The Economist (August 30, 2026) examines why programmers are the only profession that uses AI at the level of daily work on a mass scale, and whether anyone else can catch up. The publication identifies four structural factors that make code an outlier in the AI market and shows why the answer to the question of catchers determines the sustainability of the entire current investment cycle in data centers.
What Happened
The core of the article is economic, not technical: it contains no new models, benchmarks, or methods. The Economist names four factors that make programming special: the volume of open training data (open-source code makes up nearly 1/5 of training datasets), the simplicity of output verification (tests), the availability of context in digital form (codebases and APIs), and the culture of early adoption among the engineers themselves. The publication supports this with market figures: according to Stack Overflow, 4/5 of developers use AI tools for code; the combined ARR of four AI coding startups — Cognition, Cursor, Lovable, and Replit — grew from approximately $800 million in June 2025 to approximately $6 billion; and according to SemiAnalysis data for Q2 2026, coding accounts for more than half of the combined ARR of OpenAI and Anthropic.
Context
The background for the analysis is set by the economics of the data centers currently under construction. Today, the total annual revenue from AI use is estimated at approximately $150 billion, but according to The Economist's calculations, for data center construction to pay off, this figure must grow to $2.5 trillion by the end of the 2020s. This means that demand for AI must go far beyond code development, which is exactly why the publication is looking for which profession can catch up with programmers and shows how high a bar the coding segment sets.
Why This Matters for the Industry
For the industry, the sustainability of the current investment cycle depends on whether demand for AI will spread beyond programming. The nearest candidates are already showing growth: in law, Harvey, Clio, and Legora doubled their combined ARR in a year to approximately $1 billion; in finance, Rogo added 100 enterprise clients and increased ARR by 50% in a quarter; in customer service, Sierra reached $200 million ARR, and $3 billion was invested in AI for this sector this year — more than in any other AI application category. However, all these verticals are structurally lagging: they have little public data, no cheap automatic verification of results, and a significant part of knowledge exists outside digital format. In coding, the market has already moved to consolidation: OpenAI's and Anthropic's own tools are taking a dominant position, and investments in independent coding startups have stalled, so the main opportunity is shifting to building solutions on top of laboratory tools and creating an output verification layer in adjacent niches.
Why This Matters for Users
For the reader, the article provides a simple diagnostic checklist: the volume of open data, testability of output, digital availability of context, and the culture of early adoption in the profession. If your field meets these criteria, it has a chance to become the next mass niche for AI; if not, the main costs will be the verification layer and context integration. For developers, the conclusion is that their segment remains the most mature in the market: the niche has passed the pilot phase and is turning into infrastructure for the entire industry. For specialists from other fields, the practical meaning is that the obstacle to AI adoption in their work is often not the quality of the model, but the lack of automatic verification of results and public data.
What Is Still Unknown / Limitations
The $2.5 trillion figure is an extrapolation from data center construction payback, not a technical forecast: the article has no methodology, scenarios, or connection to measurable progress in model capabilities. The cited ARR metrics confirm the scale of demand, but say nothing about the level of tasks solved, the error rate, or the degree of work automation. It is still unknown whether a second niche with coding characteristics will appear and whether the problem of output verification in non-coding domains will be solved.
Sources
Author
Look at AI, editorial team