Developing software with AI

The way software is developed is changing noticeably. What used to take months can now often be done in weeks. Not because requirements have suddenly become simpler, but because modern AI tools relieve developers of routine work.

This is particularly relevant for SMEs because it shifts an old dilemma: standard software rarely fits exactly, while custom software has long been considered too expensive or too slow. With AI, we can now develop software even faster, more efficiently, and more cost-effectively, covering all of your company’s processes.

Having software developed with AI means that an experienced team develops your solution and uses AI in a targeted manner to achieve a stable result more quickly. AI primarily takes on routine tasks, leaving more time for reviews, tests, integrations, and clean architecture.

Important distinction

Many people mean two different things when they say “developing with AI”:

1. AI as a developmenttool
   AI assists in building the software: code, testing, documentation, analysis, debugging.
2. AI as a function in thesoftware
   AI is part of the solution: chat, search, classification, forecasts, knowledge systems.

This page focuses on point 1. If you want to integrate AI functions into your software, that is a separate topic because data, models, evaluation, and operation work differently.

Many SMEs work with standard software that is essentially well suited to their needs. Around 80 percent of processes are covered.

However, the remaining 20 percent are often precisely the areas that prove costly in everyday business: special cases, approvals, exceptions, missing interfaces, data comparisons. Instead of the process running smoothly, workarounds are created . Excel spreadsheets, manual checks, duplicate data entry, emails for approval.

This is exactly where AI-supported development pays off.
Not because AI “magically” solves the process, but because this 20 percent cannow be implemented faster and more professionally than before.

AI speeds up routine work in development: initial drafts, variants, tests, documentation, analysis. This allows you to reach executable intermediate results faster, feedback flows in earlier, and you build in a more targeted manner instead of in long loops.

Clear signals from practice

• You have recurring processes, but too many special cases for standard software
• You lose time due to manual transfers, comparisons, and approvals
• You need integrations that are missing or unreliable
• You cannot evaluate data properly because it is scattered or contradictory
• Changes take too long because you are dependent on providers or individuals
• Growth makes processes chaotic rather than stable

Develop faster because the basic framework is quickly in place

When it comes to new developments, AI is particularly helpful in areas where many components are standard anyway. These 80 percent often consist of building blocks that are repeated in projects: users and roles, CRUD logic, validations, standard workflows, interface patterns, logging, error handling, tests, documentation.

We have often built this in a similar form. AI does not need to “understand” new processes, but rather supports us in implementingthese familiar patterns more quickly and cleanly.

Within a few hours, a robust basic framework is created on which we can build. The result: we spend less time on routine tasks and invest earlier in what is truly specific, namely your business logic, your special cases, and clean integration into your system landscape.

Where AI really helps in everyday life

AI is particularly strong at tasks that take up a lot of time but offer little differentiation.

• Faster start: Requirements are turned into designs and prototypes more quickly
• Routine code: recurring patterns and basic frameworks are created faster
• Tests: Test cases, edge cases, and safeguards come earlier
• Troubleshooting: logs, error patterns, and possible causes are classified faster
• Documentation: technical explanations are created in parallel with implementation

What AI does not do

Technical logic, responsibility, architectural decisions, approvals, and priorities remain with the team. If these points are not clear, things will slow down, with or without AI.

Step 1: Define the goal and process

Output: Goals, user roles, priorities, success criteria, process outline
AI benefit: Requirements can be structured more quickly, leaving less room for interpretation.

Step 2: Understand data and special cases

Output: Data sources, exceptions, risks, integration points, initial test cases
AI benefit: Special case logic becomes visible more quickly, for example, via scenarios and edge cases.

Step 3: Solution outline plus cost estimate

Output: Architecture idea, integration plan, rough roadmap, risks, and assumptions.
AI benefit: Variants can be compared more quickly without having to implement everything.

