Agent Usage
"The Most versatile substance on the planet and they used it to make a Frisbee" - Ultron
Agents have highly democratized the ability to code. This has enabled various levels of programmers and non-programmers the ability to write easy and complex code. Alas, the agents still need to be told what to code, which is why despite the democratization, there's a flood of CRUD apps and other non-imaginative software. People are given the power to wield agents as good as Codex or Claude Code, and yet cannot fathom softwares beyond simple CRUD apps, implementing yet another Uber, Excel, n8n or ironically, an agent.
There's a caveat to the point above: iterative refinement of old software is always beneficial in the long run: after all, Swiggy entered the market years after Zomato did, Rapido/Ola after Uber, and yet are competing in the markets! It is Ok to compete. This was, however, before the agents era, and in order to build the software which could compete with giants in their own domains, one had to climb the mountain of programming the app by-hand, something which was both costly and time-consuming.
Now ? Writing software is neither time-consuming nor costly. Refinement of older software was earned through experience and business insights, which is no longer a hurdle to be climbed: software is cheaper then ever to make. This has inevitably led to an acceleration of a trend which has been ongoing since probably post-COVID: abundance of "mediocre" software, which is not a comment on it's objective correctness, but rather it's business and technical need: no one cares about CRUD apps or it's-uber-but-for-... apps anymore, but people are making it nonetheless.
To not be overly critical : "useless" or "mediocre" software was probably an absolute necessity in order to both learn programming and understand how software fits into various businesses. This created a right of passage for early developers to learn and grow their ability to understand and implement software. It could still be the case: non-programmers now can quickly prototype their business ideas, programmes can try pushing the boundaries of what kind of software they can build, and build a repertoire of meta-understanding about programming.
One question still remains: how ? If software can be easily built, how can one discover the edge of one's abilities, and expand on them ?
This is an open question as of my understanding. In these uncertain yet exciting times, my personal bias tends to favour the act of reading. Reading deeper into computer science, reading deeper into business, and formulating stronger frameworks which can in-turn facilitate better orchestration capabilities regarding agent usage. Nothing stops us from simply asking the LLMs anything we want, so why not tackle harder problems ?
We have a tool with seemingly infinite patience, the knowledge of the entire internet and most importantly, a means to break down walls standing between us and deep knowledge.