45 Agentic programming
Agents write code, run code, look at the results, update the code and keep going until they decide to finish the task.
It is possible to do agentic coding with local LLMs, but you will need a really good consumer grade computer that can handle much bigger models than used for autocomplete or chatting.
See this post by Simon Couch on driving coding agents on a laptop using local models such as variants of Qwen 3.5 and Gemma 4.
Qwen 3.8 (27B) is also getting great reviews for agentic coding. Simon Willison has found it works well on both a 128GB M5 Max MacBook Pro and an NVIDIA DGX Spark — both high-end machines costing several thousand dollars, so check current prices if you are considering this route.
The good news is that clever people are finding new ways to compress and use these LLMs such that they run faster on smaller computers.
Note there are some cybersecurity issues with agents because they run code on your computer. You want to be careful that you know what you’re doing before you start using agents, and get permission from your work if you’re using them on your work computer.
Look up the RAM you recorded in the Performance and hardware module. Using the model sizes on the Ollama library, work out the largest model you could realistically run for agentic coding on your own machine. If the answer is “none”, what would you need to change — and is a cloud model the more sensible option for your work?