Step 4: Short cycles with early interim results

Output: executable increments, regular feedback, updated backlog.
AI benefit: prototypes and partial implementations are created faster, feedback comes earlier.

Step 5: Ensure quality

Output: Tests, reviews, security checks, documentation, acceptance per increment.
AI benefit: Tests and documentation are created earlier and maintained more consistently.

Step 6: Go live, handover, operation

Output: Release process, monitoring, runbooks, ownership, further development plan.
AI benefits: System overviews, runbooks, and handover documents are more complete.

What you need to deliver internally to keep things moving quickly

• Process knowledge and decisions on priorities
• Access to data sources and clear terminology
• Availability for feedback and approvals per cycle

AI doesn’t just bring speed to your project. The greatest benefit comes when decisions can be made earlier and quality remains predictable. For SMEs, what ultimately counts is not that something is built quickly, but that it fits quickly, runs smoothly, and can be further developed. This is exactly where these four effects come into play.

1. Early clarity instead of late surprises

You can see earlier on whether the solution really fits your processes. This reduces the risk of spending months building in the wrong direction.

2. Less looping, less rework

When routine work is done faster, there is more time for reviews, tests, and clean integrations. This significantly reduces the need for corrections later on.

3. More quality per hour invested

When the basic framework is in place faster, the team can put more energy into data logic, architecture, and interfaces. That’s exactly where it’s decided whether it will run stably.

4. Better handover and less dependence on individuals

Structure and documentation are created earlier and more consistently. Knowledge becomes visible, handovers become easier, and dependencies decrease.

Quality assurance and maintainability

Artificial intelligence can accelerate implementation. However, the result only becomes stable and maintainable through defined standards, reviews, tests, and documentation.

Checklist: How to keep the code clean

• Coding standards and clear structure for each module
• Reviews for every change, no direct march through
• Testing strategy: unit, integration, end-to-end, depending on risk
• Definition of Done: Tests, documentation, monitoring, and security checks are mandatory
• Dependencies and libraries are consciously maintained
• Secrets management, logging, and monitoring are included from the outset
• Documentation is part of everyday work, not a final project
• Ownership is clearly defined, internally and externally

Data, security, tools, and compliance

Your data remains your data. In practice, a clear gradation helps:

• Where possible, we use anonymized or synthetic data
• For testing, we use test data or masking
• When real data is necessary, clear rules apply: who processes what, where, and for what purpose
• Access is documented in a traceable manner

Source code and ownership

You receive the source code and clear rights of use. This also includes transparency about the libraries and dependencies used, so that you don’t slip into hidden commitments.

Costs and time – what is realistic

Many people want a figure first. It makes sense once it is clear how complex the logic, data, and integrations are. Here are some possible options:

1. Clearly defined module with integration: often in the low to mid five-digit range
2. Central process solution with roles, interfaces, and data logic: often in the mid to high five-digit range
3. Larger platform with multiple process chains, rights model, reporting, and operation: often in the six-figure range

The effort and budget depend on this

• Data quality and data access
• Number of systems that need to be connected
• Rights, roles, audit requirements
• Special cases and exceptions in the logic
• Requirements for availability, operation, monitoring, and security

What effect does AI have on time and costs? 

AI does not automatically reduce complexity. Above all, it reduces the time needed for routine tasks and speeds up iteration. The biggest lever remains: clear priorities, clean interfaces, consistent quality assurance. With clear demarcation, early feedback is often possible within a few weeks, for example in the form of a prototype. The exact timing depends on how quickly decisions, data, and access are available.

Operation, maintenance, further development

Software is not a one-time project. SMEs benefit most when operation and further development are considered from the outset.

What this entails

• Clear release process
• Monitoring and alerts that help rather than annoy
• Documentation that can be used in everyday life
• Backlog and prioritization for further development
• Clear responsibilities for operation and changes

AI can help with analysis, documentation, and testing. Operational reliability is achieved through standards, responsibilities, and practiced procedures.