In a paper published in the Atlantic Economic Journal, economists Sandipa Bhattacharjee and Timothy Haase of Ramapo College of New Jersey develop a simple household production framework that explains why some time-saving innovations deliver systematic reductions in housework while others barely move the needle. Their central insight is deceptively simple: the relevant question is not whether a technology saves time, but which component of the household production problem it actually changes. Technologies that look superficially similar can have sharply divergent effects on how families allocate their hours.

The framework builds on a long tradition in economics dating back to Gary Becker’s landmark 1965 theory of time allocation, in which households are treated as small production units that combine time and market goods to produce things people value, such as clean homes and cooked meals. Later work by Robert Gronau, Jacob Mincer, and others refined this approach, and quantitative studies by Jeremy Greenwood, Valerie Ramey, and their collaborators have used it to explain long-run trends in work and leisure. What Bhattacharjee and Haase add is an explicit distinction between two classes of technology that existing treatments often blur together.

The first class consists of efficiency-improving technologies: innovations that reduce the time required to produce one unit of a home-produced good. A robotic vacuum cleaner is a canonical example. It lowers the labor cost of each square meter cleaned, but it does nothing to change the price of the market alternative, such as hiring a cleaning service. The second class consists of technologies that lower the full price of market substitutes. The full price combines the money outlay with the time a household must spend obtaining the good. On-demand food delivery platforms are the clearest modern case: they slash both the waiting time and the coordination cost of replacing a home-cooked dinner with a restaurant meal.

The crucial difference lies in what happens to demand. When a technology makes home production more efficient, it effectively makes home-produced goods cheaper to produce, and households respond by consuming more of them. A robot vacuum may clean faster, but families may respond by expecting cleaner floors, more frequent cleaning, or larger homes that generate more work. The net effect on total housework time is therefore ambiguous: the induced increase in consumption of the home-produced good may be smaller than, equal to, or larger than the direct per-unit time savings. This is why a century of appliances produced only modest reductions in housework time. Efficiency gains were largely absorbed by rising standards of home-produced output.

Technologies that lower the full price of market substitutes work through a different channel. They make it cheaper, in both money and time, to buy the good instead of making it at home. When the elasticity of substitution between home-produced and market goods exceeds the aggregate demand elasticity, households systematically substitute away from home production, and housework time falls. In the authors’ formal model, the derivative of home-produced output with respect to the full price of market goods equals the market share times the difference between the substitution elasticity and the demand elasticity, a condition that produces a clear prediction rather than an ambiguous one.

To illustrate the mechanisms, the paper examines three contemporary technologies. Robotic cleaning devices represent the efficiency-improving case: they reduce the time per unit of cleaning but leave the market alternative untouched, so their effect on total cleaning time is theoretically indeterminate. Food delivery platforms represent the full-price case: by collapsing the time cost of acquiring prepared food, they generate consistent and measurable reductions in domestic food preparation time, a finding echoed by recent empirical work on delivery platforms published in Management Science. Meal kit services occupy an intermediate position, partially outsourcing shopping and portioning while leaving cooking itself at home, so their time effects fall between the two poles.

The authors organize these patterns using data from the American Time Use Survey, the U.S. Bureau of Labor Statistics survey that has tracked how Americans spend their hours since 2003. The series covering 2003 to 2019 and, after the pandemic disruption of 2020, 2021 to 2024, reveals the divergent time-use trajectories that the framework is designed to explain: persistent housework categories that resist efficiency gains alongside food-related categories that respond visibly to the arrival of delivery platforms. The survey data are publicly available, and the paper’s supplementary appendix derives the key equations from first principles under two-stage budgeting, including a worked Cobb-Douglas example in which the substitution and demand elasticities both equal one.

The implications extend beyond household appliances. Firms and policymakers routinely promote time-saving technologies on the assumption that they will reduce the time households devote to unpaid work, and the economic value of that assumption is enormous, shaping product marketing, forecasts of labor supply, and debates about gender inequality in domestic labor. If the technology in question is an efficiency improvement, the framework warns, the promised time savings may be quietly consumed by higher standards of home production. If instead the technology lowers the full price of a market substitute, and households are willing to substitute between home and market goods, the time savings are far more likely to materialize.

