A Billion Dollars isn't cool. You know what is? Ten million dollars.

I am not a VC, an angel investor, or anything in between. I’m not even an entrepreneur! And yet, I think the idea that only hyperscale business are interesting is fundamentally incorrect¹ in the age of agentic AI. Instead, I think it’s quite possible that we’re about to see the revival of the business middle class as internet business models with unappealing economics for VCs become much more feasible.

In many of my recent conversations about the potential effects of AI on employment, jobs, and the economy, this theme has been coming up frequently. Most people are asking if AI is going to destroy jobs and I think that question is well covered in many other places (even if there isn’t a definitive answer yet).

However, even if you’re bullish that AI will create more jobs, as other technological revolutions have, explanations for how this can happen are pretty hand-wavy. “There will be new jobs we can’t conceive of yet!” And that’s almost certainly true. But regarding the more immediate future, I haven’t seen much discussion about is the kinds of business models that might be enabled by the rise of AI, specifically the rise of coding agents like Cloud Code and Codex. Justine Moore, ironically a partner at a16z, put it succinctly:

I don’t think nearly enough people understand this point: AI tools make it possible to create content that’s economically infeasible to produce with humans. You can make things - like local news or niche interest videos - that wouldn’t have generated enough $ to cover costs.

As I’ve written before, I find coding agents to be really empowering, specifically for someone with ADHD, for building things. But these agentic forms of AI also open up a new mode of internet business, which is actually a rather an old model of business. One that has been neglected in the tech sector for quite a while. By that I mean consumer-focused businesses that are both a) sole proprietorships or have a very small number of employees and b) generate significant revenue. Both of those components are important because it’s been possible to do one or the other since ~the dawn of the commerical internet but it’s been very difficult to do both. So difficult that it’s often remarked upon when a business is able to do both.

The problem

Almost everyone has had an idea for an app or business that would serve their specific niche needs, had their friends agree that it’s a great idea, but then realize they lack a) the technical skills to build it, b) don’t know if it would be a big enough market, and c) don’t know how they’d get it funded given (a) and (b)². So, the app goes unmade and needs / desires go unmet. Economists cheerfully call this deadweight loss — value from trades that could have occurred but didn’t due to some feature of the market (often unrelated to the quality of the product or service itself).

Lifestyle businesses

Let’s say you manage get over those obstacles to getting started. If you wanted your internet business to remain small, you have been somewhat pejoratively referred to as a “lifestyle business”, meaning your business generates enough income for a comfortable lifestyle but won’t enrich or enable you to reach large numbers of users. We can set aside the normative discussion as to why this would be unattractive and just take it as empirical reality that this is a common perception.

Hyperscalers

On the other hand, if you were seeking significant revenue, the primary mode of business enabled by the Internet has been that of hyperscalers. These are businesses that require enormous user bases for the unit economics to make sense or for the network effects to be sufficiently large to make the business profitable, therefore also accruing the gains to a small number of very large businesses. Typically these are backed by venture capital and require significant investment in order to survive long enough to either be acquired or to reach profitability. Customer bases have to grow huge. And to serve that giant customer base, internet companies have had to hire large numbers of employees, potentially rent large amounts of commercial real estate, spend millions on infrastructure, etc. Sadly, the reality is that building a hyperscaling business is a pipe dream for almost everyone. There’s a reason you can name all of the successful hyperscaled businesses using two hands.

B2B SaaS is an outlier here, in that the number of customers is usually small-ish with potentially high revenue, but these businesses often involve significant investments in sales teams, compliance teams, and so on. They are expensive and difficult to run successfully. Given this, most VCs have said that businesses without the potential for hyperscaling or a large B2B total addressable market are “uninteresting” (see the opening of this post) or simply aren’t economically feasible.

As a result, we tend to see a large gap in the market between small software companies and huge networked platforms. There is no internet business “middle class” or you could say there is a lack of SMBs, to use a less overloaded term.

What could come next

I think that could all change quite quickly as the cost of agentic AI goes down and output quality goes up. Coding agents allow less technical entrepreneurs to (more) easily build and maintain software. Instead of shopping around for a technical co-founder who you may or may not see eye-to-eye with, or who is looking for a giant exit, or is uninterested in the specific niche you’re trying to fill, now you can just get started on building the thing.

I’m thinking of solo founders like Claire Vo, who is building ChatPRD on her own while also hosting a great podcast. Claire is vocal about the fact that her “coworkers” are agentic AIs like Devin and shares the ups and downs of managing a fleet of virtual employees. ChatPRD even looks a little more like “present-day” B2B SaaS than the kind of SMB I’m describing here, but it’s clearly an evolution from the startups of even a couple of years ago. Claire isn’t seeking VC investment or to become a two-sided networked marketplace. She’s building a product that works very well and meets specific needs for her customers.

Lest I be accused of being too much of a Pollyanna, there are still some significant structural barriers for entrepreneurs especially in the US, that may make this an unrealized dream. Health insurance and health care are still often quite expensive for private buyers not receiving insurance through their employers. Most small businesses fail, online or not. Having a safety net available is a primary barrier to founding startups today. Childcare is not free (or even close to it). People have families to feed. Bootstrapping a business is costly in relative terms for the founder and it’s a hell of a lot easier to do so if you have a partner who can subsidize your entrepreneurship.

AI doesn’t solve these challenges; they’re political and institutional. But for those that can perhaps afford to take on the risk to start a business, AI may lower the stakes a little bit to get into a new space.

Footnotes

1. Of course, General Catalyst is a VC firm, and it’s definitionally true that giant business are the only interesting businesses to VCs. I’m just using this as an illustrative example. ↩

2. Kevin Kelly said a version of all of this way back in 2008, when discussing how anyone can be successful if they can find their 1000 true fans. Plus ça change! ↩