The Loops of August: Coding, AI, YouTube Algorithms, Vacations
We mostly sat out this August. While other people were heading for beaches, islands, mountains, and increasingly complicated travel connections, we stayed home and worked on projects. There was coding to do. There were long conversations with AI assistants. And, for reasons that will become clear, there were an alarming number of dukes.
The disappearing and reappearing of Italy is the obvious one, an annual cycle so regular that you can practically set a calendar by the appearance of handwritten chiuso per ferie signs. But once we started looking, other loops appeared around us: bursts of obsessive coding followed by months of silence; conversations with AI assistants that circle the same problem without quite landing on it; and a YouTube algorithm that has decided we have an inexhaustible appetite for overlooked stepdaughters marrying scarred members of the British aristocracy.
These are not really the same kind of loop. Some loops we create ourselves. Some loops society creates for us. Some loops we get stuck in. And some, apparently, come with a ducal seal.
The Coding Loop: Three Weeks On, Three Months Off
The evidence is sitting in our GitHub contribution charts.
We used to imagine that we worked on our Scrapbook project continuously, but slowly realized we actually didn’t, and the little green squares of the contribution chart tell that story. There will be three or four weeks of dense activity, commits stacked on commits, followed by long stretches of almost nothing. Green squares appear in a row showing activity. Then white, the digital equivalent of abandonment.
GitHub contributions going in cycles.
Scrapbook is not a toy app we occasionally tinker with. It is a long-running project built and maintained by the two of us, with roughly 80,000 lines of code, about 25,000 of them executable. It lives in the cloud, uses a range of technologies, gets fixed, acquires new features, and often just needs attention because something it depends on has changed. We have been working on it for years and, importantly, we don't want to finish it.
There is language for the pattern visible in our GitHub charts. Researchers studying human activity call one aspect of it burstiness: concentrated periods of activity separated by long periods of relative quiet. Our weeks inside those bursts can also feel a lot like hyperfocus, the state of becoming so absorbed in an activity that time and everything around it recedes.
And coding is particularly good at keeping a burst going. One fix suggests another. The new feature almost works. Then it works. Now that it works, we can finally clean up the thing beside it that has annoyed us for two years. While we are in the code, we notice something else. At some point it is midnight, and a perfectly reasonable Saturday has disappeared into a C# and JavaScript methods that should have taken twenty minutes.
Then, just as mysteriously, coding stops.
For months we might barely touch the code. Our Scrapbook project continues to run. Other interests take over. The green squares disappear.
A few years ago, we noticed this pattern and wondered whether it was something we should worry about. Now we are less sure. This is coding for pleasure. There is no product manager waiting for a sprint review, and no investor is asking about features. Perhaps the periods of inactivity are part of what makes the periods of intensity possible.
The coding loop may be one of the loops we don't need to escape. We just disappear from it for a while, until something pulls us back in again.
The August Loop: Everybody Leaves, Everybody Comes Back
Our little two-person coding cycle takes place inside a much larger and considerably more organized loop: August in Italy.
Starting in late July, the country begins its progressive emptying-out. Shop windows display hand scrawled chiuso per ferie messages. Offices become harder to reach. Automated out of office messages in email start to become the norm. Conversations increasingly revolve around when you are leaving, where you are going, and when you are coming back. Then comes Ferragosto, August 15, the symbolic center of a summer vacation season treated with a seriousness we still find impressive after living here for years.
We have written about this before: Bergamo – Street Sign Language Lesson X – Ferragosto!, Abbronzatissima: Notes on the Allure of the Suntan in Italy, and most recently Escape to Isolatos: Notes on Curated Remoteness. In the last post, we were interested in the slightly paradoxical Italian search for escape: everybody getting away from it all at approximately the same time.
But escape is only half the loop.
Toward the end of August, people begin appearing again. The shutters go up. Familiar faces return, various shades darker. There are stories about beaches and mountains and islands, mixed with complaints about traffic, ferries, delayed flights, excessive heat, and all the other inconveniences encountered while getting away to relax.
