The “don’t learn to code” message was incorrect.
For two years the pitch has been loud: AI will write the code, so why bother learning? Claude is building apps. Copilot completes your sentences. Tutorials are done.
That story is already breaking out. Companies are investing heavily in AI and getting almost nothing back BCG’s January 2026 survey of 2,360 executives found companies plan to more than double AI spending (from 0.8% to 1.7% of revenue). 94% of CEOs say they will keep investing even if they don’t see returns this year.
The results are weak on the ground. Only 31% of firms claim to have achieved any meaningful success in scaling generative AI beyond pilots. 78% of C-suite leaders expect significant returns in 18 months. Only 23% of middle managers feel prepared to deliver. BCG’s diagnosis is blunt: technology is just 30% of the work. The other 70% is people – new skills, new workflows, new ways of judging performance. Most organizations spend too much on tools and not enough on the humans who have to use those tools.
The Age of Cheap AI is already dead.
The pricing at the beginning was never sustainable. Providers subsidized usage to gain market share, then raised prices once users were hooked.
Anthropic quietly more than doubled the published cost of Claude Code in April 2026, from $6 to $13 per developer per active day (heavy users can hit $30). That one tool alone costs enterprise estimates $150-$250 per developer per month. Microsoft is shifting GitHub Copilot to token billing, and has already slashed cheaper plans Uber’s CTO said the company had spent its entire 2026 AI budget by April, mostly on Claude Code and Cursor at $500-$2,000 per engineer per month.
Relying only on these tools, every price increase and change in access will directly affect you. If you can code, you maintain leverage: local models, open-source alternatives, the ability to build and fix your own stack.
The people building AI still say you need to know code.
OpenAI co-founder and former Tesla AI director Andrej Karpathy coined the phrase “vibe coding”, prompting and accepting output without understanding it. Good for throw-away weekend projects. Good for nothing that matters. Today the real work is in steering AI agents, reviewing what they generate, and making decisions. You can’t manage what you don’t understand.
Replit’s Amjad Masad puts it this way: learn to think, break down problems and communicate clearly. That's exactly the skills coding builds.
AI is good with templates. Still weak in deep debugging, full system context & security. The developer who understands the fundamentals can tap into AI to go massively faster. If you know how to prompt and that's it, you're just one opaque error message away from being stuck.
All software companies use AI, but they still need developers.
According to a survey by GoodFirms, 90.6 percent of software companies already use AI tools in the development lifecycle. That doesn’t eliminate the need for developers. It just creates more demand for people who can make the tools productive. Cheaper, faster software expands the software market. Cloud AI is getting pricey, and you still need coding chops to use them, so open-source local models are popping up just for that reason.
All built by people that are leaving
The “don’t learn to code” message scared an entire generation, and experienced engineers are retiring. The Bureau of Labor Statistics predicts that by 2027, 18% of senior developers born between 1970 and 1980 will be gone. The loss of a single principal engineer usually requires two senior replacements that simply aren’t available in sufficient numbers. The developer shortage for 2026 is expected to be 40% worse than 2025. The BLS projects that the U.S. will have 1.4 million unfilled computing jobs by 2027; IDC sees the gap nearing 4 million by decade’s end.
Surplus illusion from headlines about layoffs. The reality is a dwindling pool of people to own complex systems and run AI in production.
Discouragement made the first
The story worked. Bootcamps and computer science programs saw a dip in interest. A whole cohort stepped aside. Meanwhile, the number of job postings related to AI rose 300% from 2023 to 2025, while the supply of qualified engineers grew at a much slower rate.
The market has now bifurcated: too many juniors fighting for entry-level jobs and a real shortage of people who can lead AI, own systems and deliver under pressure.
There are two types of people in digital world. Consumers. And architects. Consumers pay whatever the platforms charge, and they have no recourse if the access changes. Developers are the architects. They can build local, use open-source models and stay in control.
Now is the best time to start
Not because it is easy or because there is a job waiting in six months. Because the math has worked in your favor. The new-developer pipeline has been thinned by bad advice. More senior talent is leaving than is being replaced. Companies that have bet everything on “AI will do it” are discovering they still need humans who can steer, assess, correct and build on the output of the models.
The same tools that cost Uber thousands per engineer per month can teach a novice, explain concepts repeatedly and give line-by-line feedback. Never has the learning curve been shorter. Once you have real working knowledge, those tools make you far more productive than any developer from a decade ago.
BCG itself finds talent gaps as the biggest obstacle to achieving the expected returns from AI. But that's no reason to stay away. That is the strongest reason for going in.
Learning to code is no longer just about getting a job in tech. This is about digital self-reliance, about being a non-passive user in a world where platforms can raise prices and restrict access at will.
The hype said it was not worth it. The data, the economics, and the talent numbers all tell a different story.