Rails Creator: AI Still Can't Beat Junior Programmers

Ruby on Rails creator David Heinemeier Hansson has delivered a sobering assessment of AI's current coding capabilities, comparing the technology to a "flickering light bulb" that illuminates brilliantly one moment and plunges into darkness the next.

Speaking on the "Next Token" podcast, the 37 Signals co-founder expressed long-term optimism about artificial intelligence while maintaining that its present-day output falls short of human programmers. His perspective offers a reality check amid the tech industry's breathless enthusiasm for AI-powered development tools.

"I feel like it's a little bit like a flickering light bulb. You're in total darkness and then it'll flicker on and you go like, 'I can see everything.' And then two seconds later, boom, pitch black."

Hansson's skepticism stems from direct experience with AI coding assistants. When he tasked AI systems with writing lines of code, the results consistently failed to match the quality produced by human developers.

The output wasn't "as good as a job as most junior programmers I've had to deal with," he stated. This assessment carries weight given Hansson's extensive experience mentoring junior developers and building one of the web's most influential frameworks.

At 37 Signals, the company behind Basecamp and Hey, this philosophy translates into practice. For their Kanban-inspired organization product Fizzy, human developers wrote approximately 95% of the code. While the team experimented with AI-powered features, those implementations ultimately ended up on the cutting room floor.

• AI coding output inconsistent and unreliable
• Human-written code remains superior for complex tasks
• 37 Signals prioritizes quality over speed

Hansson directly challenges the narrative that AI will immediately transform software development economics. He remains unconvinced that companies can fire half their programming staff while maintaining or improving productivity.

"I'm not feeling that we're falling behind at 37 Signals in terms of our ability to produce, in terms of our ability to launch things or improve the products."

The entrepreneur views his hands-on coding approach not as a liability but as a deliberate choice. He likens it to a modern saddle maker who takes pride in craftsmanship despite being "no longer part of the main production for transportation."

"I'll continue to make my handwritten code saddles as long as I can for my enjoyment," Hansson explained, framing his approach as a luxury that prioritizes beauty and craftsmanship over pure efficiency.

Despite his skepticism about AI's coding abilities, Hansson acknowledges the technology's genuine value in specific applications. He points to Shopify, where he serves on the board, as evidence of AI's practical benefits.

The e-commerce platform's SiteKick AI agent demonstrates what's possible when artificial intelligence is applied to well-defined problems. The tool helps merchants set up and optimize their shops, delivering "truly incredible" results that Hansson describes as having "real tangible benefits."

This contrast highlights an important distinction: while AI struggles with the nuanced, creative work of programming, it excels at automating repetitive tasks and providing guided assistance within structured environments.

• Shopify's SiteKick helps merchants optimize stores
• AI excels at guided setup tasks
• Practical applications differ from creative coding

Perhaps Hansson's most striking observation concerns the US economy's massive bet on AI achieving human-level capabilities. He views the current landscape as unprecedented in its scale and uncertainty.

"This is the first one where we, in real time, know the world is going to look totally different, and we don't know what the final result is going to be."

The scale of investment staggers even an AI-positive observer like Hansson. He marvels at the conviction driving billions—potentially trillions—of dollars into AI research, with the entire American economy effectively wagering that artificial intelligence will overcome its current limitations.

"The entire American economy right now is one big bet that that's going to happen," he observed. "And I, as AI positive as I am, still marvel at that."

Yet he remains pragmatic about AI's permanence. The technology won't disappear regardless of public sentiment, and while individuals can choose to avoid generative AI, the broader economic momentum is irreversible.

Hansson's perspective offers a balanced middle ground between AI evangelism and dismissal. He neither fears the technology's immediate takeover nor dismisses its long-term potential.

His advice for navigating this uncertainty is characteristically direct: "Just try to hold onto your hat, jump on there cowboy and ride it and see where it goes."

The message resonates beyond software development. As AI continues evolving, the tension between human craftsmanship and machine efficiency will play out across industries. Hansson's "handwritten code saddles" represent a choice to preserve quality and beauty even as the world rushes toward automation.

For now, the flickering light bulb continues to illuminate intermittently—bright enough to show promise, but inconsistent enough to keep human hands firmly on the keyboard.