[oncallthrow]

My take: learning how to do LLM-assisted coding at a basic level gets you 80% of the returns, and takes about 30 minutes. It's a complete no-brainer.

Learning all of the advanced multi-agent worklows etc. etc... Maybe that gets you an extra 20%, but it costs a lot more time, and is more likely to change over time anyway. So maybe not very good ROI.

[__MatrixMan__]

It seems like you're mostly focused on the tooling for actually directing the LLM but there's a whole host of other technology which becomes relevant re: building guardrails and handcuffs for your agent. For instance I've been doing a lot of contract testing lately. It's not new tech, not changing at a blistering pace, but now that generating mountains of code is cheap, techniques for dealing with those mountains are suddenly more necessary.

[theshrike79]

1. Basic vanilla LLM Agentic coding

2. Build tools for the LLM, ones that are easy to use and don't spam stuff. Like give it tools to run tests that only return "Tests OK" if nothing failed, same with builds.

3. Look into /commands and Skills, both seem to be here to stay

Maybe a weekend of messing about and you'll be pretty well off compared to the vast masses who still copy/paste code out of ChatGPT to their editor.