25.0626 Lab notes (Gdrive), Git

This video sums up the future of software dev in an AI world. In the future devs will

Contents of this page

1. Future jobs will require "going wider" (as always)
2. The productivity bar has been raised... and many experts themselves dont realize it
3. Why I switched my focus to LLMs (2025 Jan)

In the 1960s NASA engineers used slide rules. Then in the 1970s small calculators (like the TI SR50) appeared, and my classmates and I in middle school were amazed that we could make calculations with such speed that engineers could only dream of just a few years earlier.

AI continues the tradition of new generations of binary machines doing incredible things, freeing real (human) intelligence to do far more interesting and important things. That means more jobs, not less. Strategies to Thrive as AIs get Better - Especially for programmers / Option 4. @13:00 "I've talked on this channel a lot about how I don't think that AI can replace programmers.... I think that in the long term we're going to end up with more software jobs, not less...". @15:55 he talks about "going wide" ... thats what this AI sandbox site is about.

AI excels at tasks that humans don't, and will enable humans (unburdened by cumbersome details) to focus on a much broader range of SW dev. A dev team I was working with a few years ago had a meeting where they discussed the new Git Copilot AI tool. They noted 2 main things

1. Copilot made programming much quicker and easier. This was the future.
2. Copilot might mean the end of THEIR future. All that low-level tech expertise that took them years to acquire was now available to anyone. The meeting focused on this point.

Another example of this is a recent expert blog "How I'd study AI". The author Tim churns out videos at a pretty high rate (he obviously has a team supporting him). I like his channel because he focuses on the big picture (unfortunately he often does not pay attention to getting the details all correct). He focuses on output.

But in this video "How I'd Learn ML/AI FAST If I Had to Start Over" he tells others to first focus on the nitty gritty details of the Python language. Its worth reading the comments in the video post:

• My comment.

• Its a race like this guy says. Dont waste your time on low level stuff.

I recently took a community college class in SW dev. The course was focusing on a lot of lower level tasks. I was able to use AI (Copilot) to speed up those tasks so that I could avoid spending (a lot of) time debugging and more time focusing on the big picture (deployments). Gradually AI (LLMs) became my primary focus.

It was also during the college course (Fall 24) that I was working on the Tech stacks (REST/GQL) section of the sandbox. I started to doubt the value of spending a lot of time to refresh/expand my knowledge of all the details of the various tech stacks, and instead thought maybe I should be focusing more on the AI assistants (ChatGPT/Copilot).

I was amazed at how useful and empowering these AI tools were. In particular they could

1. Answer my questions (rather than me rummaging through beautifully worded and formatted text)
2. Find answers to technical errors and bugs (rather than having to search through StackOverflow)

---

See also (these pages were deleted from the right panel 25.0326)

• Substack post https://substack.com/home/post/p-159905032
• AI concepts demo 25.0312
• Sandbox as a tutorial series

LCM/SONAR: The future of AI?

youtu.be/Le86PMGK2Uk?t=384

For a conceptual overview, see

• Substack post https://substack.com/home/post/p-159905032
  

This page sums up my personal take on AI.

• People are only human.
• AI is only a misnomer for binary logic that only simulates intelligence.
• Together (with good data) they are magic.

LLM (large language models) are basically a digital version of the old library card catalogs. But a vastly larger catalog (with thousands of indexing dimensions) that works at light speed. without AI, neither could I (or it would have taken me literally at least 10 times longer). Everything I talk about on this wiki runs on dumb binary logic that wont run unless all the details are correctly setup. I need info quick, and AI delivers.

AI has no intelligence. Garbage in, garbage out. AI could not have created this wiki.

But AI helps me avoid "Garbage in" from the sorry state of most SW product docs. It used to take days or weeks to sift through doc verbiage to find the answer to a small problem. The docs were written by tech writers who did not use the product with input from dev/marketing. To get answers to questions you have to pay (a lot of) $ and wait for an expert. Now AI answers your questions anytime for free.

---

• 1 The API sandbox uses AI (Copilot, ChatGPT) as much as possible (it just makes sense)
• 2 AI answers specific questions and gives recommendations for my actual code
• 3 The near term emergence of true AI "copilots"

The sandbox has experienced a bit of “mission creep”. Originally it was going to focus on API docs. But then I realized I could build APIs, API-docs, and stacks with the help of AI (Copilot, ChatGPT). And I realized that perhaps AI (LLMs) should be a key focus itself.

Example deployment describes how I set up a very basic demo full stack demo (frontend, backend, database; deployed the FE/BE to Render) with extensive use of Copilot (Gdrive doc #238 describes my setup steps; search “COP[“ for copilot prompts).

I write tech and API docs that show clearly how to use a product (no TDLR). But docs do not

• answer a specific question posed by a reader or
• show exactly how to modify the reader's specific code.

Copilot does. Most interesting was just how enjoyable and fast-paced it is to "chat" with Copilot while coding/debugging. You gradually start to realize you are talking with a (very capable) librarian who can navigate the entire online library of mankind's knowledge. At lightspeed and for a ridiculously low cost.

For example, in doc #238 I ask Copilot "is there yet another way to post?”. Copilot analyzes my code and gives me the code to change/insert.

AI SW copilots will be able to do 2 things in the future:

The copilot will take over low-level SW tasks. You simple describe what you need, and the copilot implements it (accurately).

The man-machine interface will evolve. The copilot will accurately interpret non-typing input (verbal, eye movement, etc).

Below: The latest tech for robot tracking of human input.