The year-end holidays are a time for reflection. But for programmers, a little deeper thought might leave them feeling uneasy.

AI Guru Karpathy Is Anxious: As a Programmer, I’ve Never Felt So Left Behind

Just now, Andrej Karpathy posted a message on X that resonated strongly with tens of thousands of programmers and industry professionals, sparking intense discussions.

Karpathy admitted frankly: “I’ve never felt so behind as a programmer as I do right now.”

He pointed out that the programming profession is being completely restructured—programmers are writing less and less code, and instead spending more time connecting various tools. If one can properly leverage the new technologies that have emerged over the past year or so, they can become 10 times more powerful; otherwise, failing to keep up will lead to skill anxiety.

There is now a new programmable abstraction layer to master, including agents, subagents, prompts, context, memory, patterns, permissions, tools, plugins, skills, hooks, MCP, LSP, slash commands, workflows, IDE integrations, and more.

Furthermore, it is necessary to establish a comprehensive mental model to understand the strengths and weaknesses of entities that are inherently random, error-prone, incomprehensible, and constantly evolving (referring to AI models)—entities that have suddenly intertwined with the excellent traditional engineering practices of the past.

Using Karpathy’s metaphor, it’s like a powerful alien tool has been distributed to everyone, but without an instruction manual. Everyone has to figure out how to use it on their own, and this transformation has brought a “magnitude 9 earthquake”-like impact to the entire industry.

In a word: roll up your sleeves and get to work—don’t get left behind.

As soon as these words were posted, they quickly garnered over 22,000 likes, more than 3,000 retweets, and 3.6 million views. Numerous developers expressed similar feelings in the comment section.

Veterans Are Also Relearning

Senior engineer Boris Cherny said: “I feel this way every week. Sometimes I start tackling a problem manually, then have to remind myself: Claude should be able to handle this.”

He also gave a specific example. Recently, while debugging a memory leak in Claude Code, he habitually resorted to old methods: attaching a profiler, using the application, pausing the profiler, and manually inspecting heap allocations.

But his colleague directly asked Claude to generate a heap dump, then had it read the file to find objects that shouldn’t have been retained. Claude got it right on the first try and submitted a PR immediately.

“This kind of thing happens almost every week.” Boris noticed an interesting phenomenon: to a certain extent, new colleagues, especially recent graduates, can use the models most effectively because they don’t have a bunch of preconceived notions about what the models “can and can’t do.”

He said that every month or two, he needs to invest a lot of mental effort to readjust his perception of the models’ capabilities, as they continue to progress in coding and engineering. Last month, for the first time as an engineer, he didn’t open his IDE at all—he wrote about 200 PRs entirely using Opus 4.5, with every line of code generated by AI.

“Software engineering is undergoing a fundamental transformation. Even for early adopters and practitioners like us, the hardest part is constantly readjusting our expectations—and this is just the beginning.”

Karpathy used a metaphor to explain this feeling: it’s like you’re walking around aiming the AI, and it fires projectiles that sometimes misfire, but occasionally when you hold it at just the right angle, a powerful laser beam suddenly bursts forth, solving your problem in an instant.

In other words, the AI tool is extremely powerful but not stable enough to be as controllable as traditional programming. You have to keep experimenting through trial and error—most of the time, it’s trivial or leads to failures—but once you find the right method, it can bring exponential productivity gains.

Igor Babuschkin, co-founder of X, specifically praised the rival Claude Opus 4.5 for its outstanding performance in the comment section. Karpathy responded that AI is evolving so fast that anyone who hasn’t kept up in the past 30 days already has outdated views.

Technology expert and venture capitalist David Galbraith said: “Spending three months this summer burning the midnight oil to learn how to use AI coding agents to deliver truly high-quality products—not the sloppy, casual code garbage—was the best thing I’ve done in my career.”

X blogger @omarsar0 took a more optimistic and relaxed attitude. He believes that the increasing sparsity of code and the rapid progress of AI don’t bother him because he doesn’t see it as a “race.” On the contrary, the field is now completely open—creative solutions and workflows can come from anyone, anywhere. This transformation isn’t limited to coding; it’s also happening in research and other knowledge-intensive fields. He advised everyone not to be anxious: spend 2 hours playing with tools every day, experiment more, share more, focus on how to feed good context to AI, and then keep building projects relentlessly.

The author of Build a Large Language Model From Scratch also took a laid-back view: the widespread anxiety about being “left behind” in skills usually stems from trying to do too many things at once rather than delving deeply into a few. For example, some people try to learn multiple programming languages instead of specializing in 1 or 2. Or they attempt to keep up with research papers in multiple fields/subfields at the same time… This isn’t inherently a bad thing, but it does cause a lot of stress.

