My last post described my attempt to generate a report on housework using Deep Research, and the way it came to a crashing halt. Over the fold, I’ve given the summary from the last version before the crash. You can read the whole report here, bearing in mind that it’s only partly done.
As I said, I chose the questions to ask and the points on which to press further. DR extracted the data (I was planning to get detail on this process before the whole thing crashed), produced graphs to my specifications and generated the first draft of the text, with a style modelled on mine.
If I were doing this to produce a report for publication, I’d initially I was about halfway there, after only a few hours of work on my part. But as with LLMs in general, I suspect the final editing would take quite a bit longer.
Still, the alternative would have been either nothing (most likely) or a half-baked blog post using not-quite-right links to the results of Google searches. So, I’m going to keep on experimenting.
Early versions of LLMs were mostly substitutes for medium-level skill. It made it easy for someone barely literate to generate an adequate business email or (in the graphics version) for a complete klutz like me to produce an obviously-AI illustration for a post (Substack expects some kind of picture)
But with Deep Research, I think there’s an amplification of general research skills. It’s ideal for topics where I have some general idea of the underlying reasoning, but am not familiar with the literature and am unaware of some important arguments
Single-person households reveal the underlying structure of domestic labour more clearly than multi-person households, because cooking, cleaning and household administration can-not be negotiated or delegated. Patterns among singles therefore show how norms, incentives and technology shape behaviour directly. Over the past four decades, both single men and single women have experienced substantial changes in domestic tasks, but the scale and nature of these changes differ sharply by gender.
Single women display long-run stability in cooking and cleaning routines. Their time spent on meal preparation remains high, though modestly reduced by more efficient appliances and improved supermarket offerings. Cleaning remains routinised, but technologies such as cordless vacuums, microfibre cloths and robot cleaners reduce the effort required to maintain prior standards, as does the the near-elimination of labour-intensive tasks such as ironing and mending.
Single men show much larger behavioural shifts. The share of meals prepared at home and weekly cooking time both increase substantially from the early 2000s onward. This shift reflects not a dramatic rise in cooking skill, but improvements in technology that substitute for skill—particularly microwaves, air fryers, rice cookers and standardised convenience foods. These appliances lower the risk of failure and expand the feasible meal set. Cleaning time among single men also rises modestly, driven by increased home cooking, higher hygiene expectations and accessible cleaning technologies.
Digitalisation reshapes household administration in two distinct ways. Pre-existing tasks—bill payment, insurance renewals, contact updates—become simpler and generally decline in time as online systems streamline formerly paper-based processes. At the same time, new digital tasks emerge, such as managing streaming services, software subscriptions and multiple platform accounts. These represent expanded digital consumption rather than an increased administrative burden in the traditional sense. For single adults, the total time spent on ad-ministrative tasks remains roughly stable, but its composition shifts from inherited analogue chores to the maintenance of digital services.
Both groups benefit from the falling time cost of provisioning through online grocery shopping, which saves roughly 45–60 minutes per week. While total domestic labour has not fallen dramatically, its composition has changed: physical labour has declined in favour of cognitive and digital administration, and routine tasks have become less burdensome through automation and improved design.
Overall, technology—not exhortations like those of Jordan Peterson—has been the central driver of change. Single women maintain stable routines with efficiency gains, while single men adapt more dramatically as skill-substituting technologies, shifting incentives and expanded digital consumption reshape their domestic production decisions.
{ 13 comments }
Alex SL 12.11.25 at 2:37 am
Still, the alternative would have been either nothing (most likely) or a half-baked blog post using not-quite-right links to the results of Google searches.
To paraphrase what my teenage daughter said when they were told by the teacher that they could use generative AI for an arts project: “I would rather hand in a crude stick figure”. As they say, the kids are alright.
And I feel the same about writing. You do whatever works for you, of course, but I am as puzzled by the idea of using AI to do literature search and writing something for me as I would be by that of having a robot watch a movie for me. I want to learn from the literature search, and I want to express myself in my own words in any text that has my name on it.
Expanding the perspective from the personal to the social, then, it is fairly predictable that grant proposal panels will increasingly be inundated by proposals from PIs who do not actually understand most of the cited literature but had an agentic system vomit out a plausible-sounding literature review and project plan. Indeed, just yesterday a colleague told me they had participated in a workshop for an LLM system for research (not the one you are using), and one of the presenters explicitly advertised it to the effect of you just have to have an idea, and the system develops a hypothesis, summarises the literature, and writes the proposal for you. My colleague’s reaction was that nobody who does that should work as a scientist. More to the point, unless funding agencies want to waste escalating amounts of reviewer and panelist time and even more escalating amounts of grant money on people who will turn out to be incapable of delivering results, they need to think about how to deal with those proposals, starting, of course, with how to spot them, where to draw the line when in doubt, and how to sanction those who do it (desk rejection? ineligible for five years?).
