A software engineer with 10 years of professional experience — most of it in finance, bookkeeping, and payment processing — has published a personal account of how large language models (LLMs) are rendering his accumulated expertise obsolete. The post, titled "LLMs are eroding my software engineering career and I don't know what to do", appeared on the blog human-in-the-loop and offers a stark warning for enterprise technology leaders who rely on deep domain knowledge in their engineering teams.
The First Pillar: Domain-Specific Knowledge
The engineer began his career in frontend web development, quickly moved to backend, and eventually worked on systems involving PCI compliance, double-entry ledgers, escrows, reconciliation, payment lifecycles, and bank transfer idempotency. He believed that specializing in this financial domain would differentiate him in a field increasingly demanding domain experts. However, after being hired by a finance-focused company that provided him with ChatGPT Enterprise and Claude Enterprise accounts on day one, his perspective shifted.
His first major project was redesigning a legacy online payment system. The company required "Design Docs" — architectural overviews readable by both engineers and product managers. Initially, he wrote the first one with minimal AI assistance, even calling LLMs "stochastic parrots." But his manager told him he was taking too long and should use more AI. Reluctantly, he complied. To his surprise, the models — despite being less capable than current versions — were able to connect the dots on how to structure complex systems. He realized: "Everything, was becoming useless. Even though the models still needed some steering, they could connect the dots on how to structure such systems, which was the hardest part that only develops in your brain after years of hands-on experience."
The Second Pillar: Debugging and Distributed Systems
Convinced that debugging — especially of race conditions and distributed systems in production — would remain a human stronghold, the engineer soon found that pillar crumbling as well. He described an incident involving network timeouts caused by misconfigured nftables rules. Claude (the Anthropic model) helped debug the issue by reading logs and deducing the root cause. In an "LLM hackathon" at the company, AI agents were used for QA and backend tasks with decent code quality.
Implications for Enterprise Technology Leaders
| Traditional Value Driver | How LLMs Now Address It |
|---|---|
| Years of domain-specific design knowledge (idempotency, escrows, reconciliation) | Models synthesize trade-offs from available training data and documentation |
| Debugging complex distributed systems (race conditions, network issues) | AI agents read logs, run commands, and present root-cause analysis |
| Learning new technologies and frameworks | Models can learn from documentation faster than humans can |
The engineer concludes that his 10 years of experience now feel "almost 0." He is not excited to learn new technology because he expects LLMs to master it quickly. He loves coding but is considering leaving software engineering altogether, unable to see a path forward in management. Part of him hopes LLMs will hit a wall, but acknowledges that could be wishful thinking.
For CTOs and digital transformation leaders, this account is not just a personal lament — it is a signal about talent retention and the changing nature of technical work. The engineer’s story suggests that companies which fully embrace LLMs may inadvertently devalue the very expertise they once prized. Leaders must consider how to redefine career paths, reward judgment over rote knowledge, and build teams where human specialists complement — rather than compete with — AI tools. The erosion of domain expertise may be inevitable, but the way enterprises manage that transition will determine whether they retain the engineers who built their most critical systems.
Sources:
