BasaltMitchell Hashimoto, the guy who built Vagrant, Terraform, and Vault, posted a thread the other day that’s been rattling around in my head.
I strongly believe there are entire companies right now under heavy AI psychosis and its impossible to have rational conversations about it with them. I can't name any specific people because they include personal friends I deeply respect, but I worry about how this plays out.
— Mitchell Hashimoto (@mitchellh) May 15, 2026
I lived through the great MTBF vs MTTR (mean-time-between-failure vs. mean-time-to-recovery) reckoning of infrastructure during the transition to cloud and cloud automation. All those arguments are rearing their ugly heads again but now its... the whole software development industry (maybe the whole world, really).
It's frightening, because the psychosis folks operate under an almost absolute 'MTTR is all you need' mentality: 'its fine to ship bugs because the agents will fix them so quickly and at a scale humans can't do!' We learned in infrastructure that MTTR is great but you can't yeet resilient systems entirely.
We already learned this lesson once in infrastructure: you can automate yourself into a very resilient catastrophe machine. Systems can appear healthy by local metrics while globally becoming incomprehensible. Bug reports can go down while latent risk explodes.
He talked about “AI psychosis” in companies where rational conversations about risk have become impossible. The whole industry is operating under what he described as a “MTTR is all you need” mentality: ship bugs fast because the agents will fix them faster.
He lived through the MTBF vs MTTR (mean-time-between-failure vs. mean-time-to-recovery) reckoning during the cloud transition. He saw how you can automate yourself into “a very resilient catastrophe machine.” Systems that look healthy by local metrics while globally becoming incomprehensible. Bug reports going down while latent risk explodes.
I think he’s right, and I think the situation is worse than he lets on.
We were already speed running toward a competence crisis before AI entered the picture. The last twenty years of software development have been a steady drift toward abstraction layers that nobody fully understands. Stack Overflow copy-paste culture. JavaScript frameworks that need a bootcamp to explain. Infrastructure that runs on configuration files that no single person has read end-to-end.
AI didn’t create this problem. It just put us in turbo mode.
The bill is coming due
Planes will crash. Power grids will fail. Entire regions will go dark. Cars will brick themselves in parking lots. Not because the underlying technology is bad, but because nobody fully understands the system they’re shipping.
The MTBF vs MTTR debate in infrastructure taught us that mean time to recovery is great until the failure mode is something nobody anticipated. When you’re shipping code that no human has read, generated by models that produce statistically plausible but semantically wrong outputs, you’re not just shipping software. You’re shipping unknown unknowns.
A system where every individual component passes tests but the emergent behavior is catastrophic. That’s what Mitchell means by a “resilient catastrophe machine.” And we’re building them at a pace that would have seemed insane five years ago.
The numbers that should scare you
A study from GitClear found that the share of code that’s reverted, updated, or otherwise changed within two weeks of being written has been steadily climbing since 2020. Code churn is accelerating. The average lifespan of a line of code before it gets rewritten is shrinking.
Combine that with AI code generation, where the acceptance rate for suggestions is over 30% in some studies, and the average developer reviews AI-generated code less critically than human-written code. The cognitive load of reviewing code you didn’t write is higher, but the perceived authority of “the computer generated this” makes people less likely to question it.
The result: more code, written faster, reviewed less carefully, generating more churn, with fewer people who understand the full picture.
This is not a sustainable trajectory.
The cost to recover
Every engineering bubble has a cleanup phase. The dot-com crash wiped out $5 trillion in market value. The 2008 financial crisis cost an estimated $22 trillion in economic damage globally. The crypto winter of 2022 erased $2 trillion.
The AI software quality bubble will have a reckoning too. Root cause analysis of critical system failures will reveal patterns nobody wants to talk about: generated code that introduced subtle state bugs, agents that fought each other in production, dependencies that nobody tracked, monitoring dashboards that showed green while the house burned down.
The cost to recover from this bubble will dwarf whatever time we saved by shipping faster.
What this means for your money
This might sound disconnected from personal finance. It’s not.
The apps you use to manage your money face the same pressures. The budgeting tools, the investment platforms, the banking interfaces. All of them. The same “ship fast, fix later” mentality. The same pressure to add features nobody asked for because the AI made it easy to build them.
Why budgeting apps sell your data is one symptom of this. Companies are incentivized to extract maximum value from your financial data because the business model demands constant growth. The AI arms race only accelerates this: more features, more data collection, more complexity, less transparency.
The three financial decisions that actually matter, salary negotiation, mortgage refinancing, and debt consolidation, all require understanding the full picture of your finances. You can’t delegate that understanding to an app that treats your data like a commodity.
The counterpoint, and the reason I built Basalt the way I did, is that some things are better when they’re simple enough for one person to understand. Your budget shouldn’t be a mystery. It shouldn’t run on infrastructure that nobody fully comprehends. It should be a file you can open and read. File-based budgeting is the deliberate slow approach in an industry that’s forgotten how to be slow.
Mitchell’s right to worry. The question is whether we’ll do anything about it before the catastrophe machines prove him right.
Basalt is a privacy-first budgeting app that stores your data as files on your device. No cloud, no AI-generated features you didn’t ask for, no company that can see your transactions. Built for people who want to understand their money, not feed another algorithm. Try the beta at basaltapp.com.