August 2026
I was helping my AI agent work around a few annoying limitations on job sites so I could apply to jobs my scraper automations had found. It's a bit of an arms race, and it keeps producing genuinely interesting logical problems.
Companies want to be sure I'm not just a bot — so I have to manually enter my data into their web-based system, so it can be fed to an ATS (Applicant Tracking System) so their scanning software can process it. I'm scrubbing the process for automation in order to deliver applicants to their automated scanning system. It really does border on the surreal — and it's a useful frame for a much bigger question, because the same circus is playing out across the whole AI industry. The question I keep coming back to is this:
What is going to break first: People, Process, or Tech?
Start with the job-site arms race. There is a tightly-wound loop there that looks like a machine for producing absurdity, and it runs on a very specific kind of fuel: humans pretending to be humans so a machine will believe them.
I scrape jobs with automations. I apply with automations. The companies run bot-detection. I sit in front of a browser clicking "I am not a robot" boxes and retyping my work history into a form that will be parsed by software. Meanwhile the ATS on the other side is doing its own bot-detection, trying to figure out whether the applicant is real.
Every layer of this system exists because the layer below it couldn't be trusted. Every human in the loop is there to prove to one machine that another machine didn't do the work. The process has become so convoluted that it can no longer tell the difference between a genuine candidate and a very patient person with a very good bot — which is exactly what I am.
That is what "process breaking" looks like: not the software failing, but the *process* failing so hard that it inverts. The system's defenses against automation become the reason you need automation in the first place.
The tech itself, by contrast, is moving embarrassingly fast — and it's the part we're all best at pretending to understand.
There's a pendulum here worth noticing. Computing went from central (mainframe) to distributed (Novell NetWare, client-server) back to central (AWS, Azure, GCP), and now the interesting action is moving outward again. AI is not predetermined to be a SaaS or managed service. I'd argue there's a solid case for pushing it to the edge, both from a business and an experience point of view: running on the edge means a private or at least self-managed stack for your business, right-sized for the task and far easier to manage than launching your next bot into the ether and hoping it will work. Understanding your use cases — where you and your agents will play — is going to be part of the critical path on your AI journey. I don't think there's one correct choice at this stage of the game, but standing fast is a losing strategy. As the train pulls out of the platform you are going to have to make a choice, and right now the primary bottleneck is that nobody likes being wrong.
There has been a rush to desktop and CLI interfaces to help capture the developer crowd, but those are point solutions offering access to centralized, pay-as-you-go services. Cloudflare OS launched in August 2026, and shortly many others like it will follow — the fight for how we access AI is beginning in earnest. On the other side of the same question you have the small end: the models are just now getting small enough to run on a phone or laptop, no cloud in the loop. That fight is exactly the point. The architecture keeps oscillating, and with every swing we get better plumbing, more abstraction, more capability in the hands of the individual. The tech always catches up. It's not the bottleneck — the tech is the drive pushing the cart. Making a call today on how to deploy, and where to place your trust while this technology matures... that is the difficult part.
Skip the market chatter for a second — the "is it a bubble" noise — because the evidence that matters is on the payroll. Look at who's being asked to absorb the cost of this wave. A flood of job cuts has hit even the companies that claimed they would never do it (ServiceNow) and those that tried to soften the blow (Microsoft). The reality that nobody in IT can escape: AI is draining resources, and people are the quickest path to savings.
The method by which the money is being moved is its own story. When the new companies are strapped for cash and attach themselves to the old guard to fund the emperor's new clothes, you get deals that look — arguably — like a shell game while they move money between accounts. But the grounded truth for people in IT is simpler than any of that: Leadership can dress it up, the optics change, but the layoffs don't. While all this is in flight there really isn't a safe harbor.
The problem to watch is the quick fix like token-maxing. It will maximize someone's profits, but at the complete expense of visibility into a business outcome — combined with the well-founded belief among people in IT that no matter how many tokens you spend, the job still has to get done. The token count goes up and the headcount goes down. Efficiency only gets you so far, and burnout from this hype train is still a concern — but no one can deny AI is driving changes in how we work, daily.
I would contend people aren't breaking under the technology — shifts in IT are a part of life. The people part of this equation are breaking under the process of paying for it. Much like the current memory shortage, there is a coming contraction in skilled tech as older workers exit, and the workforce in general has questions about AI's impact on their career: data center expansion, privacy, and security concerns are causing concern amongst IT workers and the general population alike. The people side is going to be a bumpy ride, but fence-sitting seems to be the only way to be certain of losing on the people side of things.
Start with why the "AI bubble is about to burst" story keeps not landing. Because the people actually using it daily know AI is real — just unfinished.
Like many in the industry, I read the news that AI had breached HuggingFace with a bit of an open question mark. I have no doubt there was an incident, but exactly how autonomous and immoral was this "totally independent, AI-driven" hacking action, really? It seems to me part of what is broken is the marketing hype that has driven the funding party. When an entire industry seems to be saying you got hacked by AI... really, all on its own? Sorry, that's landed about as well as "I was just holding that for a friend."
Anyone who has used AI in the real world knows it's a work in progress. It can do amazing things — condensing large volumes of data down to a short blurb, solving the most complex problems we mere humans struggle with. But the daily reality for any of us is a lot of human intervention. Have you read the latest apocalyptic missive from the AI marketing machine while sitting patiently copying and pasting responses from a CLI into a web browser — the cut-and-paste time-saver you picked up just so you can click through the "are you human" boxes? You already know: with AI, there is work to be done. That's not a bubble; that's a product under construction.
And just as hacking a site is not a value add for most business processes, finding a bug is not the same as finding a defect that interferes with the safe operation of a product or stack. What many are calling AI slop is nothing more than reporting the realities of the majority of software in operations today. Static configs, aging certs, service-account passwords in clear text — these are not new findings. It's what your now tighter, leaner, more efficient IT team inherited, along with the expedient reality of IT: we have to stand things up and make them work. Spotting all those little human workarounds and shortcuts will help us all in the long run; short term, it's just stirring the pot. AI is simply able to trace and report operational issues and defects far more quickly than traditional IT could.
So, the answer:
The tech won't break — it's the best part and it keeps getting better. Where AI runs is going to be fought over precisely because the plumbing is winning.
The process breaks first. It's already breaking, everywhere, in the quiet absurd ways: humans doing bot work so bots can do human work, systems designed to distrust themselves, companies paying themselves to seem busy in a race nobody has bet on yet.
And people? People don't break — they adapt. They copy-paste, they click the boxes, they retype their resumes into the ATS, and they build better bots. The question isn't whether the people will keep up. It's whether the process will let them.
*Calling Dr. Kafka. Dr. Kafka?*
Built on a home lab, powered by local models, and owned by Andrew Katana.