Then comes that brief September feeling that everything is beginning again. Schools reopen. Offices refill. Out of office messages just become the normal delayed responses. Plans are made. People seem faintly renewed. For a few weeks, September has the same quality of January without New Year’s resolutions.
The return completes the loop. Unlike our coding bursts, this one is based on the calendar and is collective. Everyone knows roughly when it is going to happen and participates willingly.
There is something slightly funny about the annual acceptance of it all. Italy empties out in August, businesses close because everyone is away, people complain that nothing is open and nothing can get done because everyone is away, and then the whole system starts up again. Next summer, we will do it all over.
The AI Loop: Down the Rabbit Hole
During the hotter days of August, we spent quite a bit of time indoors with air conditioning and AI assistants, sometimes while coding and sometimes asking about whatever question had wandered into our heads that day.
This introduced another kind of loop.
The transaction often starts innocently enough. Take a coding task. We ask an assistant to solve something. It misunderstands one small but important part of the problem. We correct it. The assistant apologizes, usually enthusiastically: “You’re absolutely right!” It then produces another answer based on substantially the same misunderstanding. We correct it again.
Twenty minutes later, we are no longer solving the original problem. We are trying to make the assistant understand the conversation we have already had. We see our AI credits are going to zero, and we bump our spending budgets because the problem needs to be solved.
Part of what keeps us stuck is sunk cost. We have already supplied the background, pasted the code, explained what we tried, and corrected assumptions. Surely one more prompt will do it. Meanwhile, the conversation itself is getting longer, accumulating failed approaches and explanations that may now be part of the problem. We find ourselves not so politely telling the assistant that we are burning our credits while going around in circles. Response: empathetic with yet another attempt at a fix.
The problem is that AI answers are often very plausible. A wrong answer doesn't necessarily (immediately) look wrong. It can arrive neatly formatted, confidently explained, perhaps with five helpful headings. This makes it surprisingly easy to follow an assistant a considerable distance down the wrong road before realizing that neither of you remembers where you intended to go.
We used to jokingly call some of this being “gaslit by the AI,” but that isn't really what is happening. The assistant isn't trying to convince us of an alternate reality. More often, we and it have diverged on what problem is being solved. It thinks we asked one thing; we thought we asked another. Sometimes the fault is ours because we were imprecise, withheld context we didn't realize mattered, or assumed the assistant understood something that existed only in our heads.
We are slowly getting better at recognizing the warning signs.
Sometimes the answer is to narrow the question. Sometimes it is to start a fresh conversation rather than keep dragging the accumulated baggage forward. And sometimes the best solution is simply to leave one AI assistant and ask another.
That last technique has become surprisingly useful. Instead of asking the second assistant to solve the whole problem again, we can show it what happened and ask questions: Were we talking past each other? What is the first assistant misunderstanding? Here’s a plan to fix a coding issue the first assistant gave us. What do you think of it?
We have, in other words, begun taking one artificial intelligence aside to complain about another artificial intelligence. It sounds faintly absurd translated into ordinary life, but it works. Different assistants approach the same question differently, notice different assumptions, and get stuck in different places. Increasingly, knowing how to use AI assistants means knowing not only what to ask, but when to stop asking the same one.
The AI loop is therefore unlike our coding loop. Coding can pull us forward because every solved problem exposes another interesting one. The AI loop at times can keep us in place because every attempted solution seems as though it might finally resolve the previous attempted solution but doesn’t.
The trick is noticing when supposed progress has turned into circling.
The Duke Loop: Algorithmic Pulp Fiction
Meanwhile, YouTube has discovered that apparently what I need in August is an overlooked stepdaughter, a cruel stepmother, an estate ledger, a mysterious duke with a scar who perhaps recently returned from fighting somewhere on the Continent.
And apparently YouTube is correct.
