Skip to content

AI assisted coding, in the hands of a pro

Published: September 19, 2025

What happens when a true builder tries out our (corporate approved) GitHub Copilot Agent mode, with Claude?

The scenario: The developer here (his name is Tarek, see below GitHub) is one of the key persons who built our open source on-premise analytics stack. He’s also a contributor (and speaker) in the Kubeflow community. He knows the bits and pieces of what it takes to put all this together. Now, during his Summer vacation he (of course) tried out the IDE above, to see how useful it might be. It was.

If you’re not super good at coding

AI assisted coding is really a multiplier, and it goes both ways. If you’re on ‘minus’ it will multiply that too. Personally (not very much on the plus side) I definitely see good value of starting with the context detailing, then start with that context.md sitting there alone in the IDE before inviting my AI buddy (let’s call the buddy ‘Bob’).

Context, Specs

Along the way, I ask to update the context file, not losing the ‘code’ as we (mostly Bob) is building. If you have 20 minutes, I recommend Sean Grove’s talk ‘The New Code’. Specs is a great starting point when setting the goals with the end user.

The output of an accelerated builder

So, what am I building? Basic things, tools that help me handle the tsunami of information, to keep me on top (trying to, very exhausting) of the AI space (where most people only talk about GenAI, calling it AI… let’s not discuss that for now). I also use it to test out new techniques and technologies in the AI space, to better understand what is what, how it would work in practice. I’m also using it to test out new pieces of architecture (that’s my job, architect). Basically continue with my testing things, so I can answer more detailed questions when showing (typically) powerpoints of what we built and are going towards. I do find the idea of shifting towards ‘coding then writing’ when it comes to architecting very interesting (even if not always possible, but partly).

Back to the builder, you’ll be amazed

So, what did happen when one of our great builders tried out AI assisted coding? He build a complete analytics stack…! He's very modest, so he calls it a 'DataLab' Playground. It's more, and it will evolve into something really useful, if you ask me. I tested it out, and it worked like a charm, and it will be very useful for my work.

alt text

  • Phoenix AI observability — check (extra happy about that one, we do need it for anything or we'll just 'launch and pray')
  • Ollama LLM Server, Trino, Spark, metastore, object store (MinIO), vector database All out of the box, with notebooks guiding you the way! It runs nicely on my PC (had to borrow one from the company, I’m on Mac…).

It does give a hint of the multiplier effect of AI assisted coding. Probably also about vibe coding (not discussing it here, it’s great fun, not sure anyone would to review my vibed code though).

And it is just the beginning.

Where to find Tarek’s creation: https://github.com/tarekabouzeid/data-lab-playground