broken window theory: software in the post AGI world

coding is solved, why is everything so bad?

AGI is here, software engineering has been solved, Mythos is too powerful to be released to the masses - and yet

We live in a complete and utter state of permanent software instability, perhaps to adhere with the current affairs of the world, it has grown acceptable for software to be broken, stay messy, as if nobody cares, as if nobody has any responsibility for the product.
Which in hindsight makes sense, once your engineering team embraces the “blameless culture” - as does leadership - it’s game over, the death of excellence, and you become a number in the sheet for the next round of layoffs.

If the craft is not being respected, worked on, perfected, and being automated, the decline will continue. However, this is not an ode to hand crafted code, or the lack of it, no, but rather responsibility and respect for the machine. You don’t have to understand the system you are running, few even before the LLMs did, but you have to understand how to change parts of it, evolve and grow depending on demands.
Now with agents being able to analyze and react as needed it gets easier, yet the best research labs today can’t keep inference running at scale - all this free intelligence and for what?

Perhaps is not the model, it’s the harness, perhaps it’s the agent - who cares, in the end, how many 9s do you have in your SLA?
The reality is that building software, researching ways to multiply matrixes and keeping software running are very different things, with the later being the most challenging now apparently - however I do not want to take any credit out there from my dear webdevs shipping absolute slop to this day and age still, but to be fair when they ask you to change a button that lives in micro-frontends across different platforms controlled by different teams who can really blame you for the broken state of things.

Now, back to staying online, people in 2026 are asking, is it really possible? Honestly as much as I want to blame vibe coders and infra avoiders, very few of the current systems are able to withstand the constant flow of AI agents hitting your systems faster and harder than even before, the software boom and its consequences will feel deep within every CPU cycle across your services. However the servers must keep running, the show must go on, we will figure it out.
Those who have realized this are either panicking trying to come up with new architecture, scale their systems or try to make tools for themselves and their agents to keep the lights on.

I’m in the last category with clankercloud.ai - building an AI ops workspace for you and your clankers.
It’s hard, not gonna lie, there’s a whole new way of thinking about software when you no longer make it for the human mind, but want to keep the human in the loop. Building the tools for intelligent husks better than us at information, well, parts of it.

The new era of products for others, not us, weird - but a reality. Not just products but jobs too. I wonder if we will call them something else.