The Minimum Viable Unit of Saleable Software – When Building Beats Buying in the Age of LLMs
The Minimum Viable Unit of Saleable Software – When Building Beats Buying in the Age of LLMs
TL;DR – Buying Still Beats Building for Most SaaS
Even with large language models (LLMs) that can generate code, many software products remain cheaper to license than to rebuild in‑house because the total cost of prompting, refinement, and ongoing maintenance exceeds the subscription fee. The sweet spot where a product is both novel enough to resist cheap LLM replication and priced reasonably is called the zone of viability. Brandur uses this framework to argue that his River job‑queue service can survive as a sustainable business.
1. LLMs Reduce but Do Not Eliminate Build Costs
"While LLMs have made software considerably cheaper to build, they haven’t brought it to zero."
- Building still requires an iterative feedback loop: prompt → generate → review → refine. Dozens of loops are typical before the output meets quality standards.
- Ongoing maintenance (bug fixes, feature additions) remains a human‑driven cost. The most expensive factor is the part‑time labor of the engineer overseeing the model’s output.
- Example calculation: an engineer earning $200 k/year costs $96/hour. Replacing a $400/month Jira license would require the engineer to spend no more than 4 hours per month on LLM‑generated maintenance to break even. Even a charitable estimate of 2 hours/month yields a 37‑month payback period, which is unrealistic.
2. The Build Threshold – When Rebuilding Becomes Viable
"For a $500/seat/month Salesforce license (≈ $25k/month for 50 seats), you could fund 1.5 engineers full‑time to clone the product."
- High‑priced, complex SaaS (e.g., Salesforce) can cross the threshold where the cost of licensing justifies a full‑time internal rebuild effort.
- The threshold is not static; it shifts as SaaS pricing changes and as LLM productivity improves.
- Commenters note that even expensive SaaS carries hidden costs (integration, compliance, disaster recovery) that can tilt the calculus back toward buying.
3. The Zone of Viability
"Software in the zone of viability satisfies two conditions: (1) sufficient novelty to make LLM rebuild non‑trivial, and (2) pricing not so exorbitant that rebuilding is obviously cheaper."
- Novelty: Features that require deep domain expertise, sophisticated performance tuning, or extensive API design are harder for an LLM to replicate accurately.
- Reasonable Pricing: If the subscription fee stays below the cumulative cost of prompting and maintenance, buying remains the rational choice.
- The diagram in the original post visualizes this sweet spot as a band between low‑complexity cheap tools (where DIY is trivial) and high‑complexity expensive platforms (where DIY becomes justified).
| Product | License Cost | Engineer Hours/Month to Match | Engineers Needed | Buy? | Build? |
|---|---|---|---|---|---|
| Jira | $400/mo | 4.2 hrs | 0.02 | ✔ | |
| Salesforce (50 seats) | $25k/mo | 260 hrs | 1.5 | ✔ |
Community Insights
- @zingar: Even when LLMs lower upfront effort, the ongoing motivation to maintain a side project often outweighs its utility.
- @ahamilton454: Real‑world iteration costs are higher than optimistic estimates; building “in a few days” rarely holds up under quality constraints.
- @monkeydust: Shared community contributions create externalities that individual rebuilds miss, reinforcing the value of buying.
- @bze12: A lower‑cost alternative (e.g., Linear instead of Jira) can shift the zone of viability downward.
- @jwitchel: Complex SaaS platforms consist of thousands of engineer‑years; LLMs cannot magically replace that depth.
- @ThePhysicist: Many enterprise products are essentially open‑source libraries wrapped in a management layer; the real cost lies in compliance, SSO, audit logs, etc.
4. River as a Case Study
"River is an open‑source Go + Postgres job queue with a free tier and a paid Pro version offering advanced features and billing."
- Novelty: Advanced workflow, sequential jobs, and concurrency‑limited jobs are designed with performance in mind, making them non‑trivial for an LLM to replicate with comparable fidelity.
- Pricing: Tiered pricing starts at $125/month for up to 20 developers, scaling sub‑linearly for larger teams. This places River comfortably within the zone of viability for small‑to‑medium development shops.
- Brandur plans to devote full‑time effort to River, betting that the subscription revenue will exceed the cumulative cost of LLM‑assisted development and maintenance.
5. Practical Takeaways for Engineers and Decision‑Makers
- Quantify the true cost of ownership – include licensing, integration, compliance, and the hidden labor of context switching.
- Assess novelty – if the product’s core differentiators require deep engineering insight, LLMs are less likely to replace it cheaply.
- Use the zone of viability as a decision matrix – plot your target SaaS on a cost‑vs‑complexity chart to see whether buying or building makes economic sense.
- Don’t let optimism about AI replace reality – iterative prompting still demands skilled oversight; budget for that labor.
- Consider community externalities – buying often grants access to a broader ecosystem of features, plugins, and shared maintenance.
The post’s opening anecdote about a LinkedIn user replacing a $400/month Jira bill with a Claude‑generated internal tracker illustrates the temptation to “build”. However, the detailed cost analysis shows that for most mid‑range SaaS, the math still favors buying – unless the product sits in the narrow zone where novelty and price align.