Detailed Analysis
A podcast conversation featuring four working engineers across four countries — a Google Developer Advocate in Stockholm, a Senior Software Engineer and consultant in Paris, an NVIDIA Deep Learning Institute Instructor in Morocco, and an IBM Infrastructure Engineer in Dublin — surfaces a nuanced and geographically diverse perspective on one of the most pressing career questions facing new computer science students in 2026: whether software engineering remains a worthwhile professional path in an era of rapidly advancing AI tooling. The consensus, while not uniformly optimistic, stops well short of declaring the field obsolete. Instead, the engineers converge on a more conditional answer — the career remains viable, but the terms of entry and the nature of the work are shifting in ways that reward foundational understanding over surface-level productivity.
The most practically significant advice from the discussion comes from the IBM infrastructure engineer, who draws a clear line between using AI as a cognitive aid versus using it as a crutch. The directive to never commit code one cannot explain represents a professional standard that cuts against the dominant workflow emerging in many development shops, where AI-generated code is reviewed superficially and pushed to production at speed. The Senior Software Engineer reinforces this framing by characterizing current LLMs as architecturally immature — "babies" — and warning that engineers who lack deep systems knowledge will lose the ability to critically evaluate AI output as the technology evolves. The Google advocate's framing is more structurally candid: coding itself has become a commodity in the way that manual labor in manufacturing did after industrial automation, which reframes the question from technical skill to vocational intention and specialization.
The discussion's most economically consequential thread involves Anthropic and the broader inference industry's financial sustainability. The podcast notes that at current usage levels, Anthropic is likely losing money on its heaviest Claude subscribers — a claim consistent with widely reported dynamics across the AI services sector, where compute costs for large language model inference remain stubbornly high relative to subscription revenues. The Google advocate, who works on LLM inference at scale, frames this as a systemic constraint rather than a company-specific problem: without substantial efficiency gains in inference over the next five years, the cost structure of the current AI ecosystem becomes untenable. This is a materially important observation because much of the optimism about AI's role in software development is predicated on continued and affordable access to frontier model capabilities — an assumption that has not yet been validated by unit economics.
The booing of a Silicon Valley executive's "AI is the next industrial revolution" speech by university graduates introduces a sociological dimension that is easy to underweight. The crowd reaction reflects a growing divergence between how AI is narrated by industry leadership and how it is experienced by people entering a labor market already disrupted by it. For prospective engineers, the rhetorical celebration of AI transformation often obscures what that transformation means in practice: fewer junior positions, higher expectations for demonstrated architectural knowledge, and significant uncertainty about which roles will be automated away in the medium term. The engineers in the discussion implicitly acknowledge this tension, offering advice calibrated to survival and differentiation rather than uncomplicated optimism.
Taken together, the conversation reflects a maturing discourse around AI's role in software engineering — one that has moved beyond binary arguments about replacement versus augmentation toward a more granular reckoning with infrastructure costs, skill stratification, and the professional ethics of AI-assisted development. The inclusion of NVIDIA's perspective on cost-effectiveness hints at an industry-wide recognition that the current phase of AI development is as much about making the economics viable as it is about expanding technical capability. For students choosing whether to enter software engineering, the implication is that the field remains genuinely promising, but only for those willing to engage with it at a level of depth that AI tooling cannot yet substitute for.
Read original article →