Skip to content

Technical debt

| Article: 6 min read | AI in e-commerce

Technical debt is the future cost of fixing and reworking that arises when software was built fast and without room for change. Like financial debt, it lets you move faster today, but with interest: every next change costs more and takes longer.

In plain words

Imagine a warehouse where goods were put "wherever they fit". For a couple of weeks everything is found from memory. A year later even the person who set it up spends an hour looking for an item, a new worker takes a month to train, and any change to the layout breaks something nearby. Software is the same: the code works, but it is hard to find your way around, and any change is risky.

Debt does not come only from laziness. It appears when:

  • nobody understands how everything is built (no architecture, no explanations, no tests);
  • the same code is copied in ten places, and fixing one leaves the rest untouched;
  • quick patches are layered on top of each other;
  • dependencies and libraries are not updated.

Why AI code and chat without an engineer speed up debt

When people without engineering experience "assemble a store in a chat with AI", conditions arise in which debt grows fast:

  1. No plan. Each request in the chat solves a local task, and no one can assess how it fits with the rest.
  2. No review. The result is accepted if it "works on screen". Mistakes in security, payments and handling of rare cases are not visible on a screen.
  3. No understanding. When something breaks, the owner cannot say where or why, and asks the AI to fix it, often adding a new patch on top of the old one.

This does not mean AI writes worse than a person at everything. But 2025–2026 data show patterns worth knowing.

What the data say

Security. Veracode tested more than 100 language models on tasks in four programming languages: in 45% of cases the code had vulnerabilities from the OWASP Top 10 list. Newer models wrote working code better, but not more secure code (Veracode, July 2025). In October 2026 Veracode noted the first significant improvement: the overall pass rate rose from 56% to 62%, but the weakest models remain at 44–50%, and for some types of problems (encryption, SQL injection) the report says the figures did not change (Veracode, October 2026). These are results on test tasks, not measurements of real stores.

Problems in real repositories. The academic study "Debt Behind the AI Boom" (arXiv, March 2026, latest version October 2026) analyzed 302.6 thousand AI-created commits in 6,299 GitHub repositories: for five AI assistants, 7.5% to 14.8% of commits introduced at least one issue; 484,366 issues were found in total, 89.3% of them "code smells", that is, signs of poor structure; 22.7% of those issues remain in the latest version, which the authors take as evidence of long-term maintenance costs (arXiv:2603.28592). This is about code in general, not stores, and it is not proven that a person would have written better.

Duplication and the decline of refactoring. GitClear and GitKraken analyzed 623 million code changes from 2023–2026: 81% more duplicated code than before AI, 70% fewer commits that rework existing code, and code that "swallows" errors instead of fixing them up 47%. Roughly a quarter of all commits are now created with AI involvement (LeadDev, July 2026). Earlier, GitClear analyzed 211 million lines from 2020–2024: the share of copied code rose from 8.3% to 12.3%, while the share of moved code (a sign of reuse) fell from 25% to under 10% (GitClear). These data show a trend that coincided with the spread of AI assistants, but do not prove that every duplicate was created by AI.

Quality compared with humans. CodeRabbit (December 2025, 470 open pull requests: 320 with AI involvement and 150 human-only): AI-involved requests had an average of 10.83 issues per request against 6.45 for human ones (about 1.7 times), and up to 2.74 times more security issues. The authors themselves note that the split into "AI" and "human" may be imprecise (CodeRabbit, 2025). This is one report from a company that sells a code review tool.

Teams and processes. According to the DORA 2025 report (almost 5,000 professionals), AI is now linked to higher delivery throughput, unlike the previous year, and AI itself is called "a mirror and a multiplier": in well-run organizations it amplifies effectiveness, in fragmented ones it exposes weaknesses (Google, DORA 2025 overview). So without practices, processes and review, AI does not ensure quality by itself.

Developers' own trust. In the Stack Overflow Developer Survey 2025, 66% named AI answers that are "almost right, but not quite" as a problem, and 45% said debugging AI code takes more time (Stack Overflow, 2025).

There are no separate fresh data specifically on stores built in a chat by people without engineering experience. All the studies above are about code in general.

How debt shows up in money

The link between the figures above and a store's budget is logical, but it is an inference, not a measured fact for your project:

  • Security. Vulnerable code is a risk of leaking buyers' data. A leak means not only fines and compensation but a loss of trust. See trust signals.
  • Payments. An error in payment handling, repeated charges, orders that never reach the CRM, or a wrong amount with cash on delivery. These are direct losses and extra load on support.
  • Site speed. Accumulated patches and duplicates slow pages down, which affects conversion (conversion).
  • Cost of changes. A new feature that should have cost a day costs a week, because first you have to figure things out. Sometimes it is cheaper to rewrite from scratch.
  • Dependence on one chat or one person. When the context is lost, there is nothing to restore the logic from.

An illustrative example (made-up numbers). A store was "assembled" in three weeks. Adding a new delivery method (say, a separate Nova Poshta rate) first took half a day, and a year later took six days, because the delivery rule is duplicated in four places and the developer had to find them all.

How to keep debt under control

  1. Record requirements in writing (technical specification).
  2. Have a specialist check security and payments before launch. An independent technical audit helps find weak spots.
  3. Keep the code in a repository with a change history and backups.
  4. For payments and buyers' data, use proven ready-made solutions, not code from a chat (SaaS or custom development).
  5. Set aside regular time for cleanup, not only for new features.
  6. Veracode advises combining AI tools with independent security review rather than relying on the most expensive model (Veracode, 2026).

Common mistakes

  • Counting the cost of development only for the first launch.
  • Thinking AI will "fix everything itself" in the next request.
  • Having no one who can read and explain the code.
  • Postponing updates of libraries and the platform.

FAQ

What is technical debt in plain words? The consequence of quick decisions in code: it works now, but every next change costs more.

Why does technical debt build up? From a lack of planning and review, duplicated code, quick patches and outdated dependencies. When code is written in a chat without an understanding of the architecture, these conditions appear faster.

Is it dangerous to build an online store through a chat with AI? For a prototype, no. For a store with payments and buyers' data, the risk is real: studies show vulnerabilities and accumulating problems in AI code, so a specialist must review it. See vibe coding.

How can I tell a store has a lot of technical debt? Simple changes become expensive, mistakes repeat, no one can explain how everything is built, and the site slows down.

How do I pay off technical debt? Gradually: audit, tests, removing duplicates, updating dependencies. Sometimes it is better to rewrite the problem part.

Sources

Want to check this on your store?

AI automation

An AI assistant checks the store every day, suggests fixes and makes them after your «Yes».

About the service →

Share

Related

  • Agentic commerce

    | Article: 5 min read

    Agentic commerce is shopping in which an AI agent acts on a person's behalf: it searches for and compares products, and within the permission it has been given it can also place the order and pay. For a store, this means part of the "visitors" will be…

  • Vibe coding

    | Article: 4 min read

    Vibe coding is building programs where a person describes the task in ordinary language, AI writes the code, and the person accepts the result almost without checking it and goes by whether it "looks like it works", adding new wishes in the chat. For quickly…

  • How an online store team uses AI every day

    | Article: 2 min read

    AI in a store team's daily work is a set of small, repeating tasks in which a language model prepares a draft and a person checks it and decides. The value comes not from separate "smart" features but from simple habits assigned to specific people.

  • AI automation for an online store

    | Article: 5 min read

    AI automation for an online store means handing part of the repetitive work to a language model (ChatGPT, Claude, Gemini and the like): draft descriptions and translations, a first reply to a customer, search through a knowledge base, spreadsheet analysis. It…