John Q 12.11.25 at 5:02 am
Alex SL, I’m not really clear on what you are concerned about here. I’m not after the Nobel Prize for Literature, just a way to put my ideas into readable text quickly. I wanted to find out what had happened to housework for singles and to summarise that in a document. If it had been available from a single search, I could have just quoted the results verbatim, as I’ve done heaps of times before. Would you have had a problem with that?
From my experience so far, you need to do a fair bit more than type out the idea and let the machine do the rest. If AI gets to that point, I guess we will have to rethink what we mean by research. But at present, it does the work of a human RA much more quickly, and with about the same number of errors that need to be fixed.
Moz of Yarramulla 12.11.25 at 5:39 am
Housework in share houses is even more fraught, especially when there’s couples within the household. OTOH it also grinds a lot of the rough edges off most people, making them much more likely to survive in a nuclear household should they ever be rich enough to afford one.
I lived in share housing until I was in my 40’s and never found “one system that worked”, it was more of a red queen race as people came and went so preferences changed. Gender plays a role, but so does upbringing and maturity – I’ve lived with 40 year olds who had to be told “your mum isn’t here to clean up after you” as well as people who struggled to understand that standards have to be negotiated.
As with voting systems, there is no right answer and no system is guaranteed to be fair in the eyes of everyone (Arrows Impossibility applies to cleaning!)
A tricky but fascinating study would be tracking how individuals change the amount of housework they do as their living situation changes.
Alex SL 12.11.25 at 9:25 am
John Q,
Sorry, may have tried to pack too much into one response.
On a personal level, I have no concerns; I was merely trying to say that I do not understand the attraction of outsourcing the writing. Again, like getting a robot to watch a movie for me or, perhaps a better analogy, having a robot care for the plants on my balcony. I enjoy that. I want to know the literature myself, and I want to write my thoughts down myself. Those are to me among the most enjoyable and privileged aspects of being a researcher, as opposed to, say, manually measuring characters on dozens of specimens, which I would happily apply some degree of automation to.
Two thoughts occur in this context. First, it is possible that an AI agent searching for information and/or summarising it makes more sense in a field where most literature is books as opposed to mine, where most literature is relatively concise research papers. In other words, I can see how an AI is useful to check if a certain topic is covered somewhere in one of several lengthy books, and if so, what the author wrote (if you can trust it to get that right, which I wouldn’t, based on lived experience). But if somebody thinks they need AI summaries of research papers, they would be better served with a crash course on what a “title” and an “abstract” are. Second, I have never had a research assistant and am, to be frank, not entirely clear about the concept. I am paid to be a scientist, so I would expect to do literature research myself; that is my core competence, just like I would be surprised to find that a statistician has a statistics assistant to do the statistical analysis for them. Maybe again a cultural difference between fields, or between university and science agency?
My concerns apply at the social level. What will the availability of this technology, the targeting of Deep Research and similar products to researchers as customers, do to research? Here my assumption is in the first instance that generative AI models do what they do best: generate text, ideally but not necessarily based on what is factually accurate (doi 10.1007/s10676-024-09775-5). That will lead to a mixture of well-intentioned researchers relying on agents that hallucinate references and/or use them to support statements that they do not actually support and less well-intentioned researchers who do not even care about quality but are now empowered to spam out grant proposals that they would not be able to explain if challenged directly, much less deliver on — even if you, personally, are neither of those. To use an analogy, you may find a car extremely useful and be a cautious driver, but a society that normalises car use and soon expects everybody to drive a car still has numerous serious downsides. I would rather not end up in a place where everybody around me uses generative AI agents to tell them what the literature says or to write their papers because their use has become normalised.
From my experience so far, you need to do a fair bit more than type out the idea and let the machine do the rest.
Of course; I was merely saying that the system my colleague told me about was advertised as being able to do all that. I do not believe that for a second, as I have indeed tried it out myself, and it was completely useless for any serious work. But we are still in an AI hype bubble.
Eszter 12.11.25 at 12:04 pm
Microwave has changed things in the 2000s? That seems unlikely, it was quite widespread well before then already, no?
I’ve found Deep Research to be very helpful at times, I suspect because I know some of the literature about which I have questions, and because I know how to evaluate sources. I’ve seen students use it in very bad ways and I suspect it’s when they don’t have enough core background to know how to ask the questions, how to refine the prompts, and how to make sense of what they’re given.
