unfortunately, I’m not able to take on website reviews right now.” My reaction to such requests is perhaps harsher—I find it better not to respond at all—but I appreciated the app’s attempt.
What Cora didn’t try to tackle were the layer-three messages that required more involved thinking and action. As a test, I sent myself the John Doe e-mail from another address; Cora left it untouched in my inbox for me to handle. Indeed, none of the other A.I.-powered e-mail tools I reviewed, including Superhuman, Microsoft Copilot for Outlook, and SaneBox, attempt to respond to these types of non-trivial e-mails. One assumes that they’re not close enough to winning the inbox game to risk trying.
So why can’t A.I.s auto-reply to thornier communications? A key obstacle is the way they’re built. Kieran Klaassen, the general manager and lead developer of Cora, told me that the app can be divided into two components: a standard-control program, which accesses one’s inbox and manipulates messages, and a collection of commercial large language models that the program can consult when more complicated analysis is required. For example, when Cora needs to decide whether a given message is important for a particular user, the control program creates a text-based prompt and submits it to an L.L.M. “The intelligence lives entirely in the language model,” Klaassen said. This means that an A.I. tool like Cora is not an inscrutable black box learning and evolving new abilities—it’s more like a custom software layer that’s good at using ChatGPT.
This division of labor offers some clear advantages. Cora can make use of cutting-edge language models without spending vast amounts of money to build one itself. It also allows flexibility. To change how Cora filters messages, you don’t have to update its programming but instead modify the prompts that it sends to the third-party language model. In my Cora settings, I can read the exact instructions the control program sends to Google’s Gemini Flash model when asking it to assess a message:
Emails that require the user’s personal review must stay in the inbox; examples: reader replies, media/speaking opportunities, book-related collaborations, beta-reader requests, security/account changes, and technical notifications.
If I decided that “technical notifications” were no longer important, I could delete that example; if I decided that I wanted to read positive e-mail newsletters about the Washington Nationals baseball team, I could add a few words instructing Cora to send them through. (Unfortunately, at the moment, this instruction might not get much use.) “You can actually teach it new behaviors through conversation rather than having to change code,” Klaassen said.
But a dependence on commercial L.L.M.s also presents an obstacle: they weren’t trained on information specific to me, my job, or my professional preferences. For Cora to respond to John Doe’s brother, it would have to figure out all of the relevant information—who I am, who I know, how I think about these relationships, what I’m interested in, my preferences for meeting locations and times, my upcoming availability. Packing all of that into a prompt for the model—a prerequisite for getting a satisfactory reply—would be an astoundingly complex challenge.
In a 1966 book, “The Tacit Dimension,” the polymath Michael Polanyi argued that our decisions in life and work depend heavily on unstated context and implicit assumptions, which are unique to our own experiences. What Polanyi famously dubbed “tacit knowledge” is subtler and harder to articulate than we realize. “I shall reconsider human knowledge by starting from the fact that we can know more than we can tell,” he wrote. This is precisely why current A.I.-powered e-mail tools cannot reliably respond to all of our messages. Even though language models are fantastically knowledgeable about many things, they’re ignorant of the vast quantities of tacit knowledge woven into our lives and offices—preventing any commercial model from reliably figuring out whether to say “yes” to that coffee invitation. It doesn’t matter how smart we make our machines if we cannot describe to them exactly what we want.
It’s not necessarily bad news that A.I. tools are unlikely to automate e-mail anytime soon. A machine capable of consistently winning the inbox game is a machine that might put a lot of knowledge workers out of a job. But even given their current constraints, e-mail apps might still evolve past Cora and its ilk. Srinivas Rao, an independent A.I. developer, showed me a prototype of OrchestrateInbox, a new e-mail assistant that uses commercial language-model technology to eliminate the inbox altogether, offering the user an “intelligence briefing” about the content of their messages.
In the demo I saw, the briefing began with an “executive summary,” which noted (among other things) that Rao had “received multiple pitches from founders, publicists, and strategic advisors.” This was followed by a numbered list of individuals who needed a reply, accompanied by a one-sentence description of “What they want.” Someone named Seta Z., for example, was “offering a book for possible podcast coverage or review.” Instead of manipulating individual messages, users are supposed to interact with the tool using natural language, as one would with a chatbot. Perhaps I’d ask for more information on the book—and then, if I’m not interested, I could tell the tool to decline on my behalf. All of this transpires in something like a chat interface; the user never has to see the underlying messages.