I fell into a genre of long-form historical romance stories, many clearly produced with some combination of AI-generated writing, narration, and images. They tend to be set in an elastic version of 19th-century England, somewhere between the Regency and Victorian periods. If our duke has recently returned from war, it may be the Napoleonic Wars in earlier stories or the Crimean War in the later ones. Historical precision is not necessarily the main attraction, just that vague historical events produce our broken duke.
What has been surprising is how quickly I have acquired the vocabulary of this faux-19th-century universe: reticule, ducal seal, signet ring, dowry, marriage contract, estate ledger, bombazine, duke versus earl, and the ton, the fashionable upper reaches of British society. There are carriages, neglected estates, double entries in ledgers, invisible stepdaughters, nasty stepmothers, well-tailored jackets of broadcloth, shy heroines hiding behind potted palms at the ball, marriages of convenience, and an astonishing number of men who speak in a "low scrape."
I didn't set out to learn any of this. The vocabulary arrived through repetition, one little algorithmically encouraged click at a time.
What I'm listening to is a kind of algorithmic pulp fiction. The technology may be new, but the machinery of the story is old. A wronged, overlooked woman is humiliated. Someone powerful finally sees what everyone else has failed to see. The nasty relatives are exposed. Fraud is revealed. Goodness is rewarded. And the duke eventually understands what the viewer (and most often just a listener) understood forty-five minutes earlier.
Then the story ends happily, and I click on another one.
That is the loop. YouTube learns that I will watch a story about an ignored stepdaughter and offers me another ignored stepdaughter. Perhaps this time with two cruel stepsisters and a disputed dowry. I click again, which presumably confirms that what I really need is a third ignored stepdaughter. Somewhere along the way the duke acquires a more prominent scar from temple to jawbone, just to add interest.
Why can't I stop listening?
Partly, I think, because good winning out has a universal appeal and perhaps even more so at this moment and time in my life. These 45-to-90-minute stories are auditory comfort food. We know roughly where they are going, and knowing is part of the pleasure. The attraction isn't suspense so much as promised emotional payback. However badly things go for the heroine in the first forty minutes, the ducal seal is waiting somewhere in the final twenty.
This makes the Duke Loop almost the inverse of the AI-assistant loop. In the AI loop, repetition postpones resolution: one more exchange might finally get us out. In the Duke Loop, repetition guarantees resolution. The same machine starts up again, and that is precisely what we came for.
Loops Within Loops
By the end of August, our four loops were beginning to look less alike.
The vacation loop is a ritual. It repeats because repetition (in Italy) gives shape to the year: leave, disappear, return, begin again.
The coding loop is a cycle of enthusiasm. Pleasure and curiosity arrive in clumps, followed by stretches when our attention goes somewhere else. Nothing is particularly wrong with that.
The AI-assistant loop is different. It is a trap. We stay in it because every next exchange looks as though it might finally resolve the previous one. Plus, the technology is still new, and we are all still learning how to use it efficiently.
And the Duke Loop is a reward machine. It keeps turning because it reliably gives us what we came for: recognition, justice, a happy ending, and apparently another estate ledger.
So perhaps the useful distinction isn't that loops are bad or good. Life contains plenty of repetition, and August itself is proof that repetition is not necessarily something to escape. Rituals, hobbies, seasons, stories, and habits all depend upon coming around again.
The more interesting question is how much control we have over the circle. In some loops, we are in the driver's seat. In others, we are passengers. Sometimes society has set the route long before we arrived as with vacations in Italy. And occasionally an algorithm is quietly deciding where the next turn will be as with our YouTube consumption.
That gives us a rough rule for these particular August loops. The coding loop doesn't need to be fixed. The vacation loop couldn't be fixed even if we wanted to; acceptance is better. The AI loop needs a circuit breaker. The Duke Loop...well, perhaps one more story, just until we find out whether poor quiet Lucy wins over the scarred duke. (Spoiler: she does.)
There is one final loop hiding inside this post. We wrote about AI loops using research produced through conversations with AI assistants, including research about AI-generated romance stories being fed to us by another algorithm. At some point the snake begins eating its own tail. Is the future a form of funhouse recursion? Loops within loops?
And then, while finishing this post, I remembered that I wrote a song called Endless Loop, about getting stuck inside expectations and repeating patterns around us.
Apparently even our observations about loops come in loops.




