Every SaaS business will become a harness around a model, whether or not they’ve realized it yet. That is the thesis behind a new post making the rounds, and it is a wild one.
The argument goes that SaaS companies are headed toward a future where the actual work of building and shipping software happens inside a machine — a “harness” built around a stateless model API, surrounded by infrastructure, interfaces, and context that let it do useful jobs. Engineers, product managers, and salespeople trigger agents with prompts and review the outputs. Eventually, the agents themselves decide what to do next.
“The work to produce the software service has moved from people to the harness.”
Harnesses, Agents, and the Software Factory
The post defines a harness as the entire surrounding layer of infrastructure, interfaces, context, and state that wraps a stateless model API. It can be made up of smaller, task-specific sub-harnesses, with a coordinating orchestrator called a “meta-harness.” A “software factory” is a harness whose parts are smaller harnesses — one that writes specs, one that writes code, one that reviews — plus something on top deciding what runs when.
The progression the post sketches looks like this:
- Companies sell software services built the traditional SaaS-y way, with people doing everything.
- They sell software services, but engineers pair with agents to get the work done; product and sales also pair with agents for productivity.
- Individuals operate harnesses — they trigger agents with prompts and review outputs.
- Individuals orchestrate harnesses — core tasks move to background agents running in the cloud.
- Harnesses orchestrate individuals — agents decide what to do proactively instead of humans designing the work up front.
At the end of that road, the company itself becomes the harness, and the “product” is entirely model output. The humans left are taste-holders who review and redirect the machine’s decisions.
The Slop Factory Problem
The post admits the obvious objection: if products are primarily crafted and reviewed by AI, they will be inherently low quality. The response is that the harness is not supposed to be lights-out. The core mitigation is having the harness pick where human inputs matter most.
Examples include:
- A product decision coming from an agent fanning out questions to reps in customer meetings, then synthesizing a demo for the product lead to review.
- A feature suggestion from a customer meeting turning into a major architectural decision presented to an engineering taste-holder by the agent.
- A UI redesign kicking off after aggregating feedback, with the top options presented to a design taste-holder.
The idea is that a good harness maximizes value to the customer while spending human attention only where it is needed.
Why This Matters Now
The post argues that this is already starting to happen. In-house AI developer tools like those seen from Ramp, Stripe, and DoorDash are the beginning of it. For AI-pilled companies, waiting for an SDLC tool vendor to add an integration, support a certain interface, or reach a level of cost-efficacy increasingly bottlenecks their ability to build and maintain their product.
The harness becomes the core competency, and the differentiator. Trust, distribution, efficacy, and domain context are what set companies apart in a proactive background agent world — and the harness is now what shapes all of them.
The Inversion of Org Structure
The org chart stops being a hierarchy of roles and becomes a question of where to put people so the harness gets the most taste and judgment out of them. Humans are part of the harness, not the other way around.
What We Make of It
The post is speculative, but it is not entirely empty. In-house AI tools are already a thing, and the trend toward more automation in development and operations is real. The question is whether the harness can ever be trusted to do the outer loop — the planning, the decision-making, the judgment calls.
The post acknowledges that it is quite difficult to trust agents to handle the outer loop today, even with today’s frontier models. But it bets that models will eventually be able to do it, especially as more of this planning-and-review work gets broken into tasks with verifiable rewards that labs can train on.
That is a bet, not a fact. And it is a bet that rests on the assumption that training data and reward signals can capture the kind of taste and judgment that currently lives in human brains.
The trend toward automation is real, but the idea that companies will stop doing actual work themselves is a step too far. The harness is a useful abstraction — the layer of infrastructure, interfaces, and context that surrounds a model API — but the humans who build, maintain, and judge the harness are not going away anytime soon.
Source material: “Every SaaS business will become a harness around a model,” sshh.io.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.