How to Solve Problems in Two Short Stories

Two of the finest short stories I know sit at opposite ends of the universe, literally and figuratively. One is cosmic. It spans trillions of years, outlasts the stars, and ends with a computer becoming God. The other is small and quietly vicious, and it unfolds almost entirely inside the head of one anxious man at his filing job. I have long cited them as my two favorites of the genre, but until very recently failed to see that they actually tackle the same topic. Read together, they turn out to be arguing with each other about the question political strategists are circling harder now more than ever: how do hard problems get solved?

Before going further (and even if you don't come back to this post, just for sheer enjoyment), read both. James Thurber's “The Catbird Seat” is short, wicked, and the better part of its pleasure is being ambushed by it. Then, Isaac Asimov's “The Last Question,” which Asimov rightly considered the best thing he ever wrote. What I share next assumes that you have.

The two stories disagree, completely and instructively, about how a problem yields.

Asimov's answer is computational. In “The Last Question,” humanity asks a supercomputer, generation after generation, whether the heat death of the universe can be reversed. The machine answers, each time, that it has insufficient data. It solves the problem only after cosmic time has elapsed and it has absorbed all of human consciousness into itself — only after enough information has finally accumulated. The solution, when it arrives, is an act of creation: “Let there be light.” Scale the processing high enough and run it long enough, Asimov insists, and the problem gives way. To Asimov, in other words, a powerful enough computer is not merely god-like but God: the creator that speaks the next universe into being, and leaves the unsettling possibility that we are the handiwork of a computer that came before us.

Thurber's answer is imaginative. Mr. Martin, a meek filing clerk, is being slowly destroyed by an abrasive colleague, Mrs. Ulgine Barrows. He has no leverage and no data: only an exact, almost intimate understanding of how Mrs. Barrows perceives the world, an audacious imagination, and a single lateral, absurd, perfect maneuver that turns her own perceptions against her. No model produces that move. It demanded one particular mind, with its particular anxieties and particular gifts. That is not a processing advantage but something closer to an existential one: Martin prevails because he knows what it is to be afraid of losing something. If Asimov's definition of the divine is a machine vast enough to remake the universe, Thurber's is the spark of a single human imagination, trained on the most human problem there is: how to be rid of someone you cannot stand.

This essay was not in my plans. I hadn't sat down to write anything at all. On a sunny Saturday afternoon I had simply started talking with Claude, pulling at the threads of a few things I'd loosened by reading Michael Pollan's A World Appears (his new book on the science of consciousness), for no better reason than that not many people in my life want to chase that kind of question across a perfectly good summer day. Somewhere in the conversation the two stories clicked together, and it occurred to me they are the same argument in two costumes: computation against the conscious experience of feeling. That recognition — that they belonged together — was itself a move in the mode of Thurber, arrived at in dialogue with a system built in the mode of Asimov, which is its own small joke at the expense of the thesis. Only then did I think that I should maybe write any of this down.

These are the two modes of problem-solving, and politics runs on both.

The Asimov mode is strategy: the data, the modeling, the targeting, and above all the discipline to follow what the numbers are telling you when every instinct argues otherwise. The discipline is the hard part, because the computation is most valuable precisely when it contradicts the story a candidate wants to tell about themselves.

Maryland's 2026 Democratic primaries made the point cleanly. The polling was consistent and unambiguous: primary voters cared, above all, about standing up to Trump and fighting back against ICE's cruel incursions into immigrant communities. Many candidates balked. They were certain — as candidates always are — that what voters truly wanted was their personal story and their stance on local issues. The data said something narrower and far less flattering to that instinct. Meanwhile, the independent expenditure supporting Sarah David for Baltimore County State's Attorney trusted the numbers without flinching: its mail and its television ads spoke, almost exclusively, to her opponent's partnership with ICE. David won by roughly 25 points against a five-term incumbent. The machine had sufficient data. Campaigns that overrode it with intuition did not see margins nearly so wide.

The Thurber mode is tactical, and it governs the day-to-day. It is the read of a room, the understanding that hate is often a more motivating emotion than love for a human donor or voter, the lateral move in a conversation that no targeting model would ever generate. It is the decision a candidate or staffer makes in the moment because they know, the way Mr. Martin knew Mrs. Barrows, exactly how the person across the table experiences the world.

