# HELM and the market for constructible organizations

An app or website is only part of an operating business. The question here is how to keep running a quote desk, a maintenance operation or a client delivery business. I asked an AI to test that idea against the public market evidence, and to be hard on it.

This is desk research. There were no customer interviews and no paid experiments behind it.

## What I found
- The study proposes selling the ongoing work: coordinating people and agents, delivering results the buyer accepts and recovering when something breaks. A generated form or website would support that work.
- App builders already advertise running the work as well as building it. On their own product pages, Base44 advertises workflows built through chat, with waits, conditions and connected services, and Replit advertises agents that build other agents and workflows. The study did not test either claim, but "we go further than a website builder" no longer separates a product from them.
- The evidence of demand is about starting businesses and hiring. In a Gusto survey of 1,051 founders of firms started in 2025, 60% used AI to help launch, but only 3% said they likely would not have started without it. In a Gusto study of 2,262 of its own small business customers who answered its survey, adopters had about 7% higher headcount than comparable non-adopters over the four quarters after adoption, and the authors say it does not show that AI caused the hiring. The sample is not random.
- Adoption surveys measure use. In a Federal Reserve small business survey of 6,525 employer firms, a convenience sample fielded from September to November 2025, 46% said the business or its employees use AI, and among those only 7% had fully integrated it. A Simply Business survey of 1,842 UK small business owners in July and August 2026 found 47% use. Neither says a business runs itself.
- In the study's argument, the money is in the avoidable coordination and setup work of one recurring responsibility, and not in "all the wages in an occupation". Three spending pools overlap: existing owners, new founders and implementation partners, and their revenues must not be added up as separate markets.
- In the study's illustrative model, the margin depends far more on how many cases still need a human than on the cost of the AI: raising the share of cases that need a person cut the margin much more than halving the model cost raised it. These are modeled results, not observed ones; their size depends on the inputs, which need to be published for the calculation to be reproduced.
- The study argues that the moat would come from each new deployment costing less than the last, from trustworthy state across systems, and from distribution through operators who already hold the customers. The thesis fails if buyers will not delegate real work, if setup costs more attention than the old way, if native tools match the result more cheaply, or if customers build and never go on to operate.

## How it was made
The study dates from 12 September 2026 and draws on 45 public sources, 49 customer and community observations and 18 competitor dossiers. An AI research tool gathered the sources and drafted the study. I asked the question. Figures are current at that date, and many of the survey numbers come from vendors or self-reported samples.

## Sources
- [Federal Reserve, 2026 report on employer firms](https://www.fedsmallbusiness.org/2026-report-on-employer-firms)
- [Gusto, AI and small business hiring](https://gusto.com/resources/gusto-insights/ai-and-smb-hiring-2026)
- [Gusto, new business formation](https://gusto.com/resources/gusto-insights/new-business-formation-2026)
- [Gusto, how AI is reshaping small business formation](https://gusto.com/resources/gusto-insights/how-ai-is-reshaping-small-business-formation)
- [Simply Business, small business AI usage](https://www.simplybusiness.co.uk/knowledge/business-news/small-business-ai-usage-doubles/)
- [Eurostat, enterprise AI adoption](https://ec.europa.eu/eurostat/web/products-eurostat-news/w/ddn-20251211-2)
- [IMDA Singapore, digital economy report](https://www.imda.gov.sg/resources/press-releases-factsheets-and-speeches/press-releases/2025/singapore-digital-economy)
- [Cohere, automations and their early footprint](https://cohere.com/blog/automations-early-footprint)
- [Base44, durable workflows](https://base44.com/features/workflows)
- [Replit Agent](https://replit.com/products/agent)