The analysis also speaks to a broader puzzle in economic history. Studies of the twentieth century, including Ramey’s reconstruction of home production time from old surveys and Bowden and Offer’s comparison of the United States and Britain, documented that the appliance revolution coincided with rising standards of cleanliness, childcare, and meal preparation rather than a dramatic collapse of housework. Bhattacharjee and Haase’s framework gives that historical pattern a precise analytical form: the appliance era was dominated by efficiency improvements, while the platform era, with its delivery apps and gig-economy services, is dominated by full-price reductions. The technologies of the twenty-first century may finally deliver the time savings that the twentieth century promised, not because they are better machines, but because they change a different parameter of the household problem altogether.

Subject of Research: How different classes of household technology affect time spent on home production

Article Title: Home Production and Technology: Why Time-Saving Innovations Do Not Always Save Time

Article References: Bhattacharjee, S., & Haase, T. (2026). Home Production and Technology: Why Time-Saving Innovations Do Not Always Save Time. Atlantic Economic Journal. https://doi.org/10.1007/s11293-026-09864-4

Image Credits: AI Generated

DOI: 10.1007/s11293-026-09864-4

Keywords: household production, time allocation, labor-saving technology, home appliances, food delivery platforms, meal kits, robotic vacuum cleaners, American Time Use Survey, substitution elasticity, full price, consumer behavior, economic theory

The platform, called BuildOcc and described in the journal SoftwareX, tackles a stubborn problem at the heart of building energy research. Traditional models represent occupants through fixed reference schedules, such as those published under ASHRAE Standard 90.1, which assign identical activity profiles to every day and every person. Later generations of stochastic models improved on this by sampling behavior from sensor data or national time-use diaries, using techniques like first-order Markov chains to generate realistic day-to-day variation. But these remain statistical descriptions: they can reproduce the shape of human activity without capturing any of the reasoning behind it. An occupant who turns off the lights before leaving is not following a probability distribution; they are making a decision based on context, habit, and intent.

BuildOcc’s central insight is that large language models, with their capacity for contextual reasoning and memory retention, are unusually well suited to simulating that decision-making. Each agent in the platform is built from six interacting components: a structured demographic persona, an activity scheduler, a memory stream, a reasoning engine, a simulation environment, and a persistent store that saves the agent’s cognitive state between operations. At every fifteen-minute timestep, the reasoning engine assembles the persona, the most relevant retrieved memories, and the current environment into a prompt, and the language model returns a structured action—adjust the thermostat, toggle a device, move to another room, or do nothing—along with a one-sentence explanation of why.

What separates BuildOcc from earlier demonstrations of language-model occupants is its grounding in real population data. Personas and activity schedules are drawn from the American Time Use Survey, a Bureau of Labor Statistics dataset covering more than sixteen thousand respondents, of whom 6,611 fall within the platform’s four demographic strata: employed single adults, retired couples, employed parents, and not-employed adults. Activity probabilities are aggregated by hour of day and day type, weighted so that each estimate represents the U.S. population. Appliance ownership comes from the Residential Energy Consumption Survey, and comfort bands widen for lower income brackets, reflecting the documented tendency of lower-income households to tolerate wider temperature swings to reduce heating and cooling costs. Work-from-home behavior is handled separately, with each agent carrying a probability of working remotely on any given day, drawn from where survey respondents actually reported working.

The memory architecture borrows directly from the influential generative agents work of Park and colleagues at Stanford. Every observation and action the agent takes is written to a memory stream with an importance score assigned by the model itself. Retrieval balances recency, which decays exponentially with a twenty-four-hour half-life, against that importance score. When the accumulated importance of new memories crosses a threshold, the agent pauses to reflect, distilling its recent experiences into higher-order insights that are fed back into the stream. The result is cross-timestep coherence: an agent that raised its thermostat an hour ago will recall that decision when a new request arrives, and judge the request against it. In one demonstration walkthrough, an agent raised its setpoint during peak electricity pricing while still away from home, then declined an educational demand-response signal after arriving, reasoning that thirty-five cents in savings was not worth the discomfort given the comfort band it had already stretched.

