AI-Powered Software Development for SMEs

For many mid-sized companies, custom software was long a luxury: too expensive, too slow, too risky. They made do with off-the-shelf solutions that never quite fit - Excel workarounds, media disruptions, isolated systems. AI is changing that equation. Custom software that fits a process exactly is becoming realistic for more mid-sized companies.

This article explains how AI-powered software development works, what it concretely changes for mid-sized companies - and the limits you should know before you start.

Key takeaway: AI tools mainly speed up prototypes and routine code: in a controlled experiment by GitHub, the group with Copilot finished a clearly defined task 55 percent faster on average. This makes custom software affordable for more projects in mid-sized companies, such as internal tools, integrations between existing systems and customer portals. Requirements, architecture and tests remain a human job, however, and developers check AI output too: in the Stack Overflow Developer Survey 2026, 48 percent of respondents trusted AI output when they could easily verify it, and only 16 percent for most tasks that are not important work decisions. What matters is therefore the combination of AI tools and careful software engineering.

Why does off-the-shelf software hit its limits in SMEs?

Off-the-shelf software covers the usual workflows. The processes that set a company apart from others, however, are often left out. The result:

• Workarounds instead of solutions: data is transferred manually between systems, Excel fills the gaps, knowledge lives in individual heads.
• Media disruptions: CRM, accounting, inventory and website don't talk to each other.
• License costs for unused features, while the one decisive function is missing.

Custom software solves this - but used to be expensive and slow. A classic project took months and tied up budgets many SMEs shied away from. This is exactly where AI comes in.

How is AI changing software development?

AI-powered software development does not mean an AI spits out finished software at the push of a button. It means developers work faster and more precisely with AI tools at every stage:

• Faster prototypes: an idea becomes a working prototype sooner, one you can test before large budgets are committed.
• Less routine code: coding agents handle boilerplate, tests and recurring patterns. In a controlled experiment by GitHub, 95 developers, randomly split into two groups, were asked to write an HTTP server in JavaScript; the group with Copilot finished 55 percent faster on average. That figure applies to a single, clearly defined task, not to entire projects.
• Better maintainability: AI helps with documentation, refactoring and understanding unfamiliar code - an often underestimated cost factor over a software's lifetime.

Models are developing fast, with new generations appearing within a few months of each other. In practice this means one thing above all: the bottleneck shifts from "writing code" to "defining the right problem precisely".

Developers themselves take a sober view. In the Stack Overflow Developer Survey 2025, 84 percent of respondents said they use or plan to use AI tools. At the same time, 46 percent distrusted the accuracy of the output and only 33 percent trusted it, and 66 percent named solutions that are almost right, but not quite, as their biggest frustration. A year later AI is everyday practice: in the 2026 survey, 73 percent of respondents use coding assistants and agents daily. Trust remains conditional: 48 percent trust AI output when they can easily verify it, and only 16 percent for most tasks that are not important work decisions.

What does this concretely mean for SMEs?

The most important change is economic: custom software becomes affordable for more projects. Projects that did not pay off before can pay off today. Typical use cases:

• Internal tools that model a specific process (quote calculation, scheduling, reporting) - instead of ten scattered Excel files.
• Integrations that connect existing systems (ERP ↔ shop ↔ accounting) and end double data entry.
• Customer portals and self-service solutions that relieve support.
• AI features directly in your own process - from document recognition to process automation.

The return usually comes not from the software itself, but from saved time and avoided errors - month after month.

Important: AI-powered does not mean "AI does everything alone"

Honesty is decisive here, because this is exactly where many quick AI projects fail. AI accelerates writing code - but architecture, data security, maintainability and integration decide between success and expensive legacy burdens. Anyone who has something generated "quickly" without experience often builds up technical debt that later costs more than it saved.

What matters:

• Clean architecture instead of copied snippets - so the software can grow.
• GDPR and data security considered from the start, not bolted on afterwards.
• Clear requirements: the AI is only as good as the problem you give it.
• Maintenance and handover: software lives for years - it must be documented and operable.

That is why choosing a development partner should involve not only the question of speed, but also questions about architecture, testing, documentation and handover.

Which legal questions arise in AI-assisted development?

Four, and they belong before the project starts, not at acceptance: who owns the code, which licences it brings along, who is liable and which reporting duties apply. Since autumn 2026, two of them have new rules: the Bundestag passed Germany's new Product Liability Act on 8 October 2026 and the Bundesrat cleared it on 16 October 2026, with application due from 9 December 2026; and the first reporting obligations of the Cyber Resilience Act have applied since 11 September 2026.

The first point surprises many: under Section 69a(3) of the German Copyright Act, a computer program is protected if it is the result of its author's own intellectual creation, and only a human can be an author. Code that an AI generates without a significant human contribution is not protected by copyright according to the prevailing view; who may use it is then governed by the contract alone. In contract development, the rights also do not pass automatically in full, unlike with employed developers under Section 69b.

For liability: only natural persons have a claim under the new Product Liability Act, namely for personal injury, for damage to property not used exclusively for professional purposes and for the loss of data not used professionally. For an internal tool this usually changes little, as long as it cannot endanger anyone physically; for a customer portal or a consumer app it changes a lot. What the reform means in detail is covered in the article on AI liability and the new product liability.

Conclusion

AI doesn't make software development obsolete - it makes parts of it faster and custom software therefore accessible to more SMEs. The decisive lever is not the tool, but the combination of AI tools and careful software engineering. Bring both together and you can implement solutions that would not have paid off before.

For a strategic overview for decision-makers, read the guide AI for managing directors and CEOs.

Frequently Asked Questions (FAQ)

This article was published on 2 June 2026. On 18 October 2026 we added the Stack Overflow Developer Survey 2026 and the legal questions on source code, product liability and the Cyber Resilience Act. It does not constitute legal advice.

---

Sources

• Research: quantifying GitHub Copilot's impact on developer productivity - GitHub
• Stack Overflow Developer Survey 2025: AI - Stack Overflow
• Stack Overflow Developer Survey 2026: AI - Stack Overflow; daily use of coding assistants and agents, trust in AI output
• Section 69a UrhG - subject matter of protection and Section 69b UrhG - authors in employment - Gesetze im Internet
• Section 7 UrhG - author and Section 31 UrhG - grant of rights of use, subsection 5 on the purpose-of-transfer rule
• Bundesrat: Act to modernise product liability law, printed paper 584/26 - software as a product (section 2), entry into force 9 December 2026
• Regulation (EU) 2024/2847 (Cyber Resilience Act) - EUR-Lex; overview of the reporting obligations applying since 11 September 2026: Crowell & Moring