Why I was surprised when Vera co-founder Yaniv Bernstein said he was quitting AI

Machine Learning


Australia’s great author Rodney Hall still writes by hand from a standing desk in Melbourne.

Now 90 years old, he is the author of 14 novels, poems, and photo books as well as a biography.

This week, entrepreneur and investor Yaniv Bernstein announced that he would replace AI with a pen, abandoning a technology that is on the verge of inevitable ubiquity.

“After 12 months of touting AI-assisted development, I have a confession: I was wrong. Not partially wrong, not “circumstances” wrong. That’s fundamentally wrong. Yesterday I uninstalled all the AI ​​coding tools I had. ” he writes.

The newly minted AI apostate’s dogma struck a nerve, and his LinkedIn post garnered 40,000 impressions and tons of upvotes.

This was the best April Fool’s prank. Many of us wished it were true. He also used AI to write it.

And if you look carefully at the sides of the post, you’ll find an acrostic spelling that hints at the truth.

Nevertheless, the outspoken Bernstein, who co-hosts The Startup Podcast with Chris Saad, said he was surprised by the number of experienced engineers who took this seriously and, more importantly, wanted it to be true.

“They shared their frustrations with AI tools and expressed relief that someone was ‘finally saying that,'” he said. daily startuphe also noted that there were many indescribable reactions to his heresy.

“What is interesting is not so much the differing opinions, but how deeply divided the experiences are: two groups, both experienced and both trusted, came to very different conclusions about the same technology.

“More interesting questions arise than whether AI is ‘overhyped’ or ‘transformative.’ Why are opinions and results so inconsistent? Is it because of differences in tools, workflows, and use cases? Or is there something more cultural about how teams are bringing and integrating AI into their work?”

As the co-founder of Vera, an AI app aimed at Sandwich-age adults who need advice on dealing with aging parents, Bernstein says he’s “firmly in the camp” of developers who see significant productivity gains from it and treat it as a transformational tool.

“At Vera, tools like Claude Code have truly multiplied my engineering output tenfold, and the contrast between that experience and the response to my April Fools post is something I want to explore further,” he said.

“Every business owner in Australia is now making decisions about AI adoption, but the people they are looking to for guidance disagree.”

Yaniv Bernstein will have more to say on this issue daily startup In the next few days.

In the meantime, here is his original LinkedIn post.

After 12 months of popularizing AI-assisted development, I have a confession. “I was wrong.” It’s not partially wrong, and it’s not “sometimes” wrong. That’s fundamentally wrong. Yesterday I uninstalled all the AI ​​coding tools I had.

Productivity was the first lie I told myself. Sure, we were shipping faster, but when we audited our last three projects, 40% of our code was unnecessary abstractions that the AI ​​generated because we didn’t know when it was going to stop.

By reading code, you can learn how to write code. When AI creates an implementation, it skips the part where the brain builds a mental model. I spent months coasting on the vibe.

I came up with this idea while debugging all day. The agent had written a retry loop with a subtle race condition, but having never actually read the module I couldn’t debug it. After 8 hours, I vowed to never use AI-generated code.

Lately, I’ve been using paper notebooks and pair programming. Sketch your data flow, consider edge cases, and then open the editor. Speeds were down 30%, but defect rates fell off a cliff.

Fewer files, abstractions, and dependencies. This is what my codebase currently looks like. AI agents love creating new files and wrapping them in layers. Without them, the amount of code to do the same job would be halved.

I can understand the reaction that “I was just using it incorrectly.” perhaps. But I’ve spoken quietly with more than a dozen senior engineers who have had the same experience, but they don’t speak publicly because the story is too intense.

Another unexpected thing is that I’m enjoying programming again. There’s a meditative quality to typing a well-thought-out function that autocomplete can’t replicate. Craft is the key.

Look, I’m not a Luddite. But intellectual integrity requires admitting when something goes wrong, even if you’re putting your reputation on the line.

So if you have that same nagging doubt that AI is doing the thinking you’re supposed to be doing, trust your instincts. Close the chat window. Open an empty file. Write the first line yourself.



Source link