Every SaaS business will become a harness around a model
Claims that future SaaS companies will mostly be “harnesses” around large AI models spark sharp disagreement over how far and how fast AI will reshape software businesses. Proponents argue that agentic “software factories” and in‑house AI orchestration will commoditize traditional SaaS, erode data moats, and shift advantage to those with the best harnesses and most compute. Critics counter that most firms pay for reliability, risk management, trust, and human judgment—not just code—and predict SaaS will persist, with AI used as an internal tool rather than the core runtime for entire businesses.
Concept of the “harness”
- A “harness” is described as all the infra, interfaces, context, and state around a stateless LLM.
- Several commenters argue this is basically what SaaS or any application already is, so the framing feels like a relabeling rather than a new concept.
- Others see value in explicitly treating process knowledge and org structure as something formalized and pumped into LLMs.
Owning the harness vs buying SaaS
- Some argue AI-native companies must own both model and harness or they have no moat.
- Others highlight that most businesses don’t want to run their own “agent swarms”; they pay SaaS to get rid of problems, risk, and operational overhead.
- Examples like Slack vs. self-hosted chat (Mattermost/Zulip) and GitHub vs Gitea are used to show that “you could clone it” hasn’t killed SaaS, because reliability, polish, support, and not having to maintain it matter.
“Software factories” and vibe coding
- Enthusiasts claim LLM-powered “software factories” can quickly generate internal tools or even full SaaS clones (e.g., Jira, e‑commerce sites, mobile apps) via one-shot prompts.
- Critics call this snake oil, pointing to latency, bugs, UX issues, and long-term “context rot” that accumulates technical debt.
- There’s broad skepticism that complex, mature products like Jira can be fully replicated in “two weeks of vibe coding.”
Moats, commoditization, and data
- Pro-harness commenters predict massive commoditization of software: AI reduces the value of traditional feature moats; synthetic data weakens some data moats; platform lock-in shrinks if LLMs can transform and migrate data.
- Others counter that in domains like science/biology, real-world data is becoming a stronger moat, since it’s costly and time-consuming to generate and can’t be faked reliably.
- Remaining moats mentioned: capital, network effects, relationships, research talent, proprietary sensors/data, and especially trust/brand.
Org structure, management, and workers
- Some foresee harnesses orchestrating both software and people, automating decision-making and replacing layers of management.
- Others worry this leads to surveillance, non-deterministic behavior, and stressed employees being evaluated or driven by opaque models.
- A recurring theme: AI might shift power from middle managers to executives and to whoever controls the models and tokens.
Costs, practicality, and future of SaaS
- Concerns: LLM calls are expensive, slow, and unreliable; running “everything” through a model seems wasteful and risky.
- Many expect SaaS to survive but be reshaped internally by AI, with some products commoditized and profit margins pressured, rather than SaaS disappearing entirely.