Thoughts on AI and service businesses
Sometimes people ask why I decided to work as a freelancer / one-many-agency rather than trying to start a startup. These
are a few thoughts I've had on the topic. I might come back and organize these into a proper post at some point.
- From a customer POV, hiring a person is way easier than buying a product.
- That's especially true for larger customers. I'm not taking advantage of that at the moment cause I've only sold to startups
so far, but I can imagine it would be a big advantage if you were applying this business model to larger customers that have
long procurement cycles.
- Even though models are improving at an incredible rate, I'm sceptical of the current framing of "AI agents". Right now
99% of of "AI agents" aren't autonomous agents at all - they're almost always workflows.
- AI workflows, while not a flashy term, can provide a ton of value. Just look at the recent growth of n8n. It's absolutely exploding. If you're a 17 year old looking
to make money online then building n8n workflows is the 2025 equivalent of dropshipping or running Facebook ads in the 2010s.
- Workflows are good because a lot of the time you really don't need anything fancy. The base model is super powerful. You just need to be able to deploy
it, iterate on the context, and wire it in to the relevant data.
- I think a lot of domains will require a human-in-the-loop step for a pretty long time. Software engineering isn't like most domains, it's
probably the domain most amenable to autonomy (consider how little SWE time is spent in meetings compared to other jobs).
- AI makes the service business model much more viable for a few reasons.
- Firstly it stimulates demand. As mentioned, tons of businesses want to apply this stuff and at least for now the products aren't good enough
that you can just buy one and see results.
- Secondly it makes you much more efficient. You can deliver outcomes that would previously have taken weeks in days or hours.
- This in turn makes it easier to justify simple pricing. Modern freelancers don't bill by the hour, they charge a fixed monthly price.
- In practice, the line between product and service is more blurred than it is in theory. Sure, software has 0 marginal cost of reproduction on paper. But
in reality any VC-backed startup is gonna hire a GTM team, AEs, SE, CSMs, marketers, etc. There's lots of overhead. Monthly billing, abundant funding and
high standards for customer service mean that companies in practice end up investing a lot in customer service.
- What is a freelancer? Essentially it's a tiny, vertically integrated company that bundles sales and delivery in one unit. If you're a freelance dev then
you handle sales (clients) and also implementation (coding). This is different to being an employee, where you just work in delivery (unless you're an AE, in which case it's the opposite)
- The last thing to be automated by AI will be sales.
- Groups of people are not rational and every group has its politics. That's why you need enterprise salespeople, to navigate an organizational hierarchy, read the room,
and figure out how to get stuff done. I don't think that process is going away any time soon.
- But AI makes delivery much easier. So you can be a freelancer who essentially just focuses on sales and delegates the delivery to an AI. That's novel.