ai agents

man... where do we even start. lately ive been working a lot with ai agents and large language models (llms).  let me just say this so it's recorded in history --  ai agents are going to take over user experience for casual internet users, that's without question.

so in the near term future you'll see the internals of many saas and platform applications being replaced with complex graphs and routing strategies, possibly backed by highly fine tuned large language models -- but not a requirement. with llm tooling capability becoming standardised and expected as a fundamental core capability of any llm worth its salt, we'll see these agents explode exponentially replacing legacy deterministic software engineering guts.

for the casual internet user, i highly suspect that someone will come along with something that will replace traditional search engine technology. ranked based links will clearly be a thing of the past, and only valuable to the large language model backing the platform or application you're using. as a user, you won't even realise what is going on. or care either, to be completely honest. how often do you care about how your television works? you don't, you just expect it to work. 

so in time, user experience across the internet will just turn into conversations between bots and apis -- powered and accelerated by the ever decreasing size of small language models, but at the same time have increasing context lengths. 

in other words, experience itself will turn into the output of a complex ai backed data pipelines, whether you realise that or not.

if ai agents interest you, i highly encourage you to take a look at things like langgraph, from langchain.  it offers a highly customizable sdk to make complex ai agents. its graph patterns are based off of apache beam, so if you have worked with data pipelines, you should pick up on the nodes and conditional concepts quickly.

ai agents will take over technology positions as we know it. it's an inevitability that positions like data engineer and database administrator -- roles like these will disappear to the dust heaps of time and replaced with automated ai agents.  but this just results in new jobs, jobs like creating these graphs, maintaining them, innovating with them, debugging and qa'ing them.  these should also just be seen as temporary roles, only until they're automated, which is also an inevitability.

we'll see this cycle rinse and repeat over time i suspect. clearly there will still be jobs available, but they'll certainly change in skill sets. much more emphasis will go into those who know how to prompt (if you were a philosophy major who studied logic, you are finally going to be sought after!!). and not every company will change over night, so this will be an incremental thing. in fact i have absolutely no doubt that there will be entire companies built whose entire story is ai backed automation to migrate your legacy code base into ai agent controlled micro services. i havent checked yet, but i suppose there are probably companies that exist already that advertise this sort of thing.

the future is going to be interesting...