AI coding, vibration coding AND swarm of agents have recently entered the market in a dramatic and astonishing way, and the AI Code Tools market is worth approx $4.8 billion and is expected to grow at a rate of 23% annually. Enterprises are grappling with AI coding agents and what to do with costly human programmers.
They have no shortage of advice. OpenAI CEO estimates that artificial intelligence can work over 50% of what engineers can do. Six months ago, Anthropic’s CEO said A.I would write 90% of the code in six months. Meta’s CEO said he believes AI will do just that replace mid-level engineers “soon”. Judging by recent layoffs in the tech industryit seems that many managers follow this advice.
Software engineers and data analysts are among the most costly salaries in many companies, and business and technology leaders may be tempted to replace them with artificial intelligence. However, recent high-profile failures show that engineers and their expertise remain valuable, even as artificial intelligence continues to make impressive advances.
SaaStr Disaster
Jason Lemkin, tech entrepreneur and founder of the SaaS community SaaStr, codes a SaaS web application and live-tweets his experiences. About a week into his adventure, he admitted to his listeners that something was going very wrong. Artificial intelligence deleted its production database despite his request for a “code and activity freeze”. This is a mistake no experienced (or even semi-experienced) engineer would make.
If you’ve ever worked in a professional development environment, you know how to separate your development environment from your production environment. Junior engineers have full access to the development environment (crucial for productivity), but production access is provided on a narrow basis and on an as-needed basis to a few of the most trusted senior engineers. The reason for restricted access is precisely this apply case: to prevent a junior engineer from accidentally interrupting production.
In fact, Lemkin made two mistakes. First: for something as critical as production, access to unreliable actors is simply never guaranteed (we don’t rely on a nice question from a junior engineer or AI). Second, he never separated development from production. In a subsequent public conversation on LinkedIn, Lemkin, who holds an MBA from Stanford University and a JD from Berkeley, admitted that didn’t know best practices division of development and production databases.
The lesson for business leaders is that standard software engineering best practices still apply. We should include at least the same security constraints for AI as we do for junior engineers. We should probably go beyond this and treat AI a bit adversarially: there are reports that, like HAL in Stanley Kubrick’s film 2001: A Space Odysseyartificial intelligence can try break out of the sandbox environment to complete the task. As coding becomes more lively, there will be an increasing need to have experienced engineers who understand how convoluted software systems work and can implement appropriate security measures into development processes.
Tea hack
Sean Cook is the founder and CEO of Tea, a mobile app launched in 2023 dedicated to helping women date safely. In the summer of 2025, 72,000 photos were hacked, including 13,000 verification photos and ID photos leaked on the public message board 4chan. Worse still, Tea’s privacy policy states that these images will be “immediately deleted” once users authenticate, meaning they could potentially violated its own privacy policy.
I apply the word “hacked” in quotation marks because the incident results not so much from the cleverness of the attackers, but from the incompetence of the defenders. In addition to violating its own data policies, the app left the Firebase storage bucket unsecured, sharing sensitive user data on the public Internet. It’s the digital equivalent of closing the front door, but leaving the back open and family jewelry hanging ostentatiously on the doorknob.
While we don’t know whether vibration encoding was the root cause, the Tea hack highlighted catastrophic breaches resulting from fundamental, preventable security flaws resulting from indigent programming processes. This is the type of vulnerability that a disciplined and thoughtful engineering process eliminates. Unfortunately, the ruthless financial pressure in which the “lose weight”, “act fast and break everything” culture is the complete opposite, and vibration encoding only worsens the problem.
How to safely deploy AI coding agents?
So how should enterprise and technology leaders think about artificial intelligence? First, this is not a call to abandon AI in favor of coding. MIT Sloan Study Artificial intelligence is estimated to lead to productivity increases ranging from 8% to 39%, while a McKinsey study discovered a reduction in task completion time from 10% to 50% thanks to the apply of artificial intelligence.
However, we should be aware of the threats. The senior lessons of software engineering never fade away. These include many proven best practices such as version control, automated unit and integration testing, security controls such as SAST/DAST, separation of development and production environments, code review, and secrets management. If anything, they become more expressive.
Artificial intelligence can generate code 100 times faster than humans can type, creating an illusion of productivity that is a tempting buzzword for many managers. However, the quality of the quickly generated AI store is still a matter of debate. To develop convoluted production systems, companies need thoughtful and experienced engineers.
Tianhui Michael Li is president of Pragmatic Institute and founder and president of The Data Incubator.
Read more in our guest authors. You might also consider submitting your own entry! See ours guidelines here.