Even X blogger @samswoora sighed: “The profession of software engineer is coming to an end. It might be in 5 years, or 10, but we can all feel that the end has begun.”

Famous blogger Yuchen Jin argued that artificial intelligence isn’t replacing programmers—it’s replacing programming languages.

Opposition from the Traditional Camp

However, there are also dissenting voices, represented by Rob Pike—co-founder of the Go programming language, a Unix veteran, and a pioneer of minimalism and high-quality engineering.

Rob Pike received an automatically generated holiday thank-you email from Claude Opus 4.5, which praised him for promoting simple and powerful software design and noted that his contributions to Go, Plan 9, UTF-8, and Unix have had a profound impact. But Rob was enraged and lashed out publicly on X: “You AI companies waste billions building toxic, non-recyclable hardware and destroy society, yet you have machines pretending to thank me for advocating for simple software?”

Rob Pike’s anger tapped into the complex emotions many programmers feel towards AI hype.

Some netizens fully understood and supported Rob’s attitude: such low-quality code and spam generated in bulk by AI are indeed annoying—especially for old-school geeks like Rob who pursue ultimate simplicity and pure engineering, it’s simply an insult.

Regardless, we have to admit that the development and progress of AI over the past two years have been astonishing. Although there has been intense discussion among AI experts over the past year about the end of Scaling Laws, the fierce competition among major technology companies has not slowed down AI development—it has actually accelerated it.

According to data from Epoch AI, the Epoch Capabilities Index (ECI), a composite indicator measuring general AI capabilities, has grown almost twice as fast over the past two years as it did in the two years before that, with a 90% acceleration in April 2024.

The actual exponential growth has even exceeded initial expectations, and this momentum is likely to continue into 2026.

It’s hard to imagine how far AI will advance by 2026. What predictions do you have for AI development in 2026? Welcome to share your thoughts in the comment section.

Reference Links:

https://x.com/karpathy/status/2004607146781278521?s=20
https://x.com/bcherny/status/2004626064187031831
https://x.com/daveg/status/2004661204296589480?s=20
https://x.com/nixcraft/status/2004644277859889181?s=20
https://epoch.ai/data-insights/a ... rogress-has-sped-up

THE END

The year-end holidays are a time for reflection. But for programmers, a little deeper thought might leave them feeling uneasy.

AI Guru Karpathy Is Anxious: As a Programmer, I’ve Never Felt So Left Behind

Just now, Andrej Karpathy posted a message on X that resonated strongly with tens of thousands of programmers and industry professionals, sparking intense discussions.

Karpathy admitted frankly: “I’ve never felt so behind as a programmer as I do right now.”

He pointed out that the programming profession is being completely restructured—programmers are writing less and less code, and instead spending more time connecting various tools. If one can properly leverage the new technologies that have emerged over the past year or so, they can become 10 times more powerful; otherwise, failing to keep up will lead to skill anxiety.

There is now a new programmable abstraction layer to master, including agents, subagents, prompts, context, memory, patterns, permissions, tools, plugins, skills, hooks, MCP, LSP, slash commands, workflows, IDE integrations, and more.

Furthermore, it is necessary to establish a comprehensive mental model to understand the strengths and weaknesses of entities that are inherently random, error-prone, incomprehensible, and constantly evolving (referring to AI models)—entities that have suddenly intertwined with the excellent traditional engineering practices of the past.

Using Karpathy’s metaphor, it’s like a powerful alien tool has been distributed to everyone, but without an instruction manual. Everyone has to figure out how to use it on their own, and this transformation has brought a “magnitude 9 earthquake”-like impact to the entire industry.

In a word: roll up your sleeves and get to work—don’t get left behind.

As soon as these words were posted, they quickly garnered over 22,000 likes, more than 3,000 retweets, and 3.6 million views. Numerous developers expressed similar feelings in the comment section.

Veterans Are Also Relearning

Senior engineer Boris Cherny said: “I feel this way every week. Sometimes I start tackling a problem manually, then have to remind myself: Claude should be able to handle this.”

He also gave a specific example. Recently, while debugging a memory leak in Claude Code, he habitually resorted to old methods: attaching a profiler, using the application, pausing the profiler, and manually inspecting heap allocations.

But his colleague directly asked Claude to generate a heap dump, then had it read the file to find objects that shouldn’t have been retained. Claude got it right on the first try and submitted a PR immediately.

“This kind of thing happens almost every week.” Boris noticed an interesting phenomenon: to a certain extent, new colleagues, especially recent graduates, can use the models most effectively because they don’t have a bunch of preconceived notions about what the models “can and can’t do.”

