“What can you do with it?” my friends asked me of my TRS-80 Model 3.
And again with my first IBM desktop, this time from Dad.
And many people asked when I started showing them how to get onto the internet.
Gopher was so cool, you could find any subject so fast.
But there are nuances when the question is asked about our current crop of LLMs and Diffusion image-gen models.
There’s “What can it do…?” Right Now and then there’s “What can it do…?” in 6 months time.
3 months ago Geoffrey Huntley asked Claude in a permanent Do-While loop “Hey, can you make me a programming language like Golang but all the lexical keywords are swapped so they're Gen Z slang?” And today that compiler now works, generating operating programs from Gen-Z Rizz and Pop.
It probably won’t take as long if you want to try something like it.
Let’s get personal with this question.
Back in May Mom injured her right hip, revealing extensive osteoarthritis according to the doctor, technician and X-Ray machine. I went to Grok and asked what home remedies, treatments or other options were available that fit within mom’s medical profile (things like blood thinners and diuretics complicate a lot of medical options).
So far the 2 most useful allowed things have been collagen peptides and TENS treatment of the hip area. TENS has been especially helpful. The steroid shot the doctor gave her, after making her wait an entire month but before retiring and leaving us looking for a new doctor, helped a lot too.
The medical industry is in trouble.
But what would happen if I poured 100$ worth of credits into OpenRouter. Setup my own agentic loop with websearch ability calling the new medical model, MedGemma from Google. Then gave it mom’s entire medical profile and asked it to search exhaustively through all the world’s pharmacopeia, herbal knowledge, medical, surgical, mechanical, electrical and ultrasound possibilities along with anything else it found and give me a prioritized list of options and treatments? Focusing on the easiest, cheapest and most innovative first.
We couldn’t get into the orthopedic surgeon about options or a hip replacement until mid-October.
I was really surprised with the discussion I had a few days ago with a classmate who is a far more professional, knowledgeable and experienced software engineer than I when he told me that he hadn’t spent much time with the new Agentic coding assistants. He had heard they were buggy and error prone so hadn’t bothered. I talked him into giving them a chance and he almost immediately solved a clients problem with it. But he had to patch a bug too.
The company Blitzy is converting millions of lines of very old legacy code into memory safe and/or more efficient modern code in languages like R, Rust, Python &etc. And I’m only about a week out from achieving release goals with my CodeFactory package which has some very unique code generation modes that no other IDE or Agentic CLI coder has.
World models from DeepMind, image generation so thick you can’t even go an hour online without seeing some example, protein folding, protein engineering, pharmaceutical research and drug discovery. Self-driving cars and of course, don’t forget the robots. There are many open source and DIY examples of rapidly improving robot projects and that’s ignoring the dozen or so big startup and corporate manufacturers. With the cherry on top being Optimus from Tesla.
What A.I. can do, it’s already doing and it’s probably going to do a lot more. Everybody hang on, the ride is just getting started.