AI was supposed to concentrate the economy into a handful of trillion-dollar firms. We think something stranger is happening at the other end of the market: the minimum viable company is collapsing toward two people, and the SME-dense stretches of the economy are about to get re-founded by tiny teams running on high agent leverage.
Consider the trade-off that has defined small business for a long time. In most of the real economy, revenue scales with people. To grow you hired; to hire you managed; and somewhere around employee thirty, many business owners (quite understandably) decided that “comfortable” beats “bigger.” Software companies escaped that trade-off decades ago thanks to the asset-light nature of their products and services. But the businesses that actually made, moved, or maintained real things never did.
Now AI agents can increasingly complete work at or above human standards for prices below what a small business owner would have to shell out in payroll or contractor fees. Agents’ cost structure mirrors that of electricity more so than labor or freelancers (i.e. metered and falling). Quoting, invoicing, scheduling, compliance paperwork, shipment tracking, tenant emails, timesheet chasing- generally anything under the “email job” umbrella is possible via agentic AI and minimal human oversight.
AI-Maxxed SMEs are now possible and in our view, inevitable. Humans will supply judgment, relationships, and accountability, and agents will supply almost everything else.
What does an AI-Native SME look like?
It’s early but we’ve seen two major distinctions so far in AI-Maxxed SMEs: light-touch businesses that nevertheless have some foothold in the real world, and exclusively digital firms.
Purely Digital Players
The general idea that 1-3 people can spin up some form of business with only a website (e.g. dropshipping, Amazon affiliate marketing) is not a new one. The phenomenon has just gotten more prolific since the rise of generative and especially agentic AI.
What has changed, besides just the pace of internet business creation, are its scope and level of human supervision. Generative AI has made more online business models accessible to newcomers (e.g. building websites for small businesses, various prosumer tools), and it’s also made the process of spinning up and running these apps extremely light-touch. Some entrepreneurs argue the less human intervention the better, since AI has better judgement and execution than people particularly when it comes to digital businesses (hordes of online data available → AI is better at consuming and acting on that data → Claude is a better dropshipper than you).
While interesting- particularly the thesis that AI is better at executing and indeed generating online businesses than people, what we find even more interesting is when this stops being a software-only story and tracing how agentic leverage is leaking out of the browser and into firms engaging with the “real” world.
SMEs that (Lightly) Touch Grass
These SMEs sit in the service sector of the economy but require trusting human-human relationships and/or some physical footprint (e.g. rented warehouse space) to operate. These light-touch services are a massive swath of the economy, including but not limited to categories like: wholesale and distribution, retail, freight brokerage, property management and lettings, facilities and field-service dispatch, equipment and event hire, staffing and recruitment agencies, specialist importers, self-storage operators, niche e-commerce.
Agentic AI can handle the vast majority of the email jobs-to-be-done and indeed many of the actual email jobs that would have employed a person in these types of firms. The people that remain in the firm focus on reviewing agentic outputs and owning the two things agents can’t (yet?) fake: relationships and judgement.
Picture Monday morning at a two-person freight brokerage: over the weekend, agents tracked shipments and flagged missing proof-of-delivery from the previous week. They drafted customs paperwork and emails to carriers’ whose insurance is about to lapse. The two owners of the brokerage walk into the office at 9am and spend most of their day on a queue of exceptions or drafts to review, and on the remaining workloads which still need a name attached to it (e.g. renegotiating a few lanes or talking down an angry customer on the phone).
The net result of handing off almost every task in the firm when you reach the top of the org chart is the one role that can’t be metered: the actual person taking a call with a frustrated supplier, signing a rental lease, and taking accountability for the firm’s failures and shortcomings.
Trust the (labor) market
Who is building these AI-Maxxed SMEs? Mostly it’s the young talent that used to default to banking and big tech. This is a consequence of 1) the appeal of owning a small, excellent business beats renting a desk at a large, indifferent enterprise and 2) those large, indifferent but nonetheless prestigious and high-paying enterprises have noticeably reduced junior hiring after the release of Opus 4.6.
How to get to that small, excellent business as a young person? Ideally you’re the child of the owner (some things never change), but generally the playbook for capturing some market share from these service SMEs was: sell them some software or consulting service, or offer the owner an EBITDA multiple and buy them out of it.
These plays certainly still exist (see: my previous breakdown of venture models in the AI era) and represent immense value creation for investors, entrepreneurs, and customers alike.
The past year’s leap in agentic capabilities has unleashed an additional entrepreneurial path: building a traditional business from scratch with AI-native principles and a fraction of the upfront capital and workforce required to do so.
Unlike the previous three models, an AI-Maxxed SME is not venture scale (a two-person freight brokerage cannot return a fund), but the fact that literally millions of people have a shot on goal to start, standup, and operate their own business without upfront capital, workforce, or indeed dilutive funding will be significant.
Where things stand now
We at Keen are seeing green shoots of this phenomena and believe these AI-Maxxed SMEs will be an inevitable fabric of our economy in the coming years.
None of this requires believing in AGI or a post-labor society. It just requires noticing that a huge chunk of the economy is currently run on operating models built for a world prior to agents, that the majority of tasks in these firms are within agentic reach, and that a generation of over-credentialed, under-hired talent has both the tools and the chip on their shoulder to do something about it.
After all, we are only in the first innings (T-9 months from the release of the first truly agentic models like Opus and GPT 5.3) of a very significant transformation as AI reshapes businesses, business models, and our economy as a whole. Most of the conversation has drifted towards enterprises, hyperscalers, and frontier labs- after all, it’s easier to think about a few logos than the vast majority of economic activity that takes place outside of them.
SMEs have always been the backbone of the economy, and with AI presenting the opportunity for entrepreneurs to rebuild and re-found every crevice, they’re perhaps becoming its frontier.
