Two names, one word apart, pointing at opposite futures.
Service-as-Software is the venture capital thesis: AI agents replace the people, and you sell the outcome. Service With Software is what we’re building at StudioHawk: the people stay, and the software arms them. One of these works for complex services. I’ll tell you which, and I’ll be honest about where the other one wins.
Credit where it’s due: the thesis is serious and the people behind it aren’t wrong about the size of the shift. Foundation Capital’s version is the cleanest: AI agents can now read context, generate output, set priorities and direct tasks like a skilled human performing a service. When a system of agents completes a whole job, it stops competing for software budgets and starts competing for personnel budgets, which are an order of magnitude bigger. So the play: build agents that do the labour, sell the outcome, keep SaaS margins.
In plain English: the service becomes the software. The people become optional.
I’ll steelman it properly, because pretending the other side has no point is how you end up writing reassurance content.
Some work genuinely should be close to fully agentic. In my own industry, chunks of information architecture generation can run agentically. Onboarding admin can be almost fully agentic, with a human handling the relationship while the machine handles the process. Across industries, any workflow where the context fits in the system and the cost of a wrong answer is low is agent territory, and it’s expanding every quarter.
If your service is entirely made of that work, the VCs are right about you, and you should read the 70% page with some urgency.
Here’s what the replace-the-people thesis runs into in practice: context.
A real example from inside our own system. HawkOS builds information architecture with all of a client’s context loaded: their data, their market, every IA we’ve ever built. And a human still overrides it, regularly, and the human is regularly right. The operator picks up what the machine can’t hold: this client is too small for this structure, that client is too big to do it this way, this industry has a quirk the data doesn’t show.
That’s not a temporary model weakness you wait out. There’s only so much a machine can hold in context, and only so many rules you can stack before it starts forgetting things, overwriting itself and hallucinating. The judgement that catches those failures lives in people who’ve done the reps.
We learned the same lesson the expensive way with reporting. We fully automated it, and it worked, in the sense that reports came out. But they didn’t have the insights each specific client actually cared about. The automation gets you a solid baseline, and then a human has to cut it down, sharpen it and make it about THIS business, or it’s rubbish dressed as insight. A perfect example of 70% of the way there.
So the wedge between the two models is one decision: what do you do with the people?
Service-as-Software says the people are the cost. Remove them, capture their salary as margin.
Service With Software says the people are the moat. Arm them: build the software, train it on your data, and let it lift every operator’s baseline so their time goes into the judgement layer where the value actually lives. Code, data, people. The full model is here.
Let me put a number on it, since predictions are only useful when they’re falsifiable.
By 2031 I expect 80-90% of our delivery work to be agentic, with humans doing the remaining 10-20%. And here’s the part that sounds contradictory but isn’t: I don’t expect to employ fewer people. If anything, more.
Because the baseline keeps moving. The “7 out of 10” that AI produces for free is a moving target, it rises as the infrastructure improves. Our whole bet with HawkOS is to make our automated baseline an 8 while everyone else’s generic AI produces a 7, and then to spend our humans getting clients from 8 to 9 and 10. The agentic share of the work grows forever. The human share shrinks in percentage and grows in value.
The honest close: these two theses aren’t fighting over the same businesses.
Thin services (interchangeable, low-context, low-consequence) will become Service-as-Software, and the VCs funding it will do very well. Deep services (judgement-heavy, context-heavy, trust-heavy) will become Service With Software, because in those businesses the people were never the cost. They were the product.
The question isn’t which thesis is right. It’s which kind of service you’re running, and whether you’re honest enough to answer.