Over the past week and a half, I’ve been working on the plan for an application using Cambigo. I’m doing it to use Cambigo in the “real world” so I can test and fix along the way. At the end, I’m going to see if ChatGPT, Gemini and Claude (“LLMs”) can build the application from the flow document, and also try multiple stacks like Rails, and some combination of frameworks for TypeScript and Python. I have a technique I want to try to see if I can allow the user to focus on a solid plan through Cambigo, and not on building the technology.

How it started.

The other part of this, if you are reading past the fold, is when I started Cambigo, it was to produce something that integrated a lot of work I do writing in notebooks, making diagrams, preparing various documents, evaluating different technologies, and gathering the details for new, or “remodeled,” data and systems. My plan was to help me support more people, allow others to do what I’ve been doing, and to help people build solid plans that can go to the technologists for developing and implementing. Then I started thinking that AI could help take the next step to produce a Proof of Concept. Something that is not ready to go live, but provides a tangible model to try and share with the technologists to convey ideas for a full solution.

The explosive growth of LLMs both horrifies me because of what I see it produce, and excites me because of what it CAN produce.

How it can go.

The horrifying part may not be so surprising. The work that goes into producing reliable, secure solutions requires a lot of people — with depth of expertise in multiple domains. The work done here cannot be skipped. The easy sentence to write is that we should be setting up ways for those domains to be represented in a pipeline for producing or maintaining a system or application with AI. I don’t think, like some, that this should replace people. I think it should give people the opportunity to integrate expertise so that instead of helping with two or three solutions, they can help with 200+.

For a solid decade of my federal career, I was wrangling the work of 4 to 5 people into one workload. I learned a lot about becoming comfortable drowning, and still picking the priority. If I could evaluate ideas, and collaboratively provide feedback that becomes a part of a solid plan for a full implementation, it would be invaluable. It would be even better if something could consistently cover the basics before it makes it to me.

AI eats your bones.

This can keep people working together, rather than racing to replace each other, which is where we seem to be with AI right now. It also requires experts to continue learning and using their own minds in practice — for whatever their field of expertise may be. AI is merely an exoskeleton, and has a purpose for wearing it. We need to be careful how we define that purpose.

Those that lose their bones will not grow them back fast enough to keep up in the future.