[simonw]

I like these definitions:

- AI Engineer: an engineer who builds software that makes use of LLMs and other AI models, and maybe trains models (but not required)

- Agentic Engineer: an engineer who makes use of AI tools like coding agents when writing software.

AI Engineer was quite well established in the last few years to that first meaning, mainly thanks to swyx in 2023: https://www.latent.space/p/ai-engineer - which then lead to the popular AI Engineer Summit / World's Fair series of events https://www.ai.engineer/

But this year coding agents have become much more widely spread (the category didn't exist when AI Engineer was coined in 2023), so there's a possibility the term is being redefined to describe people who use those. I think that's a bad redefinition, personally.

("Agentic Engineer" is much less widely used, there may be other names for that category of engineer that I've not encountered yet.)

[janalsncm]

AI engineer: makes API calls to a hosted LLM.

ML engineer: builds models and deploys them.

Hosted models have eaten a lot of the domain of ML but the difference is pretty clear in industries like recommendation, where LLMs are slower, less accurate, and cannot be personalized, not to mention orders of magnitude more expensive.

Agentic engineer would be someone who builds agents not just someone who uses them. Anyone can use Claude code.

[simonw]

"Anyone can use Claude code"

I'll happily push back against that. Using Claude Code and similar tools effectively is way, way harder than people expect.

Anyone can pick up a guitar and strum the strings, but it takes a whole lot of work to actually get good with it.

[Cherryontop11]

> I'll happily push back against that. Using Claude Code and similar tools effectively is way, way harder than people expect.

No it's actually not that difficult. Even if you are not a "prompt engineer" you will get out what is needed from claude if you use simple logic (no need to be technical). In fact this whole "Prompt Engineering" scam will go extinct soon, as SOTA LLM's already catching up on most edge cases and prompting issues.

> Anyone can pick up a guitar and strum the strings, but it takes a whole lot of work to actually get good with it.

Really the wrong analogy here. You are comparing a human who learned guitar through years of practice. When with AI who itself already knows how to play. You're not learning an instrument. You're more like someone who just hired a virtuoso musician. Yes, you need to tell them what song you want,but you don't need to teach them chord progressions. The model already knows the hard part.

[seattle_spring]

Considering 99% of engineers are using AI tools, that would mean all engineers are now "Agentic Engineers." Are we really no longer putting any value on someone who has expertise in understanding the code it produces?

[simonw]

Right, that's why I don't think that class particularly needs a name. It's trending towards "software engineers" now.

We need a name for engineers who don't use coding agents.