A Chicken Game With No Brakes: Does Everyone on Earth Know How to Code?

The dissonance between Big Tech's 730-billion-dollar money burning and the flood of AI agents. Right now the global tech ecosystem is caught in a genuine chicken game of "you die, I live." The money Big Tech is pouring in to seize AI leadership has long since left the category of plain "investment." It is a war of firepower, literally piling money into crates and setting it alight.

But what the public witnesses behind this frenzied speed race is genuinely strange. More than 500 AI agent services flood the market day after day, and every one of them shouts the same two things: "coding automation" and "work automation." As if all 8 billion people on Earth live only to touch code lines and build automation pipelines.

This article examines the astronomical scale of Big Tech's AI investment, the real data from the developer market, and the enormous gap that exists between that and actual general-public usage. All figures follow publicly announced guidance and published surveys, and may carry some error depending on exchange rates and aggregation timing.

1. The 730-billion-dollar AI storm Big Tech is burning through

Big Tech's capital expenditure (capex) on AI infrastructure exceeds the annual defense budget or infrastructure budget of entire countries.

Big 4 hyperscaler 2026 capex guidance: the combined 2026 capital expenditure of Amazon, Alphabet (Google), Microsoft, and Meta comes to roughly 725 to 730 billion dollars (about 970 trillion to 1,000 trillion won), a jump of more than 78 percent in a single year from 410 billion dollars in 2025.

Add cumulative investment on top of that, and the money the major seven Big Tech companies poured into AI data centers, GPUs, foundries, and talent acquisition from 2020 through 2026 has surpassed 1.5 trillion dollars (about 2,000 trillion won).

At this point it is fair to say that FOMO, not a cold calculation of return on investment, governs these boards: "if we fall behind now, it's over." The problem is that there is still no evidence that the party on the receiving end of that spending is the party actually making money.

2. Why do some 500 agents shout only about "coding"

According to estimates from market research firms and developer communities, the number of commercial AI agents and frameworks in existence has passed 500. Cursor, Claude Code, Devin, Windsurf, Cline, v0 — new names release daily. The problem is that their marketing copy is word-for-word identical.

"Give it instructions in natural language and the AI writes the script, fixes the bug, and completes the deployment automatically."

There are three reasons.

Measurability. Software development has clear inputs (instructions) and outputs (code, pass/fail). It is by far the easiest domain in which to test and market LLM performance. By contrast, areas like tax filing or hospital appointments have success criteria entangled with law, agencies, and regulations, so vendors have to build the benchmarks and evaluation criteria from scratch before anything else.

Where enterprise spending concentrates. According to a Menlo Ventures survey, developer tools (AI coding) take the largest share of enterprise generative AI B2B spending. A significant portion of Anthropic's and OpenAI's enterprise API revenue comes from coding agents. Products cluster where the money is; that is a plain economic result.

The buyer structure of the earliest customers. The first organizations to purchase agents are internal development teams. Approval paths are short, user counts are measurable, and if it fails the cost of retrying is low. In other words, the early revenue structure is locked into the shape of "developers buy developer tools."

Judging by the developer ecosystem alone, AI adoption is explosive. But that is exactly where Big Tech's narcissistic illusion begins.

3. Illusion and reality, in numbers

The problem is one of ratios. That 84% figure from the developer world is not 84% of the whole population. It is not 84% of the entire enterprise market either. Put the whole population in the denominator and developers are a very small slice, and the tools they use are developer tools.

Coding is 4.2% of what the public actually asks about. In an analysis of 1.5 million real ChatGPT conversations by researchers from Harvard and Duke together with OpenAI, questions about computer programming accounted for only 4.2 percent. The majority of users (49 percent) turn to AI for information search, summarization, sentence writing, and idea generation.

Agent washing. Gartner forecasts that agent capabilities will be embedded in 40 percent of enterprise applications by 2027, while simultaneously warning that more than 40 percent of real enterprise agent projects will be abandoned or discontinued. The causes are soaring API costs, unclear ROI, and agent washing — nothing but a plain chatbot with the signboard swapped to say "agent."

Productivity collision in the field. For repositories handed more than simple code generation and real system maintenance, research (MSR academic analysis) reports code complexity rising 41 percent and technical debt plus error rates climbing 30 percent. Generation gets easier while review stays exactly as it was, so the system drifts in the direction of getting slower. The correct signal for detecting this case is not "how many agents did you attach" but "by how much did the proportion of code reviewed by humans increase."

4. So where is the real market

What the public actually wants is not code generation. That 4.2 percent from above means the other 95 percent-plus are carrying entirely different problems. And most of those problems are not about code at all. They are about procedure and procurement.

The commonalities are three. First, the result can be verified as a number or a state. Second, the points requiring human intervention are clear. Third, the cost saving converts directly into cash. Unlike coding agents, here "human verification" is not a stage but a design element. That difference is one of product design, not of foundation model choice.

Conclusion: 8 billion people are not code authors

The AI competition Big Tech is running with 730 billion dollars is currently trapped inside a closed worldview of "developers build tools for developers, and those tools automate yet other developer tools." Ninety-nine percent of the world's population are not developers, and they do not wake up wanting the code to generate itself. What they want is intelligence that handles their tax filing smoothly, convenience that books a complicated hospital appointment for them, and practical automation that untangles the snarls on a logistics floor.

The flood of coding agents is not evidence that AI technology is universally capable. It is far more likely a symptom of overcapacity — technology that failed to find a business model crowding into the easiest development domain to sell.

The larger the scale of the money being burned, the more the market's question narrows to one.

"Beyond the coding playground that belongs only to developers, what real value will you create in the lives of ordinary people?"

If Big Tech cannot answer that question, the 730-billion-dollar fireworks will leave behind nothing but the most expensive technological bubble in history.

References

• Each company's 2026 capex guidance: from each company's earnings materials
• Stack Overflow Developer Survey; Menlo Ventures, The State of Generative AI in the Enterprise
• Analysis of 1.5 million ChatGPT conversations: joint research by Harvard / Duke / OpenAI
• Gartner report on enterprise agent outlook and the abandonment rate warning
• Rising complexity and error rates in repositories that adopted agents: MSR academic analysis