Expert View 6 min read

AI makes code cheaper. What should businesses pay software vendors for?

An expert analysis on software economics in the AI era: evaluating in-house development vs. SaaS subscriptions, total cost of ownership, and vendor value propositions.

If two developers can replace a product for which a company has paid a subscription for years in just a few months, the vendor should take a close look at its offering. Perhaps the client was buying significantly more features than they used. Perhaps their needs have changed. Or perhaps the price genuinely no longer aligns with the value provided.

The occasion for this discussion was a publication on Habr about the economics of the software business in the era of vibe coding. It retells a message from the comments: an unnamed clinic allegedly replaced its subscription to an off-the-shelf app with in-house development in three months using two specialists. This is testimony from a discussion participant; without a verified estimate, comparison of features, and data on subsequent operation, it does not prove the economic superiority of either approach.

Nevertheless, the question is valid: what should a business pay for when creating the necessary software becomes easier? In my view, vendors will have to explain their pricing more precisely, while clients will need to learn to calculate the cost of in-house development with the same care they use when checking a subscription invoice.

Compare the same work over several years

I would start with a list of features that the company actually uses. For a clinic, these might include appointment booking, rescheduling, and patient notifications. This is a hypothetical example: we do not know the requirements of the facility from the discussion. Replicating the vendor's entire product might be unnecessary. However, demonstrating a successful booking is not enough if a production system requires access control, data exchange with other applications, and disaster recovery.

The next step is to agree on an acceptable level of service. How long can a process be unavailable? What data can be restored from a backup? Who responds to an error and within what timeframe? These requirements depend on the task. An internal tool used once a week does not necessarily need round-the-clock support. For a system whose downtime blocks customer service, the requirements will be different.

For an initial comparison, I would use a three-year horizon. For in-house development, it is worth factoring in requirements analysis, application creation and testing, data migration, infrastructure, external services, support, and modifications. In-house staff time also has a cost: while working on this solution, they postpone other tasks. For an off-the-shelf product, implementation, integrations, training, paid custom enhancements, and potential migration to another system are added to licenses or subscriptions.

It is useful to test such an estimate under several scenarios: standard operations, load increases, changing requirements, and replacing the maintainer. If the conclusion about savings relies solely on the assumption that nothing will change for three years, the decision still requires verification. This applies equally to custom code and a seemingly advantageous SaaS pricing tier.

In-house development can be the right choice

A niche tool with clear boundaries can be more cost-effective to build in-house. This is especially true if off-the-shelf products require expensive customization and the company has a team capable of supporting the result. Process uniqueness is a compelling argument for custom development, yet even a standard task can have a simple and economically justified internal solution.

An off-the-shelf product should be considered when it already covers the necessary scenarios, deploys quickly, and its support terms meet the company's needs. There is also a third option: leave standard features on the platform and implement the specific part of the process with a custom module. In that case, the estimate must include integration maintenance and compatibility testing after updates.

The use of AI itself does not determine the quality of either option. A model can help write code, prepare tests, documentation, and debug errors. The team must verify whether the result meets requirements and can be used safely. Unchecked acceptance of generated content and engineering work with AI provide different grounds for trust.

Vendor lock-in and dependency exist in both models. A vendor can raise prices, restrict data export, or stop developing a required feature. An in-house team can leave behind a system that is difficult to hand over. The mere fact of owning the source code does not guarantee independence. It should be tested practically: can another specialist deploy the system based on documentation, restore data, and make an agreed-upon change?

Vendors will have to show what is included in the price

Ukrainian product companies should view cheaper software development as a reason to rethink their value proposition for clients. Industry expertise, ready-made integrations, and implementation experience carry value when they help launch processes faster and reduce client expenses. Support carries value when it is clear which requests it covers, how quickly it responds, and what happens after a major failure.

These commitments must be verified in the terms of cooperation. Having a subscription does not guarantee data recovery or compensation for losses, and the label "SLA" does not explain all service limitations. On its end, the vendor can use AI to lower its own costs, release changes faster, and improve service. The client has good reason to ask how these gains affect their terms.

Before deciding to replace a system, I would ask the team and potential vendor five questions:

  1. What scope of features and level of service do we need, and how will we verify them before launch?
  2. How much will each option cost over three years, accounting for migration, support, modifications, and our employees' time?
  3. Who will restore operations after an outage, within what timeframe, and how will we test data recovery?
  4. How will we hand over maintenance to another team or extract data and migrate to another product?
  5. Under what changes in price, load, or business needs will the chosen option cease to be cost-effective?

If you are moving away from Russian software, the Software Ukraine Navigator offers a starting point for finding Ukrainian alternatives. Each proposal then needs to be checked against your requirements, tested on your own workflow and included in the same cost comparison.

A client who decides to build an in-house system is not necessarily mistaken. A vendor charging for an off-the-shelf product must demonstrate what work and commitments it takes on. The quality of this discussion will determine who gets the budget: the one who promises fast code more persuasively, or the one who can deliver the result the business needs throughout its entire lifespan.

Sources & materials

Materials and sources used in this article.

  1. publication on Habr about the economics of the software business in the era of vibe coding — habr.com
Tags: AI