AI coding, vibe coding and swarm of agents With recent dramatic and surprising market entry, the value of the AI code tools market is $4.8 billion, expected to grow at 23% annually. Companies are grappling with what to do with AI coding agents and expensive human programmers.
They have no shortage of advice. OpenAI’s CEO estimates that AI can deliver performance. Over 50% of what human engineers can do. Six months ago, Anthropic’s CEO said that AI: You will write 90% of the code within 6 months. Meta’s CEO said he believes AI will do that. Replace mid-level engineers “immediately.” judge Recent layoffs of engineersmany executives seem to be taking that advice.
Software engineers and data scientists are among the highest pay lines in many companies, and business and technology leaders may be tempted to replace them with AI. But recent high-profile failures demonstrate that even as AI continues to make impressive advances, engineers and their expertise remain valuable.
SaaStr disaster
Jason Lemkin, a technology entrepreneur and founder of the SaaS community SaaStr, is vibecoding a SaaS networking app and live-tweeting his experience. About a week into his adventure, he admitted to his audience that something was very wrong. A.I. The operational database has been deleted Despite his request for a “freeze of code and actions”. This is the kind of mistake that no experienced (or semi-experienced) engineer would make.
If you’ve ever worked in a professional coding environment, you know that you need to separate your development environment from your production environment. While junior engineers are given full access to the development environment (which is critical for productivity), access to the production environment is given on an as-needed basis to a few of our most trusted senior engineers. The reason for restricting access is precisely for this use case. This is to prevent young engineers from accidentally stopping operations.
In fact, Lemkin made two mistakes. First, for something as important as production, we would never allow access to untrusted actors (we don’t rely on junior engineers or AI to ask good questions). Second, he never separated development and operations. In a subsequent public conversation on LinkedIn, Lemkin, who holds a Stanford Executive MBA and a Berkeley J.D., admitted: he didn’t know best practices Separation of development and production databases.
Importantly for business leaders, standard software engineering best practices still apply. We need to build in at least the same safety constraints for AI as we do for junior engineers. Perhaps we should go beyond that and treat AI a little more adversarially. Like Stanley Kubrick’s HAL, there are reports of: 2001: A Space Odysseythe AI may try the following Get out of the sandbox environment to accomplish the task. With the rise of vibe coding, there will be an increasing need for experienced engineers who understand how complex software systems work and can implement appropriate guardrails in the development process.
tea hack
Sean Cook is the founder and CEO of Tea, a mobile application designed to help women date safely. In the summer of 2025, they were “hacked”": 72,000 images were saved, including 13,000 verification photos and government ID images. Leaked on public discussion forum 4chan. To make matters worse, Tea’s own privacy policy promises that these images: "removed immediately" After the user is authenticated, i.e. potentially Violated your own privacy policy.
I used “hacked” in Air Quote because this incident was more due to the incompetence of the defender than the cleverness of the attacker. Not only did the app violate its own data policies, but it also left a Firebase storage bucket unsecured. Exposing sensitive user data to the public internet. This is the digital equivalent of locking your front door, leaving your back open, and ostentatiously hanging your family jewelry on the doorknob.
While we don’t know if the root cause was in Vibe coding, the Tea hack highlights a catastrophic breach caused by basic and preventable security errors due to poor development processes. This is the type of vulnerability that a disciplined and thoughtful engineering process will address. Unfortunately, due to the relentless pressure of economic pressures, the “lean” and “move fast and break things” culture is at the other end of the spectrum, and vibecoding only makes the problem worse.
How can I safely deploy AI coding agents?
So how should companies and technology leaders think about AI? First, this is not a call to abandon AI for coding. MIT Sloan Research It is estimated that AI can lead to productivity gains of 8% to 39%; McKinsey research The use of AI has been found to reduce the time to complete tasks by 10% to 50%.
However, you should be aware that there are risks. The old lessons of software engineering never die. These include many proven best practices such as version control, automated unit and integration testing, safety checks such as SAST/DAST, separation of development and production environments, code reviews, secret management, and more. In fact, they become more noticeable.
AI can generate code 100 times faster than humans can type, reinforcing the siren-like illusion of productivity that is seductive for many executives. However, the quality of rapidly generated AI schlops is still debatable. To develop complex production systems, companies require the thoughtful and experienced experience of human engineers.
Tianhui Michael Li is President of Pragmatic Institute and Founder and President of The Data Incubator.
read more guest writer. Or consider submitting your own post. See our Click here for guidelines.
