According to a post on sshh.io, every SaaS business will eventually become what the author calls a “harness”—the infrastructure, interfaces, context, and state that surrounds a stateless large language model.
The author defines a harness broadly to include all the tooling and frameworks that wrap AI models in enough capability to produce usable work. Examples include LangGraph, Codex, Claude Code, and OpenCode. The concept extends beyond coding: a “software factory” is described as a harness composed of smaller harnesses—one writing specs, one writing code, one reviewing—plus orchestration deciding what runs when.
The post outlines a predicted trajectory for how SaaS businesses will evolve:
Stage one: Traditional SaaS with no harness, humans doing all work.
Stage two: Engineers pair with agents for productivity; other functions like product and sales follow suit. Individuals operate harnesses.
Stage three: Many core tasks move to background agents running in the cloud, with humans writing prompts and reviewing outputs. Individuals orchestrate harnesses.
Stage four: Agents work proactively with humans moving to sampled reviews. Agents increasingly decide what to do rather than waiting for human direction. Harnesses orchestrate individuals.
Stage five: The company becomes a harness. Core tasks are entirely agent-driven; humans focus on taste and judgment where it matters most. Domain knowledge, tooling, and review loops become the business’s competitive advantage.
The author addresses concerns about quality, arguing that well-designed harnesses can route human attention to high-impact decisions. A product agent might synthesize customer feedback for a product lead to review; feature suggestions become architectural decisions for engineers to judge; design variants get presented to taste-holders.
The post notes this shift is already beginning. Companies like Ramp, Stripe, and DoorDash have built in-house AI tools because waiting for vendor support increasingly bottlenecks their ability to ship products. The author predicts companies will own their top-level harness while plugging vendor products into specific workflows, rather than relying entirely on third-party solutions.
Key facts
- A ‘harness’ is defined as all infrastructure, interfaces, context, and state surrounding an AI model
- The framework predicts five stages of evolution from traditional SaaS to AI-orchestrated business operations
- Companies are already building in-house AI tools because third-party vendors can’t match their speed or integration needs
- Successful harnesses route human attention to decisions requiring taste and judgment rather than routine work
- If an entire business can be run by a third-party harness, the business becomes commoditized