That demand-response interface is one of the platform’s most experimentally useful features. BuildOcc supports three mutually exclusive signal types drawn from established behavioral intervention typologies: direct commands, educational price information, and social norm comparisons. In validation experiments, the results were strikingly patterned. Educational price signals were the only type accepted in every demographic stratum, accepted by between one and three agents in five depending on the group. Direct commands were declined in all but one case, with agents reasoning that a thirty-minute air conditioning shutdown on a hot afternoon would push the zone past its comfort limits. Social norm messages—telling agents that most similar households had reduced their use—failed to persuade a single agent in any stratum. A fixed-schedule model, which has no mechanism to respond to a signal at all, could never produce these differences.

The platform’s validation proceeded in two tiers. The first tested the activity scheduler alone, comparing simulated activity distributions against the survey reference data using Kullback-Leibler divergence, a measure of how far one probability distribution departs from another. Every survey-grounded condition fell within or below the sampling noise floor expected from a correct sampler, while a deterministic rule-based baseline diverged by two orders of magnitude—worst for retired and not-employed strata, whose real schedules depart furthest from the standard working-adult assumption baked into conventional models. The second tier showed that demographic priors propagate into distinguishable behavior: retired-couple agents, home all day, were the most active, toggling devices and moving between rooms far more often than employed agents under identical environmental conditions.

Practical considerations have not been ignored. Because the agent queries a language model at nearly every timestep, cost scales linearly with simulated agent-days; measured runs came to roughly twenty cents per agent-day using Anthropic’s models, with wall-clock time dominated by provider latency. Researchers can eliminate per-call costs entirely by running a local open-weight model through Ollama, and the default temperature of zero means a fixed model version and prompt reproduce the same action on every run—though that determinism does not survive across providers or model updates, so long-term studies are advised to pin a local model. Integration is handled through three layers: a Python library for batch simulations, a stateless REST API suited to EnergyPlus co-simulation, and a Model Context Protocol server that lets tools like Claude Desktop, LangChain, or Home Assistant drive simulations without custom integration code. A plugin registry allows other groups to add new demographic strata, substitute alternative time-use surveys such as the Harmonised European Time Use Surveys, or swap in different memory architectures without touching the core library.

The limitations are stated with unusual candor. Each timestep is drawn independently from the hourly activity distribution, so simulated activity episodes are far shorter than the diaries record—simulated sleep averages under seventy minutes against more than five hours in the survey data—and no duration model exists yet. Agents take one action per timestep, the strata cover only U.S. demographics, and the memory importance scores are self-assigned by the agent with no external calibration. The validation establishes internal consistency, not behavioral realism; comparing simulated agents against measured occupant data, including field data from smart thermostats, is left for future work, along with multi-agent household simulation. Still, the conceptual shift is considerable. By treating occupant behavior as a structured, demographically parameterizable experimental variable rather than an uncontrolled source of uncertainty, BuildOcc turns the messiest part of building energy modeling into something researchers can finally hold constant, vary deliberately, and reproduce across laboratories—a small step toward buildings designed not for ghosts, but for us.

Subject of Research: Large language model-based occupant agents for building energy simulation

Article Title: BuildOcc: A large language model occupant agent platform for building energy research

Article References: BuildOcc: A large language model occupant agent platform for building energy research. (n.d.). https://doi.org/10.1016/j.softx.2026.103068

Image Credits: AI Generated

DOI: 10.1016/j.softx.2026.103068

Keywords: BuildOcc, large language models, building energy simulation, occupant behavior, demand response, American Time Use Survey, generative agents, EnergyPlus, open-source software, HVAC, artificial intelligence, energy modeling