Introduction

In 23 Things They Don’t Tell You About Capitalism, development economist Ha‑Joon Chang offered a provocative thought experiment: the washing machine changed the world more than the internet. At first glance this feels absurd—we live in a world where every aspect of modern life is mediated by digital communication. Yet Chang’s argument was that household appliances like the washing machine freed people (particularly women) from hours of domestic labour and effectively doubled the labour force. The internet, by contrast, mostly improved our ability to communicate and entertain ourselves; its productivity gains are real but subtle.

Over the last year I’ve found myself coming back to Chang’s analogy when thinking about AI agents. At least once a week I discover some mind-blowing task that AI can do for me. In my 5-minute walk home, I can have AI do deep research into any topic that would normally take hours. It doesn’t just summarise results; it crawls academic sites, pulls data, compares metrics, highlights limitations and produces a structured report while I walk home. Companies like Sierra have AI do customer service tasks end-to-end. Coplay has AI do the mundane boiler plate tasks of creating a game in Unity. AI is doing work for us, freeing us up to focus on the more interesting/fun parts of the task. In these moments, AI feels less like a search engine and more like a washing machine.

In this post I’ll unpack the washing‑machine analogy, compare the market structures of appliances and the internet, and consider where AI agents might fit.

How the washing machine transformed the economy

In the mid‑20th century doing laundry could occupy an entire day. Automatic washers and dryers liberated people from this chore and enabled them to join the paid workforce. Chang argues that this shift—alongside other “mundane” household technologies—doubled the effective labour supply and drove a significant share of 20th‑century economic growth. Less time washing clothes meant more time for education, childcare, community work and paid employment. Sound familiar? Washing machines didn’t just make life convenient; they changed labour participation rates, household structures and gender roles.

Now the capex needed to start a washing machine company is slightly lower than what’s required to build the next frontier AI model. So the market has remained quite fragmented to this day with multiple key players such as Whirlpool, Electrolux, GE Appliances, LG, Panasonic, Samsung, Sharp, and Toshiba. But since everyone claims the AI wave of today feels like the dotcom era, maybe it will look more like the market landscape spawned by the internet?

The internet: ubiquitous yet concentrated

The internet’s impact is obvious: e‑mail replaced letters, social media redefined social life, search engines surface information instantly and SaaS tools run entire businesses. However, its productivity gains are harder to isolate. Chang noted that the internet’s biggest influence has been in leisure—access to news, entertainment, and socialisation. Even digital advertising, one of the internet’s most lucrative sectors, is dominated by just a few firms. In 2025 Google and Facebook together account for more than 60 % of global digital advertising spend. Their scale and network effects create high barriers to entry, allowing them to extract enormous rents.

So while the internet has enabled thousands of small SaaS companies, the core platforms—search, social, advertising—are controlled by a handful of giants. This concentration contrasts sharply with the washing machine industry’s fragmentation.

Will the agent market resemble washing machines or the internet?

The AI application layer is like the washing machine and foundation model companies are like Google and Facebook:

1. Fragmented, appliance‑like market: There are countless ways to embed agents—legal, medical, design, marketing, game development, etc. Like washing machines, switching costs could be low (you can always try a different agent), leading to intense competition. With open‑source frameworks and fine‑tuned models proliferating, we might see dozens of vendors carving out niches, just as multiple appliance manufacturers co‑exist.
   
   

2. Platform‑dominated market: On the other hand, agents rely heavily on foundation models, data pipelines and platform integrations. Few companies can train trillion‑parameter models or index the open web. If the key bottlenecks are compute, proprietary data and distribution channels, then a few giants could capture most of the value—much like Google and Facebook dominate digital ads. Network effects (the more users an agent has, the better its feedback loop) could create feedback flywheels, with Cursor being a great example of this in action.
   
   

My guess is there will be foundational models provided by a few hyperscalers, but an ecosystem of specialised agents built on top. I guess like washing machines connected to a shared power grid. Some agents will be built into operating systems and development environments; others will be stand‑alone SaaS products. The battle will be less about who controls the agent concept and more about who has access to training data, platform distribution and user trust.

Conclusion

When my grandmother bought her first washing machine, she didn’t marvel at the mechanical engineering; she just enjoyed having her day back. AI agents offer a similar promise: a chance to reclaim time from drudgery. The history of the washing machine reminds us that labour‑saving technologies can have profound social and economic impacts, yet the markets that produce them are often fiercely competitive. Meanwhile, the internet’s story warns us that a handful of platforms can capture enormous value.