Ask HN: How to Transform to Agentic Coding?
I'd like to shift our development workflow to an agentic approach, enabling us to use LLM models more efficiently in actual software development. We are a team of very senior experts in a Python-heavy niche industry. Currently, our adoption level is low. We are stuck asking for snippets and help, and have not managed to get agents to build larger modules beyond disposable scripts. When asked to do anything more serious, agents usually mess up the results, and it is just faster to write it by hand.
Because the agentic domain is so new, and most tutorials out there already seem outdated, I'd like to ask experts on the leading forum: what would be a good starting point for converting codebases into a format that agents can work on?
This includes
- Set up for the large codebase needed: where to put files, what hooks to install
- What tools to utilise (as in CLI/SaaS) to boost the intelligence of agents to a meaningful level
- Strengths and weaknesses of different models, and how to pick a model
- What kind of instructions to write, where and in what format
- How to set up the "loop" where the agent can work independently for longer periods of time
- How to utilise multiple agents The only requirement is curiosity. There's no "do this" with these tools. You have to learn them through constant experimentation.