Baldur Bjarnason once built a career around teaching others how to build for the web. He wrote books. He ran training programs. He consulted and produced content that helped thousands grasp HTML, CSS, JavaScript and the shifting frameworks that power modern sites. Then the bottom fell out.

His training projects dried up one after another. Peers who sold courses and workshops shut their businesses. Demand for web development ebooks cratered. “All my training and education projects had dropped off one by one,” he wrote in a recent post on his site. Most of his peers who had been selling courses and training were shutting down their businesses. The demand for web dev ebooks had cratered.

Software developers haven’t all switched to using agents for their coding. But they do seem to have all switched to error-prone, backwards-facing, nondeterministic chatbots for what now passes as web dev education. The observation, drawn from conversations across the independent education scene, points to a stark new reality. Generative AI has gutted the market that once supported those who explained the web.

Axel Rauschmayer felt it sharply. One of the most prolific writers on JavaScript and TypeScript, he watched book sales plunge. Income from those sales went from enough to live on in 2024 to zero in 2026, he reported in the same discussion. Traffic to his blog and free online books increased beyond what he could afford. Virtually all of it comes from AI crawlers. No ad income followed. He eventually removed much of the content to reassess his options.

Josh W. Comeau sells online courses, research articles and interactive tutorials on web technologies. He heard similar stories from other creators. Revenue fell more than 50 percent. People stopped engaging with the content. They turned instead to large language models that regurgitate the work without consent or compensation. Comeau argued the shift removes any incentive to create high-quality free material. The output of that effort simply trains competitors that pay nothing. His comments, echoed across forums and republished by LavX News on October 2, 2026, capture a widespread frustration.

Kyle Cook, host of the YouTube channel Web Dev Simplified, saw his programming tutorial views and income drop by roughly half. He described the economics plainly. Making a video about the latest AI model takes less time, earns more money and attracts more views than teaching programming fundamentals. The pattern holds across platforms. Creators who once built audiences through patient explanation now compete with tools that synthesize their knowledge instantly.

But the damage runs deeper than lost income. It threatens the quality of future developers. Universities and boot camps built curricula around the idea that writing code represented the central challenge. That assumption no longer holds. With generative AI, code production is no longer the bottleneck. The real difficulties sit before and after that step: specifying requirements with precision, evaluating generated output, understanding tradeoffs, verifying correctness, and maintaining systems over time.

Communications of the ACM explored this shift in its August 2026 article “Computing Education When Writing Code Is No Longer the Challenge.” Authors noted that most current curricula rest on the notion that writing code is the hard part. GenAI changes the equation. Educators must now redesign objectives, assignments, assessments, courses and entire programs. Some suggest holding the line by banning AI tools for another generation. Others call for urgent reinvention. The piece argues for haste in updating introductory programming courses while extending changes across the curriculum.

Similar warnings appear in research. A Russian-language conference paper from August 2026 examined problems of teaching web development in universities amid widespread AI assistants such as GitHub Copilot, ChatGPT and Cursor. It identified two clusters of issues. Didactic problems include superficial learning, the illusion of competence, and difficulties in assessment. Security problems encompass vulnerabilities in generated code, hallucinated packages that enable supply-chain attacks, data leaks through prompts, and prompt injection risks. Roughly 40 percent of code generated by Copilot in vulnerable scenarios from the CWE Top 25 list contained exploitable defects, the authors cited from earlier studies.

Yet job market data tells a more nuanced story. The U.S. Bureau of Labor Statistics projects web developer and digital designer jobs to grow 5 percent from 2025 to 2035, faster than average for all occupations. Median pay stood at $92,650 in May 2025. Globally, developer numbers keep rising. GitHub reported more than 36 million new developers joined in 2025, up 23 percent year over year. And 84 percent of developers now use or plan to use AI coding tools, according to a roundup of 60-plus statistics published by Colorlib on October 1, 2026.

Sixty percent of developers, along with majorities of executives and recruiters, say a degree should not be required for tech roles. AI fluency has become baseline. Nearly 80 percent of new GitHub users try Copilot in their first week. The role itself evolves toward higher-level architecture, AI integration and full-stack expertise. Human judgment remains central. Forty-six percent of developers still distrust the accuracy of AI output.

Entry-level positions have tightened. Unemployment for computer science graduates rose to 7 percent in 2024 from 6.1 percent the year before. Underemployment exceeded 19 percent. Tech job postings on Indeed fell 36 percent between 2020 and 2025. More than 600,000 U.S. tech workers lost jobs since ChatGPT appeared, according to Layoff.fyi data cited in an August 2026 WebProNews analysis titled “Why ‘Learn to Code’ Feels Dated in 2026 as AI Rewrites Software Careers.” Google reported that 75 percent of its new code came from AI by April 2026, up from 50 percent the previous year. Demand for AI skills in entry-level postings jumped from 10.5 percent in fall 2025 to 16.5 percent by spring 2026.

