“Vibe-coded this on the weekend — now it makes $10K MRR while I sleep.”
Introduction
I started my career as a UNIX sysadmin in the pre-internet days.
Back then, if you wanted something online, you had to know it all. You built your own web server, configured DNS by hand, ran email servers, tuned spam filters, hardened operating systems, and deployed application servers. There was no managed service, no “one-click deploy.” If you didn’t understand every moving part, your system simply didn’t run.
That experience gave me a lifelong respect for what it takes to put a system into production. Which is why I view today’s culture of fast builds and AI hype with a skeptical eye.
The Myth of “Finished”
We’ve all seen the posts: “I spun this up in a weekend it’s live & making $$$!”
But live isn’t the same as production.
In development, “done” means the code runs, the UI looks passable, the demo flows. In production, “done” means the system can withstand real-world traffic, hostile environments, and the constant churn of updates.
If you haven’t looked at your logs, tested your endpoints, or built for resilience, you don’t have a product. You have code exposed to the entire internet.
Prototypes Have Their Place
Fast builds and MVP demos are valuable. They spark ideas, inspire teams, and open doors. But we need to stop confusing these prototypes with production systems.
I say this as someone who still enjoys hackathons. Not long ago, I built AIvoxveri.com, an experimental voice interview platform aimed at separating genuine human applicants from AI-generated impostors and reducing recruiter load by 70%. It combined agentic AI for evaluation and fairness with Hedera blockchain to provide verifiable, immutable evidence of each interview.
That build process was exhilarating: 2 weeks of night time designing, coding, and seeing something come alive. But it was never meant to be “production.” It was a proof-of-concept, a conversation starter.
Could AIvoxveri.com handle enterprise-scale data pipelines, hardened endpoints, auditability for compliance, and the scrutiny of regulators? Not yet. But that’s the point. A hackathon prototype can prove potential — but real engineering turns potential into a system you can trust.
How AI Can Avoid the Same Mistakes
Let me be clear. I absolutely believe the world will be AI enhanced and that agents will be at the centre of that revolution. But at the moment the risk now is that hyped AI and agentic systems fall into the same trap at enterprise scale.
Right now, companies are pouring millions into flashy demos: proof-of-concepts that impress in boardrooms. AI agents that summarize documents, draft emails, even mimic decision-making.
The danger is that we stop there. That the demo becomes the “finished product,” and the hard work of engineering never happens.
But it doesn’t have to be that way. We can prevent AI from becoming just another cycle of hype by shifting the focus early:
- Security: Define who owns the data flows and how they’re logged.
- Reliability: Design for what happens when APIs fail, not if.
- Compliance: Build explainability into the system from the start.
- Operations: Resource teams to patch, monitor, and maintain AI pipelines as living systems.
If we embed this discipline from the beginning, AI projects won’t quietly die after pilots: they’ll mature into production systems that create real value.
“We Don’t Need Engineers Anymore”
One of the most dangerous narratives in the AI hype cycle is the idea that agents and LLMs mean we no longer need traditional engineers.
It’s wrong.
AI can generate code. Agents can automate glue work. But none of that replaces the discipline of production engineering: logging, monitoring, scaling, securing, explaining.
Without engineers — and I mean “real” engineers — AI projects hold up at prototype stage. Which is why companies spend heavily on “AI transformations” only to abandon them when the gap to production becomes obvious.
Lessons from 30 Years of Hype
I’ve seen this cycle so many times before.
Some hyped technologies burned investor money in pilots and never reached production (Push Technology, Portals, Second Life, Enterprise Blockchain, Metaverse). Others eventually matured into core infrastructure (E-commerce, Wi-Fi, Cloud, Mobile, Big Data).
The pattern is simple:
- When hype leads to infrastructure (the internet, cloud, mobile), it survives.
- When hype is just a demo in search of a problem (Second Life campuses, VR everywhere, corporate blockchains), it fizzles.
AI is at a crossroads. It is clearly real — generative AI is already embedded in Office, Google Workspace, and developer tools. But whether agentic AI becomes infrastructure or just another abandoned hype wave will depend on whether companies build production-grade systems around it.
What Production Really Means
Whether it’s 1995 and you’re running your own mail server, or 2025 and you’re deploying multi-agent AI workflows, the principles haven’t changed:
- Defence in depth: Assume vicious attacks from day one.
- Monitoring and logs: If you can’t see it, you can’t trust it.
- Operational discipline: Backups, patches, failover.
- Human oversight: Not just for ethics, but for resilience.
- Testing, testing: If it wasn’t tested, you can’t prove explainability.
These are not optional. They are what make a system a system.
Closing Thoughts
I’ve been building long enough to know that nothing is ever really “done.” The internet is hostile by default. Demos have their place. Hackathons have their place. But prototypes and MVPs are not production. The real engineering work begins after the demo. Without it, AI projects don’t transform companies — they just drain budgets. So the next time someone says “AI means we don’t need engineers,” remember this: hype doesn’t keep systems running. Logs do. Engineers do.
This post originally appeared on Edward Johnson’s blog on Medium