Agents, tasks, tokens, skills, Claude, Cursor, Gemini, Antigravity, OpenCode… Ralph?! If the start of your 2026 wasn't a tidal wave of AI hype, changing currents day to day, hour to hour, maybe I slightly envy you. It feels like a hyper-accelerated version of the 2010's Javascript framework boom on steroids and a touch of meth.

Despite the whiplash, it's incredibly exciting… and inspiring to see people excited about building. Building tools that are turning every engineer into the lauded "10x" engineer when wielded correctly. Updating Claude Code has become a daily/semi-daily ritual because the team at Anthropic is as moving as fast as their tools are letting software professionals everywhere move. And yes, I know I can enable auto-update but the act feels something like a digital sacrament, awaiting perhaps minor bug fixes but maybe, just maybe truly impactful workflow changing features (see yesterday's Tasks release).

The constant change is destabilizing and deliberately so. There is no guarantee that either the model or tool you're using will be the best tool for the job next week or even tomorrow. This might make it difficult for organizations to standardize on one toolset. We don't want to be burning cash for licenses that are redundant and/or unused but we also do not want to stifle the creative energy and gains that are coming from the constant flow of novel models and tools. In my opinion, there will still be some brief time before true vendor lock-in occurs and teams are confident they can go all in on one platform and yet with open source solutions like OpenCode, they may not have to. Regardless, I am relishing this time precisely because it is so uncertain. Even in a leadership role, I am producing features nearly as fast as they can enter my mind which as a scrappy builder with 10+ years in the industry is a true delight.

So it's fun to tinker and experiment but organizations everywhere are now starting to see the completeness in which they can ship fixes, tweaks, features and even entire apps. What exactly does this mean for the software professional? If you're like me, the awe-inspiring one-shot prompt that produced a high-quality solution in a few short minutes that would've taken at minimum a few hours comes with excitement but also dread.

What's my purpose? I spent years crafting well-organized, performant, DRY.. and beautiful code and now I'm going to turn over the reins to, at best a skilled imitator and at worst a slop-generating tech debt machine. I found myself waxing eloquent for a bit, I thought back to the grind of being a non-CS grad self-learning Python, demystifying the bits of logic the simple syntax controlled. I thought of optimizations that tripled runtime performance, mastering concurrent programming with goroutines and even learning the skill of leading other engineers. I love these things. I've made a living doing them, yes but I ultimately find immense pleasure and satisfaction in the act of making computer do things, do useful things and empowering others to do the same.

The good news is that is not changing or will change. Humans still need computers to do useful things and even more so. How we produce useful things is changing. Which leads me to the title of this article, your job is obsolescence. The word choice is purposeful, your job is not obsolete but it is to make the manual, time-consuming parts of your job obsolete and to appropriately delegate them to an agent. Wrangling bits of state in your enterprise React app, updating structs for new fields, migrating to new versions of packages, these are things you can now confidently do with an agent's help.

Keeping yourself keenly aware of the changes being made but not actually having to type the characters on the screen is the new art. And that's a good thing, it keeps you focused on what needs to be built and how it should be built. Your decades and years of experience and studying are more important than ever. Non-technical folk will ship code using agents but much of it will be brittle and painful to maintain. You will produce code just as fast as those people but ensure that the technical guardrails and indispensable principles that make good software good (and secure) are in place. You're eliminating the tedium, not the art of quality software. Ultimately, you are in control of what is produced, the same as it ever was. How you enforce that control might look different but an agent will not replace that.

The levers you pull might look different but they must be skillfully pulled nonetheless. Put simply, you will architect the environment in which the agent thrives.

No AI was used in the production of this article.