A perspective from inside CrossKnowledge
Artificial intelligence has moved from experimentation to daily usage in many software teams. At CrossKnowledge, AI is no longer a side experiment. It is part of our day-to-day development workflow.
This article is not about hype or future promises. It reflects how we actually use AI in practice, where it helps, where it does not, and what we have learned along the way.
How AI adoption started at CrossKnowledge
AI adoption did not start with a company-wide plan or a formal transformation program.
It started in a much more organic way:
- Developers experimenting on their own
- Using AI to unblock themselves
- Asking questions when documentation was missing or unclear
Over time, these individual experiments became shared practices. Today, AI is used across the team as a support tool, not as a replacement for engineering judgment.
How we use AI in our daily development work
Explaining code and understanding legacy systems
One of the most immediate benefits has been code comprehension.
We regularly use AI to:
- Explain unfamiliar or legacy code
- Clarify complex logic
- Understand the intent behind older implementations
This is particularly useful when onboarding new developers, working in large codebases, or revisiting parts of the system that have not been touched for a long time.
Starting projects from scratch
When starting a new project or component, AI helps reduce the initial friction:
- Generating an initial project structure
- Proposing basic architectural skeletons
- Creating boilerplate code
This does not replace design work. It simply helps teams move faster to the point where real decisions start.
First-pass code reviews
AI is also used as a first reviewer, before human review takes place.
We use it to:
- Spot obvious issues
- Suggest improvements
- Highlight potential edge cases
This works well as a pre-filter, but it never replaces peer review. Human review remains essential for correctness, domain knowledge, and long-term maintainability.
Generating documentation
Documentation is often postponed due to time pressure. AI helps us close that gap by:
- Turning rough notes into readable documentation
- Generating first drafts of README files
- Explaining APIs and workflows in clear language
AI gives us a starting point. The final version is always reviewed and adjusted.
Creating unit tests
AI is also used to reduce friction around testing:
- Generating initial unit test structures
- Suggesting common edge cases
- Improving test readability
Tests are always reviewed, but AI helps teams get started faster, especially in existing codebases.
Debugging and problem analysis
For debugging, AI is used as a reasoning aid:
- Interpreting error messages and stack traces
- Suggesting possible causes
- Helping narrow down hypotheses
It rarely provides the final answer, but it often helps engineers focus their investigation earlier.
Tools we use at CrossKnowledge
GitHub Copilot with Claude and GPT models
For development-related use cases, we mainly use GitHub Copilot, backed by Claude and GPT-based language models.
Typical use cases include:
- Code suggestions
- Refactoring ideas
- Test generation
- Code explanations
Copilot integrates directly into the developer workflow, which makes adoption simple and natural.
Microsoft Copilot for non-coding use cases
Outside of pure development, we also use Microsoft Copilot for:
- Writing and summarizing documents
- Structuring ideas
- Supporting communication and documentation tasks
This helps beyond engineering and supports broader collaboration.
What we deliberately do not use AI for
Despite daily usage, clear boundaries remain.
We do not rely on AI for:
- Final architectural decisions
- Security-sensitive implementations without review
- Business logic without domain validation
AI is treated as a support tool, not a source of authority.
Impact we have observed
The biggest impact is not writing more code, but:
- Faster understanding
- Reduced cognitive load
- Quicker exploration of solutions
- Better starting points for discussion
At the same time, we have learned that blind trust leads to inconsistent quality. Review practices and engineering discipline remain essential.
AI tends to amplify existing habits, whether they are good or bad.
Management perspective
From a management point of view, AI does not change the fundamentals:
- Ownership still matters
- Code reviews are still required
- Quality standards still apply
What changes is the pace. Teams can explore more options faster, which makes clear boundaries and expectations even more important.
Conclusion
At CrossKnowledge, AI is used as a practical accelerator that is deeply integrated into daily workflows while remaining under human control.
Used well, it reduces friction, improves understanding, and supports better decisions. Used poorly, it simply speeds up mistakes.
As with any tool, the real value of AI comes from how thoughtfully teams choose to use it.