Lately I keep hearing the same story from Portainer customers; a long-standing idea that never got funded is handed to one person, that person is given an AI coding tool (and a pretty healthy token budget), and what comes back is a complete new software solution that gets rolled out enterprise-wide.
The person is usually a senior, experienced software engineer, and their role has quietly changed into driving an AI software factory that generates a codebase far larger than one human could have written, or can now fully read, let alone understand. One example I heard was a file management platform along the lines of SharePoint, housing many millions of files (all MOVED from legacy file shares into the app), full text indexed, AI summarized, searchable, and fed into an LLM as a content source. That is a genuinely impressive advance in what IT can deliver, and it hands the CIO the unconstrained masterpiece they always wanted. My question though, is that really enterprise ready, and enterprise supportable? Once it becomes a mission critical piece of core software, with 99.99% availability as an expectation, is that one engineer willing to dedicate their life to supporting it 24x7?
What got cheap, and what did not
The cost of producing software has collapsed; the cost of owning it has not moved. Owning means patching dependencies, fixing bugs, reviewing security, getting permissions right, backing up and restoring, upgrading, passing an audit, handling the critical incident at 7am just as everyone is logging in and expecting the product to work, and coping with the day the engineer takes another job. Think of the build as the free puppy; the expense is the next ten years of feeding, vet bills, and finding someone to look after it when you go away đ.
When you buy a âSharePointâ class product you inherit a vendor security team, a patch cadence, and a very large installed base finding bugs before you do. A bespoke platform has exactly one installed base, yours, so the bugs are found by your own users or by an attacker, whichever arrives first.
Reading is the new bottleneck
Generation speed scaled, review speed did not. The engineer is now reviewing code they did not write, at a volume nobody can read line by line, so they test behavior instead; that is reasonable, but testing finds what you thought to test. The failures that matter in a platform holding millions of files tend to be the ones nobody thought of. How does permission inheritance behave across that many files? What does the indexer do with a corrupted document? When access to a file is revoked, does the AI summary of that file, or the copy of it sitting in the LLMâs content source, get revoked too?
Does anyone need to understand the code?
Most of us have never read the source of the products we run, so reading every line is not the requirement; the requirement is that someone is accountable for how the system behaves and can explain it when it fails. An AI generated summary of what the app does cannot give you that, because it describes intent and is written by the same system that wrote the code, which makes the author the marker of its own homework.
There is real pride, and a real business win, in a CIO finally getting the platform they always wanted without a three year budget cycle, and I would not want to take that away. The line between a useful tool and a mission critical system is crossed when the data cannot be lost, when the service cannot be paused for a week without the business noticing, or when it holds anything subject to retention or legal obligations. Past that line the recommendation would still be to procure software from a company who has teams of people responsible for all of the above. There is still great value in outsourcing these obligation.
I get the fact that many CIOâs and even CEOâs now see no value in paying for software that can be vibe coded âin an afternoonâ, but realistically, companies spend millions of dollars on product research, development, testing, certifications, compliance, etc, purely to give confidence in service availability, reliability, and security. You dont get a âfree passâ to ignore these requirements just because you have a single engineer who knows how to drive an AI vibe-coding factory.
So if your engineer resigned tomorrow, who would you call? If your system got compromised tomorrow, which vendor would you set your legal team onto ? Who picks up the damages?
Sometimes, just because you can do something, doesnât mean you should.

đŻ I am currently supporting those vibe coders and providing them a IT managed playground with all security guardrails. Once their project passes the management and architecture gates the solution is move to a internal OTA environment. Once moving to mission critical outside a small them than their project is handed over to DevOps feature team. Still in development the whole delivery process but works so far