Namraj Pudasaini
May 17, 2026
Most coding tasks are done with Claude code or Codex and other AI agents. The agentic world is full of skills, prompts, automation and llm magic. The number of companies offering agentic operating systems and business chatbots are increasing. The future is AI taking most of the jobs not only coding. That is all seen and felt across industries at large.
Every week new releases of emerging models and every month new versions of existing agentic models is hard to catch up. To facilitate the learners, substack, tldr.tech or hundreds of newsletters are coming into existence on a monthly subscription basis. The race is on between the big labs.
The prediction is rise of AI not its disappearance. Merging of many labs are happening and people are invested with time and money in this new field.
Here is what actually changed. Before AI, an idea died in the notebook because starting it was too much work. Now it gets a repo, a README, a landing page and a database schema before lunch, and then it dies anyway. The friction that used to kill bad ideas early has moved to the end, where it costs more.
Look at your own machine. Count the folders with a node_modules in them and a last commit from a weekend three months ago. Every one of them got further than it would have in 2020. None of them shipped. Writing the code was never the hard part of finishing software. Deciding what it should do, keeping it working, and still caring about it after the novelty wears off is the hard part, and no model does that for you.
It is genuinely good at work you can describe completely. Write this function. Convert this to TypeScript. Explain this stack trace. Write the tests I was going to skip. That is a real slice of the job and it is gone.
It stalls elsewhere. Drop it into a large codebase it has never seen and it will confidently copy a pattern that was deprecated two years ago, because that pattern is still sitting in the repo. Give it an ambiguous requirement and it will not push back — it picks an interpretation and builds it, and you find out which one three days later. Ask it what to build and you get the average of everything anyone has already built.
Less typing, more deciding. More time reading code you did not write and reviewing changes you did not make. The person who can hold the whole system in their head and tell good output from merely plausible output ends up doing more, not less.
The only 0.01 percent of people using AI heavily and rest of the AI users are between 70 percent who dabble with it occasionally. The productive nature of AI will produce more automation, but intelligence of human is always required.