This is also why, nearly twenty years in, I still staff call time myself — those unglamorous hours when a candidate sits in a small room and dials donors, one after another, to ask for money. Most consultants who run a firm handed that off long ago. I never have, because call time is where people are at their most human. Asking another person for money is stressful and exposing, and the stress strips away the performance. In that room I learn who a candidate actually is: what they are proud of, what frightens them, what they need to get better at.

One of my most capable clients was already a gifted fundraiser when I caught a habit in call time: she would name the amount and then keep talking through the ask and therefore softening it by reassuring the donor that less or another time was fine, no pressure. The whole game is to say the number and then stop, letting the silence belong to the donor. I told her, bluntly, that this one habit was the only thing standing between her and the totals other clients were posting. A pro, she corrected it at once. Her reporting numbers have only climbed since.

The detection was the easy part. A model might well have flagged the habit from a transcript. What it could not have done is the part that actually mattered: the trust that let me say it to her that bluntly, and the relationship that let her hear it without flinching and correct it with the next call. And the deeper version of the work is smaller still, and harder to admit. It is knowing which single sentence will move a particular donor, and which will send them reaching for a donation to your opponent instead. It is knowing whether a given candidate will get into the chair faster for the promise of a record haul or the fear of being outraised and humiliated. It is reading each specific person or situation closely enough to know exactly which lever to pull. You can call it manipulation, or you can call it craft. Thurber never bothered to name it. He simply made Mr. Martin's gambit a small masterpiece of turning a person's own mind against them, and made us love him as the hero of the story anyway.

The best strategists keep both methods at their fingertips, both alive in their heads as possible routes, and know, reliably, which one a given moment demands. But the balance between them is shifting, and shifting fast.

When I pressed Claude on exactly this, it was, to its credit, candid about its own limits. It could generate what looked like lateral thinking and model other minds well enough to be useful. Whether there was anything it was like to be the thing doing so — whether it possessed any version of Mr. Martin's quietly desperate ingenuity, or only a very sophisticated pattern of outputs that resembled it — it openly admitted it could not tell me. That gap is the entire question.

Which returns us to Asimov, and to the stakes beneath all of it. The live question for our field is whether the computational mode eventually absorbs the imaginative one: whether, as the processing power trained on campaigns keeps compounding, it crosses some threshold where it can also do the situated, in-the-room, irreducibly human thing. “The Last Question” suggests the machine gets there in the end, but only after it has taken in everything human first. “The Catbird Seat” suggests some problems will always require one specific, irreplaceable person who knows what it is to be afraid.

The honest objection lives inside my own argument. The most staggering win of my cycle was an Asimov win, where the data was plain to all of us. And the reads I just called irreducibly human, knowing which sentence moves a donor or what a candidate needs to hear, are exactly what the models get better at every month. None of it is sacred, and anyone who says computation has hit its ceiling there isn't paying attention.

None of this makes the Asimov mode the junior partner; the win I just described is proof of how much it matters. If I find myself dwelling on Thurber, it is not because it outranks Asimov. It is because the computational mode already has every advantage and every evangelist, while the Thurber mode is the one we are at risk of forgetting how to do as the machines get better at the other. It is also the harder of the two to automate. Computation is strongest where the past predicts the present, and the problems most worth solving are often the ones where it doesn't, where there is no comparable precedent to extrapolate from. The Catbird Seat move is the move with no training data: the novel situation, the sample size of one, the person who refuses to behave like the average of ten thousand others. A model interpolates from what has already happened; Mr. Martin invented something that never had. And even a flawless read still has to be delivered by someone a candidate trusts at decision deadlines or in small rooms stacked high with call sheets, which is a relationship, not an output. 

For what it's worth, it is where Claude landed too: the work ahead — persuading actual people, reading actual rooms, finding the move no one else would have found — is exactly the work that computation has not learned to do, and may never do alone. That is either very good news for the people in our profession or a serious challenge to whatever comes next. Possibly it is both.

So, the next time you are staring down something the data cannot quite crack, the question worth asking is the one Thurber answered for himself eighty years ago: what is your Catbird Seat move?