There’s a paradox many of us have noticed: a couple of years ago the end of programmers was announced, but it turns out that the people now spending the most on AI are these very professionals.

Will programming disappear?

The short answer is no. The long answer is this: programming is not coding. Coding agents, like the ones I’ve mentioned, help you with only one part of the development process.

Another way to look at it is this: to succeed with a coding agent, you have to structure your work, plan a strategy for tackling the problem, define the implementation stages, burn through each stage in a session with the AI, review the code, test, fix, and deliver.

All of that is programming. You haven’t stopped doing it; it’s just that you no longer code, or you write less code than you used to (even though more code gets generated).

Vibe coders are programming, but they aren’t aware that they’re doing it. And over time they learn to program, which is nothing more than ordering your thoughts and conveying them to the machine.

Everything should evolve into something like what we see in Star Trek:

DAX: Computer, create a data base for all historical references to the Orbs, including all reports of any unexplained phenomena in Bajoran space.

COMPUTER: Time parameters?

DAX: Ten millennia.

COMPUTER: Initialising data base. Requested function will require two hours to complete.

I also think coding should disappear, or evolve into a single language that all AIs master.

Reinforce your process

MCPs are an innovation that’s having a big impact on the development of AI agents. At work we use Shortcut, and of course they have an MCP.

When you configure Claude Code or Cursor to use it, you add a powerful tool to your workflow.

If the story’s description is well written and complete, you can use it in your prompts to help you in your “vibe coding” sessions.

We’re at an intermediate stage in the evolution of AI. In essence it’s “autocomplete” on steroids; we don’t yet have an artificial intelligence that reasons, and that may still be a long way off.

In the meantime, we have to modify our process slightly, but more than anything, reinforce it. Companies that adopt no-holds-barred vibe coding are going to fail.

Today, more than ever, you have to reinforce quality control, security checks, the continuous delivery process, and so on.

AI, in its current state, can’t fully take over your software development process. Those who believe programmers will be laid off en masse are wrong. Good programmers will come to the rescue of those projects created by vibe coders. The latter will learn to program. And everyone will realize that we’ll get an increase in productivity, but it won’t be exponential as some assume.

The stages of the development process are there for a reason. That’s why I say that companies with bad development processes that bring in AI as a magic recipe won’t just be disappointed; they’ll also make their situation worse.

Improve your process using AI

Here’s one of the many ways you could improve your development process:

• Use vibe coding to prototype. Hold one or several sessions together with your clients or users, and iterate until you’ve built a quick prototype that lets them validate their ideas.
• Capture the feedback, record it, make transcripts, use ChatGPT to prepare a good PRD.
• Hand this to your development team. That information can be split into use cases, or stories, that you can enter into a system like Shortcut (even better if that system has an MCP).
• Iterate in coding sessions with Claude Code, Cursor or whichever coding tools suit you best.
• Reinforce your DevOps process. It’s vital to have version control and a fast way to go back, with good backups and an automatic rollback process.
• Do you have testers on your QA team who don’t code? Teach them to use vibe coding tools. They’ll be very good at giving the AI feedback so it makes the corrections.
• Reinforce pen-testing and security testing, because there’s probably a lot of generated code that isn’t going to be reviewed in depth.
• Give your programmers access to the Pro or Max modes of these tools; budget between 50 and 200 dollars a month for each one.

This is just one idea. Every company has its own challenges, but a lot of improvements can be achieved by adopting AI in your development cycle.

I’d like to read your opinions; you can comment here or on the newsletter.