It's not happening yet. There's accelerating growth due to increased interest and understanding from humans seeing what this stuff can do, but the exponential growth associated with a true singularity will come from the AI being capable of improving itself much better than humans can. The AI improves itself, gets better at improving itself due to the improvements, improves itself even more, and so on recursively.

The capability of AI in computer programming right now is impressive, but it's not at the level of understanding really complex programs (like itself) well enough to debug them, let alone reason about how to improve them. AI is scary good at one-off programming puzzles that are easily to fully and briefly specify, but that's a very different task from understanding how all the parts of a large, complex program work together and coming up with novel ideas to rearrange them to serve some over-arching goal.

I think some of the recursive self-improvement will begin with some combination of human and machine intelligence, but right now the AI is really just a time-saver to make human coders somewhat more efficient, rather than something that greatly expands their capabilities.