He said that every month or two, he needs to invest a lot of mental effort to readjust his perception of the models’ capabilities, as they continue to progress in coding and engineering. Last month, for the first time as an engineer, he didn’t open his IDE at all—he wrote about 200 PRs entirely using Opus 4.5, with every line of code generated by AI.

“Software engineering is undergoing a fundamental transformation. Even for early adopters and practitioners like us, the hardest part is constantly readjusting our expectations—and this is just the beginning.”

Karpathy used a metaphor to explain this feeling: it’s like you’re walking around aiming the AI, and it fires projectiles that sometimes misfire, but occasionally when you hold it at just the right angle, a powerful laser beam suddenly bursts forth, solving your problem in an instant.

In other words, the AI tool is extremely powerful but not stable enough to be as controllable as traditional programming. You have to keep experimenting through trial and error—most of the time, it’s trivial or leads to failures—but once you find the right method, it can bring exponential productivity gains.

Igor Babuschkin, co-founder of X, specifically praised the rival Claude Opus 4.5 for its outstanding performance in the comment section. Karpathy responded that AI is evolving so fast that anyone who hasn’t kept up in the past 30 days already has outdated views.

Technology expert and venture capitalist David Galbraith said: “Spending three months this summer burning the midnight oil to learn how to use AI coding agents to deliver truly high-quality products—not the sloppy, casual code garbage—was the best thing I’ve done in my career.”

X blogger @omarsar0 took a more optimistic and relaxed attitude. He believes that the increasing sparsity of code and the rapid progress of AI don’t bother him because he doesn’t see it as a “race.” On the contrary, the field is now completely open—creative solutions and workflows can come from anyone, anywhere. This transformation isn’t limited to coding; it’s also happening in research and other knowledge-intensive fields. He advised everyone not to be anxious: spend 2 hours playing with tools every day, experiment more, share more, focus on how to feed good context to AI, and then keep building projects relentlessly.

The author of Build a Large Language Model From Scratch also took a laid-back view: the widespread anxiety about being “left behind” in skills usually stems from trying to do too many things at once rather than delving deeply into a few. For example, some people try to learn multiple programming languages instead of specializing in 1 or 2. Or they attempt to keep up with research papers in multiple fields/subfields at the same time… This isn’t inherently a bad thing, but it does cause a lot of stress.

Even X blogger @samswoora sighed: “The profession of software engineer is coming to an end. It might be in 5 years, or 10, but we can all feel that the end has begun.”

Famous blogger Yuchen Jin argued that artificial intelligence isn’t replacing programmers—it’s replacing programming languages.

Opposition from the Traditional Camp

However, there are also dissenting voices, represented by Rob Pike—co-founder of the Go programming language, a Unix veteran, and a pioneer of minimalism and high-quality engineering.

Rob Pike received an automatically generated holiday thank-you email from Claude Opus 4.5, which praised him for promoting simple and powerful software design and noted that his contributions to Go, Plan 9, UTF-8, and Unix have had a profound impact. But Rob was enraged and lashed out publicly on X: “You AI companies waste billions building toxic, non-recyclable hardware and destroy society, yet you have machines pretending to thank me for advocating for simple software?”

Rob Pike’s anger tapped into the complex emotions many programmers feel towards AI hype.

Some netizens fully understood and supported Rob’s attitude: such low-quality code and spam generated in bulk by AI are indeed annoying—especially for old-school geeks like Rob who pursue ultimate simplicity and pure engineering, it’s simply an insult.

Regardless, we have to admit that the development and progress of AI over the past two years have been astonishing. Although there has been intense discussion among AI experts over the past year about the end of Scaling Laws, the fierce competition among major technology companies has not slowed down AI development—it has actually accelerated it.

According to data from Epoch AI, the Epoch Capabilities Index (ECI), a composite indicator measuring general AI capabilities, has grown almost twice as fast over the past two years as it did in the two years before that, with a 90% acceleration in April 2024.

The actual exponential growth has even exceeded initial expectations, and this momentum is likely to continue into 2026.

It’s hard to imagine how far AI will advance by 2026. What predictions do you have for AI development in 2026? Welcome to share your thoughts in the comment section.

Reference Links:

https://x.com/karpathy/status/2004607146781278521?s=20
https://x.com/bcherny/status/2004626064187031831
https://x.com/daveg/status/2004661204296589480?s=20
https://x.com/nixcraft/status/2004644277859889181?s=20
https://epoch.ai/data-insights/a ... rogress-has-sped-up

THE END