![[Asteraceae] Carlina acaulis [Asteraceae] Carlina acaulis](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiFyEGN5E9JGykYVH0_Qvht-AwJxFAa0cgCBKGyNCfrdjnFt_RBpLaefrjx6oX9g6TRYl23BP1VhFFXM6S8gfDpBj_dDTWnbSXp7ctPDStNRBtvl6GSIpYEMo22bC9OreMjCbI37lDn8wO4Zl-XvTyGISY76bVb2mM4QvXKbdn0l_mysJY7MI-HJtIpog/w200-h151/%5BAsteraceae%5D%20Carlina%20acaulis%201.jpg)
![[Asteraceae] Carlina acaulis [Asteraceae] Carlina acaulis](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiw7JaUsZB94FnGgV8SSscIHOrh0hTAYwKEiSH7goDNWVPC8UvzVYL6_GmOBklXASQJJiKgHGRcYWnjxr3NWyokdj6wkIVLMKVZ7LR-nl1reWrBRNo-vXj3q-R5ETT-VM1ZjTdJRZm6rxiJzJ4pPsAiBXkn7-Vpokd3YCTcu_eKpGPljZ14mzCj6e5MyA/w200-h151/%5BAsteraceae%5D%20Carlina%20acaulis%202.jpg)
![[Asteraceae] Centaurea sp. [Asteraceae] Centaurea sp.](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgOSCsv4Rx310HRQgdiOqrYe8WM0a1L3YE92moqXi0l_5NtTwe9vLBgvQfjWtnytTG7Q3qLm7iBHzu322-WVX_MkHOoUJ0p5EpPT7nMaFULL4_XwrzbFa2GHh6PNy1fjOnp2khnz20vBG2Kgqs6sWECCYAMvE1whwo3i9mOQnpl6jBGuiMoK2IRV5wwuQ/w150-h200/%5BAsteraceae%5D%20Centaurea%20sp..jpg)
![[Asteraceae] Solidago sp. with hoverfly [Asteraceae] Solidago sp. with hoverfly](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgD-cjiQDaDDTzRPQmpKKLf-aqoByDIws52j7dbTYeWtMIOy4Rr27Fo32kGFtsX5TO5ksFyNHRxtkDhnB9rnT_qbkRdSQChmKuVP3_nqTYKeWx_SSPxONmCxFkbnD-B2vA2kKTSbAKerzY7hMBjlgHJsSz9-sdOS9iGw3hQSWFdSM7HUz1rowvWrrgfww/w200-h151/%5BAsteraceae%5D%20Solidago%20sp.%20with%20hoverfly%201.jpg)
![[Asteraceae] Solidago sp. with hoverfly [Asteraceae] Solidago sp. with hoverfly](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhuROfBnj54gv-wEASVTCK5jJghSkC95uaZr0d6b7H0bPOlq88eVtEVjBj-Yci9dfdWSPqakb9PZ_NiGnces3sEw4TITAbkOMlHZ1i8lR6AuB9U5buNFrX-FMd9pyE5wgtNFyxwHCyaiDSaofSOdtUUthK-uK8nmZinsKv8S_w7ADC1VOfDld4C_2XMRQ/w200-h151/%5BAsteraceae%5D%20Solidago%20sp.%20with%20hoverfly%202.jpg)
![[Gentianaceae] Gentiana asclepiadea [Gentianaceae] Gentiana asclepiadea](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhq6CxfR7olmN4Y2Cg94dS4jEgK-XYIteEEfQzh8TUaE9qcRLt1NyAD1HuRxDoLId4lmFlfULnDiI2lYJRkOwzD-82qXtGipQMDB6bWkUKbrqciZrYeJ5Q5pYuXRwmlNzixVB30EQmXZ_PPMzboO8esCuJSH-OkIJUQhq_vufY3I_foUqiJUWe6aOm-1w/w150-h200/%5BGentianaceae%5D%20Gentiana%20asclepiadea%201.jpg)
![[Gentianaceae] Gentiana asclepiadea [Gentianaceae] Gentiana asclepiadea](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhtr-yopBy8Kf0WBKPJt_Q3sgHFKZW0tEWcZ1pyXIfn_aDWlDVl5U5dui8YyHpW3EUqPlaK2tCK5MbP8RaJBDeTp038_1jQKtZNbshPcXIr4A8QWGCcnB_FImtcj-G9jNHTb8rfdbZybaEAwgdBmcmso6hL79c-puO-9-H9VpYRoHjjH6MhzS-iF48REA/w264-h199/%5BGentianaceae%5D%20Gentiana%20asclepiadea%203.jpg)
![[Ranunculaceae] Aconitum napellus [Ranunculaceae] Aconitum napellus](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgKdaVhj4lXSpq0Vy6aaxvBWSnmqleIn2LHGJW_ZWiiCtJdDLUerKG0LPY2mkk_SsRDrapIsgxO9ZuQ7pBTaTx6MFmGtBniXpcZd2pAEessK8kozLnocOJHfJnI5o6XPKdWNSeLP1hXOyVDTdtWp4VhAO81YzmYr6MJzFzuwiW6saJ4JM4ujHDwSBBE5w/w150-h200/%5BRanunculaceae%5D%20Aconitum%20napellus%201.jpg)
![[Ranunculaceae] Aconitum napellus [Ranunculaceae] Aconitum napellus](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEj7PyXZIbEKcm4czNv_Ey8O_6wXTvpy433s0tS1skuDRcfADRBH618bwalEmA-OlfU9uG_UUBymFca9AXh6OpHp5c2G03doFAhJpN9EIKl-b2Q6FvXcoqqRCeA6EDTQULItQSpMNEEtwwRJstcJBoZH5M0vnim1q_31j6Q6foOg_YCIiHVadJbvJYQtsA/w265-h200/%5BRanunculaceae%5D%20Aconitum%20napellus%202.jpg)















.jpg)

%2001.jpg)

.jpg)





%20-%20five%20vases.jpg)








.jpg)
%20-%20Circulograph%20(1983)%2002.jpg)
%20-%20The%20Drone%20(2021).jpg)
%2002.jpg)


%2001.jpg)
%2002.jpg)
%20-%20Counting%20the%20Rice%20-%20R%C3%A4kna%20ris%20(2005)%2002.jpg)