So what replaces the old model of web development education? Some institutions experiment with new formats. arXiv papers from September 2026 propose judgment-centered software engineering education and AI-augmented frameworks that emphasize prompting, orchestrating, validating and accounting for AI output. One framework outlines five non-linear integration levels with obligations to explain, verify, modify and account for decisions. The scarce skills now center on decomposition of problems, system design, evaluation of tradeoffs, debugging complex behavior and determining whether a system is correct, secure, maintainable and fit for purpose.

Boot camps once promised fast entry through intensive practical training. Many focused on current frameworks and portfolio projects. Yet the same economic pressures hit them. Revenue at some providers grew modestly in 2026 after stronger gains the prior year, but the overall industry faces headwinds. Discussions on Hacker News threads linked to Bjarnason’s post revealed mixed views. Some founders reported their own businesses holding steady or growing slightly. Others warned that much of what ed-tech companies taught had become counterproductive, overemphasizing complex tools at the expense of core HTML, JavaScript, CSS and DOM understanding.

Traditional four-year degrees offer theory and broader foundations. They carry higher cost and longer timelines. Boot camps deliver faster, cheaper entry with strong placement rates in normal times. Both models now wrestle with the same question: how to teach when AI can generate code faster than any student. Some programs integrate AI tools deliberately, teaching verification alongside generation. Others maintain strict no-AI policies to build fundamental understanding first.

Independent creators who supplied the free and low-cost layers of the education stack face the hardest squeeze. YouTube views, newsletter subscriptions, ebook sales and course enrollments all suffered. The incentive structure that once rewarded clear explanation and generous sharing has weakened. When models trained on public content compete directly with their creators, the economics turn hostile. Traffic arrives, but it brings no revenue and consumes bandwidth.

Comeau captured the bind. Without compensation or consent, the system discourages the very content that improves the models and educates the next cohort. High-quality material still appears, but the volume and sustainability of that output look uncertain. Some creators pivot to topics around AI itself. Videos explaining the newest models perform better. Others leave the field. A few experiment with paid communities, private newsletters or enterprise training that values human insight over generated text.

The web itself grew from open exchange. Documentation, blog posts, Stack Overflow answers and conference talks formed a commons that accelerated progress. That commons fed the training data for today’s models. Now the models threaten to exhaust the supply. Without fresh, accurate, human-curated material grounded in real experience, the quality of AI assistance may stagnate or degrade. Slop, hallucinated packages and subtle security flaws already appear in generated code at scale.

Industry leaders and educators search for answers. Some propose new business models: sponsored deep technical series, certification programs tied to verification skills, or corporate-funded open content that treats education as infrastructure. Others advocate for licensing changes or technical measures to limit AI crawling of educational sites. Practical success remains limited so far.

Universities ponder curriculum overhaul. Introductory courses could shift emphasis from syntax to specification, testing, architecture and critical evaluation of AI suggestions. Advanced classes might explore human-AI collaboration, technical debt accumulation in AI-augmented codebases, and governance of automated systems. The ACM piece urges speed. Institutional inertia, however, runs strong. Policy changes, faculty training and assessment redesign all take time.

Students sit in the middle. Many already use ChatGPT, Claude or Cursor for assignments. Some gain speed. Others develop the illusion of competence while missing core concepts. Assessment becomes harder. Traditional exams and take-home projects leak answers easily. New methods such as oral defenses, in-class live coding, portfolio reviews with detailed explanations, or AI-specific tasks that require modification and justification of generated code gain attention.

Web development retains appeal. It offers tangible results, creative expression and relatively quick feedback compared with other engineering disciplines. Demand for skilled practitioners who understand browsers, networks, performance, accessibility and security persists. The difference lies in what skill means in 2026. The ability to prompt effectively matters. The discipline to review, test, secure and maintain what emerges from those prompts matters more.

Bjarnason’s post and the chorus of responses it provoked serve as an alarm. The old education economy for web technologies has broken. Creators who supplied knowledge, examples and patient walkthroughs find their work commoditized and their livelihoods undercut. The content that once taught humans now primarily trains machines. And the machines, for all their speed, still produce code that requires human oversight to avoid costly mistakes.

Rebuilding the system will demand fresh thinking from universities, boot camps, corporations, independent creators and the open-source community alike. It will require new incentives, updated pedagogies and perhaps new economic arrangements that value verification, judgment and long-term stewardship over raw code generation. The web isn’t going away. The question is whether the knowledge required to build and sustain it can survive the transition.

Recent coverage in WebProNews from August 2026 and the Colorlib statistics compilation from October 1 reinforce that the pressure is not temporary. AI adoption accelerates. Job requirements shift. Fundamentals endure, but the path to mastering them has changed. Educators and learners alike must adapt or risk being left with tools they no longer fully understand.