The Impact of AI on Individual Growth

Anthropic recently published a study called How AI Assistance Impacts the Formation of Coding Skills. Basically, they ran a randomized controlled trial with 52 software developers learning a new Python library. The group that used AI assistance scored 17% lower on a comprehension quiz than the group that coded by hand.

I discovered this study mainly because of chatter on the socials, where people were talking about how this proves we’re getting dumber, or that AI only saved 2 minutes, so it’s not worth the hype. Since everyone else has opined on this, I thought I should too.

Life Moves Fast…

As I was writing this article, news about Block laying off 4,000 employees broke. I’m afraid we’re going to see more such announcements. It would be naive of me to think that I won’t be affected by these developments as well. I recommend we all get subscriptions to the Max or equivalent plans at one of the major providers and we learn how to leverage AI to augment our lives.

Back to my original thoughts.

Blaming The Wrench Instead Of The Mechanic.

The study found that how someone used AI mattered more than whether they used it. The people who scored well with AI assistance were the ones asking follow-up questions, requesting explanations, and using AI as a sounding board rather than a code generator. The people who scored poorly were the ones who essentially said “write this for me” and copied the output.

This isn’t surprising, we’ve seen this pattern before.

GPS navigation comes to mind. Researchers at McGill University found that habitual GPS use is associated with worse spatial memory, and that people who relied on GPS more didn’t do so because they had a bad sense of direction — the GPS usage itself led to the decline.

Personally, I’ve found that because of GPS, I’m more adventurous about trying new routes and exploring unfamiliar places. I know I can always ask Maps to take me home. The more I venture out, the more I learn about my surroundings and over time, the less GPS I need to get around. In fact, I’ve learned more shortcuts around town because I knew I wouldn’t be lost for too long.

Another anecdote would be how many phone numbers people have memorized versus those who rely on the contacts app on their phone. In an emergency, how many of us could dial a loved one without looking up numbers on our phones?

Language syntax is the new phone number — you don’t need it memorized, but you still need to know what you’re looking at when something unfamiliar rings.

Did Curiosity Actually Kill The Cat?

Not every programmer is a hacker at heart. Some of us got into programming because we love solving problems. Others got into it because the job market was favorable. Both are valid career paths, but they produce very different relationships with tools.

As I read the study, two things stood out to me.

1. Trio is a thing, and these smart people thought it was a good way to teach about asynchronous programming, even though I don’t code in Python, I will run through their tutorial over the weekend.
2. I need to look into Output Styles. I’ve been using normal mode to work on the linkblog, I’ve actually found the way the agents are thinking most informative, as they help me think. I’m going to spend the next few weeks in Explanatory or Learning mode to see how things go.

The study’s qualitative findings hint at this. They identified seven participants who fell into what they called the “Conceptual Inquiry” group — people who only asked conceptual questions and relied on their improved understanding to complete the task. This group was the fastest among the high-scoring patterns and second fastest overall. These are the tinkerers. They used AI to understand, not to avoid understanding. These were the curious cats…

Meanwhile, the “AI Delegation” group of four participants wholly relied on AI to write code. They finished fastest and learned the least. I’d bet good money these are the same people who, in a pre-AI world, would have copy-pasted from Stack Overflow.

The Big Picture

I think if we ran these same studies across writing, design, legal research, financial analysis, or any other knowledge work, we’d find the same pattern. As someone said,

you’re not going to lose your job to AI, you’re going to lose your job to someone who knows how to leverage AI.

I know Star Trek is fiction, but think about it. As powerful as the AI is in that timeline, they still needed Scotty, Geordi, Data, and sometimes Wesley or Barclay in engineering to know when to divert power from the main deflector dish into the dilithium crystals that would then create a subspace relay going through the wormhole into the gamma quadrant.

“It is hard to communicate how much programming has changed due to AI in the last 2 months: not gradually and over time in the ‘progress as usual’ way, but specifically this last December. There are a number of asterisks but imo coding agents basically didn’t work before December…”

— Andrej Karpathy (@karpathy), February 25, 2026

“Some times AI dazzles you with spectacular feats of ingenuity, but much of the time it just clears away all the little toils that can compound to steal your day.”

— DHH (@dhh), February 26, 2026

“The hottest new programming language is English”

— Andrej Karpathy (@karpathy), January 24, 2023

What Can We Do?

I still think this is the time to invest in ourselves.

Get a subscription at one of the major providers.

If you’re an advanced user, look into ollama and hosting your own models.

1. claude.com/pricing
2. chatgpt.com/pricing
3. gemini.google.com

Then figure out how to leverage these tools in your day to day workflow.