An especially big issue is to recognize what DR missed entirely. You have to know the literature at least somewhat to come to that realization. For areas where I have little knowledge (like medical domains where I’ve run some queries) I wouldn’t necessarily know if I’m missing an entire angle.
MisterMr 12.11.25 at 5:22 pm
@Alex Sl
“More to the point, unless funding agencies want to waste escalating amounts of reviewer and panelist time and even more escalating amounts of grant money on people who will turn out to be incapable of delivering results, they need to think about how to deal with those proposals”
Simple: they’ll feed all the proposals to nother AI which will do the summarizing and the choosing. What could go wrong.
KT2 12.14.25 at 3:32 am
The future is here, just not evenly washed
“Some five billion people in remote and developing regions still wash their clothes by hand. It’s a task that unfairly burdens women and young girls, who can spend up to 20 hours a week on the chore.’
“Flat-pack washing machine spins a fairer future
“A former Dyson engineer is rolling out a revolution for household chores in deprived communities after inventing an off-grid, flat-packable washing machine
Words by Robin Eveleigh, Fern McErlane December 3, 2025
…
Some five billion people in remote and developing regions still wash their clothes by hand. It’s a task that unfairly burdens women and young girls, who can spend up to 20 hours a week on the chore.
Enter Navjot Sawhney, who founded the UK-based social enterprise The Washing Machine Project(TWMP) to tackle this, and has now shipped almost 500 of his hand-crank Divya machines to 13 countries, including Mexico, Ghana, Iraq and the US.
The Divya washing machine, made up of an outer drum and an inner one which rotates, operates a 30-minute wash cycle where it completes a 5kg load needing only a few minutes of manual turning.
…
https://www.positive.news/society/flat-pack-washing-machine-spins-a-fairer-future/
David in Tokyo 12.14.25 at 9:11 am
The whole AI thing is quite hilarious.
The idea that someone would expect me to be willing to read a paper that they had an AI write for them is completely insane.
But the people who like having AIs write their papers are enthralled by it.
Go figure.
Of course, one has to work really hard to keep one’s sense of humor operating when faced with this insanity.
David in Tokyo 12.14.25 at 9:30 am
When I left the Unix world in 1990, one of the things I thought of doing was management consulting. It’s an interesting game: they research stuff using in-house expertise and write up recommendations for their clients. For big money.
The thing is, the assumption in this game is that the consultant company is going to put some original thought into the problem being written up, and thus provide some value added.
But.
If they start, as Delloite has recently been caught doing, handing in AI generated reports, the customers are not going to be amused at the idea of paying big bucks for this.
I find this hilarious, although a company that big going belly up might not be hilarious to anyone else…
SusanC 12.14.25 at 2:36 pm
For example, an AI can suggest which research papers you ought to read for your research topic; and the set of all research h papers ever written in clearly too vast for any human being to have read all of it.
SusanC 12.14.25 at 2:41 pm
As a practical matter, I run into LLM hallucinations a lot when trying to use them for research.
LLMs will typically give you the correct answer when the answer to your question is well-known, and some made-up nonsense otherwise.
Problem: if the thing you are trying to write about is novel enough to be publishable research, related questions about it will be obscure enough that an LLM will fabricate nonsense answers.
John Q 12.16.25 at 9:41 pm
Eszter, you’re right that microwave ovens were in general use before 2000. It would be more precise to say that the cumulative effect of these innovations has made it much easier to produce good food with limited skill.
On the point of what DR missed completely, that’s always a possibility, and also the case if you do a literature search yourself or get it done by a research assistant or grad student. At some point, you need to present what you have done to (putative) experts and be prepared to be told that you’ve missed whole areas of the literature. As someone who spreads themselves pretty thinly across a lot of topics, I’m used to this.
John Q 12.16.25 at 9:56 pm
Alex “I am paid to be a scientist, so I would expect to do literature research myself; that is my core competence”
So, you go to the library, read papers there, look at the citations in those papers, and find them in the library, taking notes by hand?
Or do you use computer-generated citation indexes, Google scholar etc?
I’m guessing the latter. This then becomes one of those arguments (relevant to the topic here, and discussed in the 2012 thread) like priding yourself on “cooking from scratch”, which never means “making pasta from wheat you have personally grown and harvested”.
Seriously, I’m sure you don’t believe that searching literature is your core competence. The core competence of a scientist is formulating and testing hypotheses in your field. Checking what other people have done, and using that to inform your own work, is an important subsidiary task, like setting up the lab equipment, but that’s all. If you can do these subsidiary tasks more quickly and efficiently using automation, you should
Comments on this entry are closed.