Having a fun discussion with colleagues on LinkedIn about our approach to coding with concepts. See my recent blog post here. Happy to answer any questions folks may have here, and excited to hear your ideas and reactions.
I remember you saying that LLMs can understand concepts. That was before I was happy about coding with LLMs, but when I did warm up to coding with LLMs, that was one of my guiding principles.
Concept design and formal methods completely opened up AI coding for me. I see the people micro-managing all these tasks (coding-related or not, could be film editing, document preparation etc.), and for me, it’s so much nicer and easier to iterate up front on a formal description of concepts. Agreeing on the ontology, concepts and the state machine description for critical abstraction levels and components of the system eliminates a huge class of defects and waste.
And once the LLM knows the concepts, it can build everything with minimum oversight. You can walk away and come back to a high-quality v1 that actually does what you want it to.
I’m preaching to the choir here, everyone here knows much more about this stuff than I do. I’m just reporting-in, as a blue-collar programmer carrying around a copy of Using Z, that out here in the real-world LLMs + concept design is definitely the killer combo.
2 Likes
This is great to hear, @tom . Do feel free to share any details.
We plan to release our own harness for LLM powered concept design coding very soon…