Our approach to AI at typdy.
At typdy, we believe great software is built by people. AI is a tool we approach with curiosity, but also with a healthy dose of scepticism.
Photo by Serge Taeymans
We're Sceptical
We are pretty sceptical about AI at typdy. LLMs are powerful models capable of really impressive things, but they are still prone to error, and we suspect they always will be.
The problem, as we see it, is that LLMs do not and can not reason. At least not in the traditional sense of the word. Their 'reasoning' is a lot like looking at lots of examples and picking the parts that fit the best, rather than any deep understanding of the task at hand. Can this work? Sure, but it can be unpredictable.
That's not to say it's useless though, just potentially dangerous when handled inappropriately.
Our Current Policy
Given our concerns, we do not currently use AI to write any of the code behind the typdy Headless CMS. That is, beyond the kind of basic autocomplete that has been a mainstay in IDEs for a very long time.
Outside of code, our use of AI is largely limited to early-stage idea exploration and inspiration. For us, it's about taking the best from AI without delegating every thought to it (which some seem to advocate).
The Future
Should we start using AI significantly as part of our development process in the future, there will be strict rules and limitations. This will include an additional multi-step review process to ensure all AI code is reviewed by more than one person.
We would also require developers using AI to take responsibility for their use of AI, and be held accountable for egregious errors. We will never force a developer on our team to use AI or meet any kind of ridiculous token quotas.
