Viewpoint on Generative AI

Generative AI in digital service companies: synonymous with productivity gains, myth or reality?

In this article,  Jean-Louis Vila, CTO and Associate Director of Coexya, explores the real impact of artificial intelligence on developer productivity, demystifying preconceived ideas and offering a pragmatic analysis of the benefits and challenges of adopting AI in software development.

Generative AI in digital service companies: synonymous with productivity gains, myth or reality?

That’s the question I’ve had to face repeatedly over the past year as CTO of an IT services company. Although I have a bit of trouble quantifying intellectual services, I’ve given it a try.

Some are predicting the end of developers, others their impoverishment, others wild productivity gains… In short, as with any new technology, when faced with the unknown it provokes surreal reactions, accompanied by sharp marketing and a natural instinct for self-preservation. To take a step back, I’ll start with a simple observation: every time there’s a universal scientific advance (i.e. one that concerns the majority of humans), we naturally think in terms of the world before, whereas quite often there’s an explicit or more subtle paradigm shift. That’s why, in response to the statement “With AI, there will be huge productivity gains and therefore lower production costs“, I simply reply “Yes, you’re right. My answer is simply “Yes, you’re right about productivity, and the same goes for lower costs“. Then, of course, the question “How much do you think it will be? I simply answer “between 5 and 15%” for people who use AI. You can see a look of disappointment on my interviewer’s face.


But I’d like to follow this up with a different line of reasoning, one that is certainly not new, but worth remembering. Generative AI is your virtual assistant, your exo-cortex that :

  • Speaks your language;
  • Has a memory you’ll never have;
  • Re-learns and specialises ;
  • Allows you to “Think out of the box” without taking illegal substances 😉
  • Has no conscience and no state of mind…

In the world of developers, it is a “virtual peer programmer” capable of :

  • Provide contextualised and semantic unit tests that go far beyond simple templating;
  • Generate objectives based on written or oral expression;
  • Suggest;
  • Correct errors/exceptions;
  • Detect potential bugs;
  • Suggest optimisations;
  • Explain;
  • Transcode ;
  • Learn;
  • + a whole host of other things reserved for ingenious people who will save even more time.

To do all this, you need to interact. So it’s easy to understand why I estimate a 5-15% gain in productivity and no more, because this interaction costs … time. QED!

However, once again, other aspects need to be considered:

  • Improving the quality and coverage of unit tests as a minimum;
  • The real ‘presence’ of a peer programmer (this in no way precludes the use of humans 😉.

To date, quantifying the improvement in quality is a tricky business, and I won’t get into it without more information. On the other hand, for the quantification aficionados, if we reason on the basis of the “peer programmer”, here is a projection that I’ll let you appreciate and comment on:

Gain/Savings = (Cost of human peer programmer) – (Cost of virtual peer programmer)

If we assume a beginner engineer in France/province with a production cost of €150/day and 20% of an FTE for the peer programmer, then over a year this represents around 44 days or €6,540 and if we subtract €1,000 for the virtual assistant per year, this gives a saving of €5,540.

For an IT Company with 500 developers, this represents €2,790,000 for one year. I’ll leave you to draw your own conclusions, but hesitating means delaying the inevitable!


AI at the core of our business solutions