Whether or not Rao’s particular vision spreads, there’s a bigger lesson here. Although A.I. e-mail tools will probably remain constrained by the tacit-knowledge problem, they can still have a profound impact on our relationship with a fundamental communication technology. Dan Shipper, the founder and C.E.O. of the company that produced Cora, told me that the telling question for our current moment is not “Do I do e-mail anymore?” but, rather, “How different does my e-mail look than it used to?” Recently, I returned from a four-day trip and opened my Cora-managed inbox. I found only twenty-four new e-mails waiting for my attention, every one of them relevant. I was still thrilled by this novel cleanliness. Soon, a new thought, tinged with some unease, crept in: This is great—but how could we make it better? I’m impatient for what comes next. ♦
New Yorker Favorites
In the weeks before John Wayne Gacy’s scheduled execution, he was far from reconciled to his fate.
What HBO’s “Chernobyl” got right, and what it got terribly wrong.
Why does the Bible end that way?
A new era of strength competitions is testing the limits of the human body.
How an unemployed blogger confirmed that Syria had used chemical weapons.
An essay by Toni Morrison: “The Work You Do, the Person You Are.”
Sign up for our daily newsletter to receive the best stories from The New Yorker.
Cal Newport is a contributing writer for The New Yorker and a professor of computer science at Georgetown University.
More:
Organization
Efficiency
Artificial Intelligence
Read More
What Comes After Starvation in Gaza?
The Lede
What Comes After Starvation in Gaza?
For the severely malnourished, simply starting to eat normal meals again can cause sickness—even death.
By Clayton Dalton
What Was the American Revolution For?
American Chronicles
What Was the American Revolution For?
Amid plans to mark the nation’s semiquincentennial, many are asking whether or not the people really do rule, and whether the law is still king.
By Jill Lepore
How Far Can Donald Trump Take Emergency Power?
The Lede
How Far Can Donald Trump Take Emergency Power?
In the Supreme Court’s tariffs case, the conservative Justices will weigh two conflicting impulses regarding Presidential authority.
By Jeannie Suk Gersen
The Allure&-and the Policing&-of Subway Surfing
Critic’s Notebook
The Allure—and the Policing—of Subway Surfing
Mayor Eric Adams’s administration has wrapped an expansion of invasive surveillance in the apolitical packaging of saving teen-agers from their addled selves.
By Doreen St. Félix
Is Gambling Really Threatening the Integrity of Sports?
Fault Lines
Is Gambling Really Threatening the Integrity of Sports?
After a recent N.B.A. scandal, more writers and pundits have come out against legalized betting. But the case that they’re making is weaker than it appears.
By Jay Caspian Kang
What Did Men Do to Deserve This?
The Weekend Essay
What Did Men Do to Deserve This?
Changes in the economy and in the culture seem to have hit them hard. Scott Galloway believes they need an “aspirational vision of masculinity.”
By Jessica Winter
Rebels Post Videos of Mass Killings in Darfur as the World Watches
The Lede
Rebels Post Videos of Mass Killings in Darfur as the World Watches
After the fall of El Fasher to the R.S.F., observers fear for the next target in the war.
By Nicolas Niarchos
Rian Johnson Is an Agatha Christie for the Netflix Age
Onward and Upward with the Arts
Rian Johnson Is an Agatha Christie for the Netflix Age
The director revived the cozy mystery with “Knives Out.” In a new sequel, can he find his way to the end of the maze?
By Anna Russell
Solvej Balle’s Novels Rewire the Time Loop
Books
Solvej Balle’s Novels Rewire the Time Loop
Most stories in the genre build to a moment of escape. “On the Calculation of Volume” imagines a woman making a life inside an infinitely repeating November 18th.
By Katy Waldman
The Hidden Devastation of Hurricanes
The Lede
The Hidden Devastation of Hurricanes
Their health effects extend far beyond official death tolls.
By Clayton Dalton
The Real Target of Trump’s War on Drug Boats
The Lede
The Real Target of Trump’s War on Drug Boats
The Administration has blown up seven vessels in the Caribbean in recent weeks, but the President has been pushing for more dramatic military action in Latin America since his first term.
By Jonathan Blitzer
How the Supreme Court Defines Liberty
Under Review
How the Supreme Court Defines Liberty
Recent memoirs by the Justices reveal how a new vision of restraint has led to radical outcomes.
By Ruth Marcus
The New Yorker
Sections
NewsBooks & CultureFiction & PoetryHumor & CartoonsMagazineCrosswordVideoPodcasts100th AnniversaryGoings On
More
Manage AccountShop The New YorkerBuy Covers and CartoonsCondé Nast StoreDigital AccessNewslettersJigsaw PuzzleRSSSite Map
AboutCareersContactF.A.Q.Media KitPressAccessibility HelpUser AgreementPrivacy PolicyYour California Privacy Rights
© 2025 Condé Nast. All rights reserved. The New Yorker may earn a portion of sales from products that are purchased through our site as part of our Affiliate Partnerships with retailers. The material on this site may not be reproduced, distributed, transmitted, cached or otherwise used, except with the prior written permission of Condé Nast. Ad Choices
Privacy Information