When we don't use AI in software development

By Max Ikaheimo

October 1st, 2026

We use AI coding tools every day, and we've written about how we use Claude Code in production. Being clear about where we don't use AI is just as important. It's the difference between using a tool and being used by it.

High-fidelity, detail-heavy work

Some work is all about the details: pixel-precise interfaces, complex animations and interactions where the exact feel is the deliverable. AI tools get these close quickly, but close isn't the goal. Fixing the last details in generated code often takes longer than writing it carefully from the start, and the result is harder to maintain.

For this kind of work, our developers write the code themselves.

Security-critical code

Authentication, authorization and access control, payments and anything that protects sensitive data share one property: a subtle mistake can be expensive, and it may not show up in testing. Generated code tends to look correct, which makes subtle security flaws easy to miss in review.

Code like this is written and reviewed by our developers rather than generated, and it gets the strictest review in the project.

Personal and confidential data

Under GDPR, processing personal data requires a legal basis, and a service that processes it on your behalf acts as a data processor that needs an agreement covering that processing. Pasting customer data into an AI tool doesn't remove those obligations.

Our rules are simple:

• We agree up front which data may be processed by AI services and on what terms.

• Personal and confidential data stay out of prompts unless there's a legal basis and an agreement covering it.

• We use commercial AI services whose terms exclude training on customer data.

• Test and development data is kept to what the work actually needs.

When the client says no

Some organizations don't allow AI tools on their projects at all, for regulatory, contractual or policy reasons. That's a legitimate choice. If your policy rules out AI tools for a project, we don't use them.

How we decide

For any task, we ask three questions:

• Can the output be verified properly? If a developer can't review and test the result thoroughly, AI shouldn't write it.

• What does a subtle mistake cost? The higher the cost, the less generated code belongs there.

• Does it involve personal or confidential data? If so, the data agreement decides, not convenience.

Used this way, AI makes development faster where speed is safe, and stays out of the places where it isn't. Read more about our AI development services.