this post was submitted on 25 Feb 2026
48 points (76.1% liked)
Programming
25780 readers
384 users here now
Welcome to the main community in programming.dev! Feel free to post anything relating to programming here!
Cross posting is strongly encouraged in the instance. If you feel your post or another person's post makes sense in another community cross post into it.
Hope you enjoy the instance!
Rules
Rules
- Follow the programming.dev instance rules
- Keep content related to programming in some way
- If you're posting long videos try to add in some form of tldr for those who don't want to watch videos
Wormhole
Follow the wormhole through a path of communities !webdev@programming.dev
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
It's not that I've never tried it, I've dabbled in it consistently over the last few years. If you had said there was a major difference compared to 2 years or maybe even a year ago, sure. In the last 4 months, I guess we've gotten stuff like Claude 4.6, which saw an increase in coding performance by 2.5% according to SWE benchmarks. An improvement, sure, but certainly not an exponential one and not one which will fix the fundamental weaknesses of AI coding. Maybe I'm out of the loop though, so I'm curious, what are those exponential improvements you've seen over the last 4 months? Any concrete models or tools?
I only ever started using and evaluating LLM agents this past December, but in my experience it kinda works now. I can't ascertain exactly why and how, but I was taken away how a simple prompt could get workable results.
I decided to try Qwen 3.5 Plus via Qwen Code CLI (Gemini CLI fork) and it's bizarre what it can do.
It can figure out when it's struggling to something, look on the internet for questions and docs to understand things better. It takes a lot of actions by itself, not like that bad models from 4 months ago that gets stuck on endless thinking and tweaking and never fix anything.
Recent models are thinking each time more like human programmers.
I think you're mistaking improvements in tooling as improvements in the LLMs. LLMs are plateauing. The idea of exponential growth is an illusion. We took 20+ year old technology, geared it toward text (the LLM), and trained it on the entire Internet. Then, it's popularity grew exponentially.
This is the hype narrative that Altman, Dario, Jensen, etc. push. They are trying to convince everyone that what we have is Model T Ford of AI. Just imagine where we'll be in 6 months!