Fintech engineer warns of engineering commoditisation as LLMs automate domain knowledge

A software engineer working in the fintech sector has published a detailed follow-up to their viral blog post, "LLMs are eroding my career," outlining the practical implications of large language models on daily engineering workflows. The author argues that the automation of domain-specific knowledge, such as local tax regulations and ledger implementation specifics, is rapidly reducing the need for human input and eroding the traditional competitive advantage held by experienced developers.

The engineer describes a workplace environment where management has recommended accelerating design documentation with AI, a move they characterise as careless for a business handling money. To maintain quality control, the author has implemented several workarounds, including keeping documentation generic on implementation details to allow for thoughtful coding, splitting sensitive implementation parts into smaller tickets for cautious review, and adding end-to-end test tickets to identify bugs before release.

A central concern raised in the post is the commoditisation of engineering roles. The author notes that domain knowledge, which previously set them apart from coders who only understood syntax, is now "promptable" using tools like ChatGPT Pro. This shift has reduced the need to consult senior colleagues for details on accounting processes or ledger specifics, leading to a situation where less human input is required to complete tasks.

The post draws direct parallels between the current disruption in software engineering and the earlier displacement of copywriting and UX writing roles. The author points out that large organisations have laid off UX writers because AI can generate acceptable text labels 90 per cent of the time, effectively allowing one professional to do the work of ten. They argue that while demand for software may have an upper limit, the supply of AI-generated content is increasing, leading to a market where only the top tier of professionals remain employable.

While the author identifies as an "AI-native engineer" who is committed to improving agentic tooling and using adversarial code reviews, they warn that the profession is heading toward being commoditised. They reference Turing AI, a company hiring engineers to write code for reinforcement learning, suggesting that the "human moat" on engineering principles may not last. The author contends that this technological shift is more profound than previous industry waves, such as the adoption of object-oriented programming, and could have significant implications for other knowledge-based sectors including finance, biology, and law.

The engineer dismisses comparisons to past technological cycles, such as the rise of object-oriented programming in the 1990s and 2000s, arguing that those waves did not make knowledge promptable or show signs of fast, compounding improvements aimed at replacing workers across multiple fields. They urge the industry to recognise the scale of the current change, likening the dismissal of the threat to the initial underestimation of the impact of the pandemic, and caution against using bad past examples to predict the future.