Every SaaS Business Will Become a Harness Around a Model

The Harness Is the Company

Every SaaS Business Will Become a Harness Around a Model

The author argues that all SaaS companies will eventually become harnesses around stateless LLMs, progressing from engineers pairing with agents to proactive background agents that orchestrate humans. The company's domain knowledge, tooling, permissions, and review loops become the business harness, and differentiation shifts to how well a company builds that harness rather than the software itself.

Following this trajectory, you've actually turned your company into a harness. The work to produce the software service has moved from people to the harness.
  1. socializer

    I've heard some variant of the SaaS doomsday prophecy dozens of times over the past year. Somehow, SaaS is still here and more or less looks the same.

    It will probably continue to exist because most businesses are perfectly content to outsource tech problems to someone else for a reasonable fee, and if you think that managing an agent swarm is a seamless replacement for that, I don't know what you tell you. The vast majority of business owners don't want the added complexity in their life.

    And for the same reason, most SaaS will probably not migrate to some radically new UX paradigm. Some of it will be vibecoded, but the UX will probably continue to be deterministic. I think this might change once a new breed of "LLM-native" business owners takes charge, but that's going to be a slow process, and it will be hampered by prosaic concerns about interoperability, support, predictable cost, and liability for mistakes (from "the agent did a bit of tax fraud" to "the agent decided to hack my competition").

  2. munchbunny

    I go back and forth on this.

    But to pull something from the post:

    > If this is the right mental model, you should expect to see:

    > An unusual amount of in-house harness building on both the build side and the sell side

    Yup.

    > Org structures and individual roles being reshaped around their place in the business harness

    Not yet, this is the one I am most skeptical about because it's still way too easy for the harness-driven processes I've seen to go off the rails, so it's still very much human SME-driven. However, the number of butts in seats required is going down.

    > AI-native startups beating incumbents in domains where the “moat” can be easily harness-ified

    In most cases I've seen the "moat" is not easily harness-ified, and increasingly the money and energy seems to be going into getting to the starting line.

    > All software a software company uses (on or tied to the core build or sell paths) needing to be headless so the outer harness can run it

    Tech companies are definitely doing it, but I don't think this is actually realistic yet for companies whose core competence isn't software. But this feels like a matter of when, not if.

  3. hammock

    A lot of (non-software) business is already like this, without the AI part.

    There are 40,000 McDonald’s franchisees who basically just put money into a coinop machine and the machine returns with exactly the food, drinks, restaurant decor, prices, etc that it wants them to sell.

    Toyota’s production system coordinates what gets made and when, detects abnormalities and directs human attention toward problems and improvement. They explicitly eliminate the need for people to continuously watch machines, while preserving human judgment

  4. ylisav

    Where will systems of record go? back to excel again, i guess?

  5. ilaksh

    This is a side issue, but he mentions that the harness is stateless. But no agentic task is actually stateless.

    I suspect for purposes of optimization and performance etc., sometime in the next several months we will start to see new popular ML architectures or variations of LLMs/VLMs that are designed to be _stateful_. So a lot of the infrastructure for managing state and memory etc. gets sucked into the model somehow.

    They may end up changing or expanding the concept of an ML model to enable that.

    That type of belief is what is making me anxious about AI engineering staying relevant for much longer. I just wonder if pretty soon we need to be able to build and train or customize ML models that just do everything, or at least know how to prompt an agent to do that.

  6. dkarl

    This is the AI companies' fantasy, to replace simple, efficient, deterministic computing with expensive LLM calls. It'll be bad for customers, bad for users, bad for the environment, bad for everybody except the companies selling it.

    I'm a heavy AI user and haven't written code by hand in a long time (and probably won't until Advent of Code, honestly) but I'm already sick of people using AI to replace code at things that code excels at and AI is no better at, just slower and more expensive.

    It isn't just driven by the AI companies, either. I've already seen it internally, with developers who need diagnostic tools creating Claude skills, and in the excitement of adopting AI, barely noticing that they're incredibly slow, expensive, unreliable, and even slow to create compared to the script or diagnostic endpoint they would created in the past. And they're surprised to learn that Claude skills suffer from bitrot and cross-repo dependencies just like traditional software: somebody's Claude skill fails because it wasn't updated to match changes elsewhere, and it takes much longer to realize that a conversation with Claude is going wrong than if a script or an endpoint returns an error.

    When code is the best solution, by all means use AI to write it! But don't use AI instead when AI isn't fit for purpose.

  7. rrook

    it's all software, not just sass. the "application harness" is emerging, which bundles up the application, the operations of the application, and the sdlc of the application all into one deployable thing. agents/people just plug into it, and it looks _mostly_ like an org chart / task board.

  8. 2001zhaozhao

    This couldn't be more true. Companies will be driven by AI harnesses (as defined by this article) that automate decision-making and prioritization. In the medium term, may the company with the best harness win.

    In the longer term, the downstream impact is massive commoditization of software and invalidation of most existing moats. Data moats are gone if you can simulate the data with AI. Even platform effects can be sidestepped if AI replaces one side of the platform.

    In addition, while right now agile startups have the advantage, at some point the balance will start tilting towards whoever has the most tokens (OR perhaps durable moats will trump even near-infinite tokens; we will have to see). Startups have a limited time window to have whatever impact in the world they are hoping to have, or to build a moat that won't be disrupted by AI, but there are few of them left in the world.

    The upside is that when there is a lot of commoditization, then the consumer benefits.

More from this day

2026-10-02