The Diffusion of AI Across the Real Economy

What does it mean for AI to diffuse across the real economy?

When I say the real economy, I mean the businesses most people work for. A regional distributor with 300 employees. A law firm with four offices. A credit union, a home health agency, a manufacturer that has made the same part for forty years. Together they employ most of the workforce and produce most of the output, and almost none of them have an AI team.

Diffusion is the point where a technology stops being impressive and starts being ordinary. Electricity is the classic case. Factories had electric power for decades before productivity moved, because the gains only came once owners redesigned their floors and their work around it. Having the technology was the easy part. Rebuilding how work got done took a generation.

AI is at the start of that same curve. The models are remarkable. The demos are everywhere. But the test of whether AI matters to the economy is simple: does the accounts payable clerk at a 200-person company spend their week differently? For most businesses today, the honest answer is no.

Why hasn't AI reached most businesses yet?

The capability is far ahead of its use. McKinsey's 2026 survey found that only 37% of respondents attribute any effect on earnings to AI.¹ Most companies have tried something. Far fewer have changed how the work actually gets done.

The reason is almost always the same. A model is raw capability, and a business runs on specifics. The model does not know your systems, your documents, your approval rules or which exceptions matter. Closing that distance is engineering work: connecting the systems, handling the edge cases, testing against real data, and keeping it running when something upstream changes.

Companies with their own engineers can do that work, which is why the early gains have gone to technology companies and the largest enterprises. Everyone else is left holding a login. On almost every first call we hear the same thing from people running established businesses. They are sure AI matters, they worry about falling behind, and they have no idea where to start or whom to trust.

Why won't better software close the gap on its own?

Software assumes someone on your side will operate it. Someone has to choose the tool, set it up, connect it to everything else, train the team and keep it working as the business changes. For a company without technical staff, every new tool adds to that load. A dozen AI products, each solving a slice of the problem, can leave a business more tangled than before.

There is also a mismatch in where the money goes. Businesses spend far more paying people to get work done than they spend on software to help with it. The software industry has always been able to sell only the smaller of those two budgets. What a business owner wants is the larger one handled: the reports done, the intake cleared, the claims processed.

AI makes it possible, for the first time, to sell that work itself at software-like cost. The question is who does the selling and the delivering.

What is an AI-native services firm?

It is a firm that delivers finished work to a business and uses AI to do much of that work itself. At Soren, AI agents do a large share of the building, testing and upkeep on every deployment, directed and checked by our engineers. A traditional consulting firm grows by adding people and billing for their hours. We grow by getting more done with a small team, and every improvement in AI makes that team more capable.

The last two rows are where the model earns its keep. AI changes every few months. A business that bought a tool in the spring may be behind by the autumn. A business with a partner whose job is to stay current stays current too, and each new model makes its systems better without a new project or a new invoice.

Why are AI-native services the way AI diffuses?

Because this is how every major technology has reached ordinary businesses. A family-owned distributor did not get online by studying networking. Someone built the website, set up email, connected payments and kept it all running. A whole layer of firms grew up to do that work, and the internet reached the wider economy through them. The same was true of accounting software, phone systems and the cloud.

AI needs its own version of that layer, and the old version will not do. A traditional consultancy cannot price AI work for a 200-person company, because its costs are people and its people are expensive. A software vendor cannot fit its product to how each business really runs. A firm built around AI from the start can do both: deliver work that is specific to one business, at a cost a mid-size business can carry.

1. It meets businesses where they are. The client describes a result in plain words. Nobody has to become an AI expert first.
2. It carries the engineering. The firm brings the people who connect systems, handle exceptions and keep things running.
3. It gets cheaper and better over time. AI does more of the delivery with each model release, and those gains reach the client.
4. It compounds. What the firm learns on one deployment makes the next one faster, across every client it serves.

Put those together and the economics change. A 200-person firm with no technical staff can get the same capability as a company fifty times its size. That is what diffusion looks like in practice, and I think it is the most important thing AI will do over the next decade.

What is Soren building toward?

Most of the attention in AI goes to products that take over one kind of task in one profession. That work matters, but an established business does not have one task to hand off. It has dozens of processes, and each one depends on its own systems, documents and habits. No single product covers how a particular company runs. Deciding what to change first, and then building it, takes engineers who sit with the business and learn it.

So we built Soren as an AI-native engineering firm. To a client we look like a services firm: you get a team, a relationship and a result we are responsible for. Inside, we run like a software company, with AI agents doing much of the building and upkeep. We have done this with law firms, financial firms and nonprofits, and with teams in fields that have no industry software of their own.

Our goal is for AI to reach the businesses the technology industry tends to overlook. A first workflow is typically live two to three weeks after we get access to the systems it needs, and the same team stays to run and improve it.

What should a business owner take from this?

1. Buy results. Describe what you want in terms of your business, and ask whoever you hire to be responsible for it.
2. Ask how the firm is paid. If it earns more when the work takes longer, your interests and theirs point in different directions.
3. Ask what happens when AI improves. The answer should be that your systems improve and nobody sends you a new invoice for it.
4. Start with one result. A single working workflow in a few weeks will teach you more than a year of reading about AI.

I am bullish on this because I see it work. A team that has done the same reporting by hand for a decade gets it in five minutes. People go back to the work they were hired for. Multiply that across millions of businesses and you have AI changing the real economy. I think the firms that make it happen will be some of the most important companies of this period.