I trained developers to use AI in software development for analysis, breaking down work, implementation and review. I have personally worked with tools including Codex, Claude and ChatGPT. The aim was a better workflow: faster progress, careful review and clear developer ownership of the code we delivered.
Training and workflow
I also see a leadership task in making the learning shared. When one developer finds a good way of working, the rest of the team should be able to understand when and why it works. We set aside time every other Friday for learning and knowledge sharing. That gave us a place to discuss both the possibilities and the limitations of the tools we used.
What we observed about time
In one task, an estimate of around 100 hours of work came down to about 15 hours. In our own work, we estimated that AI made development about 50 percent faster in many tasks, including review and checking. This is an experience-based estimate from our team. The 100-to-15-hour example was an unusually strong result, not an average.
Testing, review and accountability
For me, AI does not change the need for experienced specialists. If anything, it places new demands on their judgement. A faster first version is only an improvement if the result fits the task, can be maintained and has been checked. That is why we worked with the whole process, from describing the problem properly to examining the output and fixing what did not hold up.
If I join a new software team, I will start with concrete tasks and clear quality requirements. Then we can try out, measure and adjust the use of AI together with the people who will stand behind the solution. The goal is better work and more room for the developers' judgement, not a big number without an explanation.
Contact: kontakt@tomc.dk · +45 299 297 67