Death to “vibe coding”

I am trying to think a physical-world corollary to the mindless reliance on vibe coding and I cannot seem to find one. In the real world we seek out professionals, for our day-to-day chores. And demand them, for our important tasks; from attorneys, accountants, and doctors, to plumbers, electricians, and engineers we expect those practicing a trade to have mastered those skills. 

We encourage our children to get an education, go to university, and earn a degree; one that shows corporate America that they have invested in the their future and are ready to begin their career. 

For those who sit outside of the interests of the classroom, technical and vocational training are a must for our HVAC, mechanics, and other trades that power much of America’s economy. Both are equally important to our economy and centuries of training have solidified standards and practices that ensure the safety and competency of those in these fields. 

What these paths remind us is that a career in a chosen profession requires a diligent focus on education. Learning the basics of the craft through proper application of theoretical concepts in real world situations. Understanding that choices made in the daily role have an impact further down the line for your colleagues, peers, and customers. And over time, experience in those industries breeds wisdom that enables you to be more efficient, effective, and profitable. 

Would you allow your next major investment in a house to be built by someone who just went to Ace Hardware and bought a hammer and watched a video on YouTube? How about allowing someone with no understanding of the complex financial markets today to invest your entire fortune for your future because meme coins are hot? Or to bring this absurdist notion to its end, would anyone of us select anyone less than a board-certified surgeon to perform the next operation on us or a loved one?

The answer to these questions seem obvious – No, as they should. But that same logic seems to escape those in the business world today; especially those in the C-suites.

Most companies have rigorous hiring practices with extensive requirements for new hires. Those requirements become more painstaking as the roles inch closer to the C-suite of any corporation. However, in a rush to integrate machine learning into their businesses, as increasing shareholder value demands from the Wall Street peanut gallery, businesses are progressively taking on riskier practices that defy common sense. 

Companies are not only allowing but actively encouraging those untrained in complex programming to take those roles on to build their next business applications. 

For example, without proper training in development and an in-depth understanding of the impacts integrating open-source libraries into these vibe coded applications, they are opening up their businesses to: cybersecurity breaches, deployment and maintenance issues, and user-experience nightmares, that are planned, prevented, and mitigated by their more experienced counterparts. 

Let’s take the request for a simple timestamp added to the application that may come down from the executive to its newly ordained vibe coder. 

At its core, this seems like a simple request for a timestamp; an alphanumeric sequence that usually applies to the date and time of the file being reviewed.

This simple request becomes immediately more complex when you begin to work with large scale enterprise applications due their integrated nature and how they read and write data from other systems. Some of those systems may be internal to the company while others are more than likely derived from other networks, partners, and businesses that supply services to the company.

That complexity increases when you start to breakdown what a timestamp actually looks like. At its simplest for, the timestamp might look like [HH:MM:SS] where HH, MM, and SS are hours, minutes, and seconds. However, that simplicity disappears when you begin to work across enterprises, timezones, and integrated systems that have been developed over time, possibly decades. 

If we examine the day of the week as a component of a timestamp, that seems like a simple thing to tackle. But, should it be stated DAY, fully spelling out each day – Sunday, Monday, Tuesday, Wednesday, Thursday, Friday, Saturday? Or, should it be something simpler and abbreviated like, Sun, Mon, Tue, Wed, Thu, Fri, Sat. 

Some of this is determined by the systems we might be integrating the data with, but at the outset of the design process we need to investigate these seemingly small details to ensure that our designs work with the data needed to support the business goals of the enterprise.

And the gamut of details that can be enabled on a timestamp can be daunting:

Component – Julian, Format – year, month, day, Description –
Component – seconds, Format – SS, Description – seconds (00 – 59)
Component – minute, Format – MI, Description – minutes per hour (00 – 59)
Component – hour, Format – MERIDIAN, Description – AM or PM  periods
Component – hour, Format – HH, Description – hour of the day (01 – 12)
Component – hour, Format – HH12, Description – operates the same way as HH above
Component – hour, Format – HH24, Description – 24 hour format (00 – 24)
Component – hours, minutes, seconds, Format – SSSSS, Description – seconds in 1-day (00000 – 86400)
Component – day, Format – DAY, Description – Day of the week (Sunday, Monday, Tuesday, Wednesday, Thursday, Friday, Saturday)
Component – day, Format – D, Description – Abbreviated day of the week (Sun, Mon, Tue, Wed, Thu, Fri, Sat)
Component – day, Format – DD, Description – Numeric day of the month (1 – 31)
Component – month, Format – MON, Description – Abbreviated name of the month (Jan, Feb, Mar, Apr, May, Jun, Jul, Aug, Sep, Oct, Nov, Dec)
Component – month, Format – MM, Description – months of the calendar (01 – 12)
Component – month, Format – MONTH, Description – Name of month (January, February, March, April, May, June, July, August, September, October, November, December)
Component – month+day, Format – DDD, Description – Numeric day of the year (1 – 365)
Component – year, Format – YY, Description – last 2-digits of the year
Component – year, Format – YYYY, Description – 4-digit year (0000-9999)
Component – week, Format – W, Description – week of the year (1 – 52)

If these formats do not match the integrated systems being used for business, compliance, and legal requirements then the application(s) and its data becomes immediately suspect. This simple request is anything but, and its impact it can have on the business are far reaching and potential expensive.  

Added to this are the costs necessary to alleviate any of the numerous issues that will inevitably creep into the codebase created by those with little to no understanding of programmatic concepts and we see that AI has not delivered on the promise of its value proposition. 

Ford has recently rehired many of the engineers they fired in service to the AI Overlords due to the utter failure of these systems to truly integrate successfully into their business. This experiment cost 3 years, billions of dollars, and damage to Ford’s brand and reputation. Learnings that other companies are/have discovered on their own and should be surveying how important it is to integrate the next best thing if the hidden costs truly damage the company’s business. 

Not only are companies claiming it as a key differentiator, which it is not, but as a business initiative today but one without a clear vision on its significance nor how it should, and more importantly should not, be implemented into its business. The end result it at best buyer’s remorse, but in the case of the Ford example above true damage to the company’s brand, budget, and share price. 

To those that consider this screed rantings from a luddite¹, let me assure you that these impacts have already affected almost every company that is rushing to implement AI. 

There is a place for machine learning in the business of today and tomorrow. However, it is not in the replacement of the workers and the value they bring to every company. Institutional knowledge is as important as skills like experience and intuition that can only be delivered through a human workforce. Gartner has a recent study on some of the pullback on these technologies, like AI-agents, due to lack of ROI and inability to achieve complex goals. 

The real challenge that I see, and have faced much of my career as a creative professional, is the degradation of skills, talents, and experience by those who wish to co-opt the discipline. As the maxim goes, every one with a pencil thinks they are an art director – now, everyone with a computer, with access to these tools, believes they are a coder. 

It takes more than the tools to do the job. As Malcolm Gladwell’s book, Outliers: The Story of Success, tells us, it takes about 10,000 hours of practice to reach a level of expertise that would be considered professional, or world-class. Industry is learning, or more accurately re-learning, a valuable lesson in today’s era of AI. Let’s hope they are fast learners, for all of our sakes. 

¹ And let us remember the Luddites were not anti-technology, they were concerned with the impact of new technologies on the workers and society and the implementation by the wealthy robber-barons to expand their wealth and power at the expense of the worker.