A Programmer’s Confession: My Ten‑Year Career Pillars Are Crumbling—What’s Next?

A software engineer with ten years of experience describes his career as built on three pillars: deep domain knowledge in finance and payments, expertise in debugging distributed systems, and a commitment to high‑quality code and architecture.

First Pillar: Domain Expertise

After joining a fintech company that fully embraces AI, he received enterprise accounts for ChatGPT and Claude and was encouraged to use them for research, design, and coding. He was tasked with refactoring a legacy online payment system and was asked to produce a design document that both engineers and product managers could understand, avoiding a deep‑technical dive.

Despite minimal AI assistance, he completed the document, initially dismissing large language models as “random parrots.” His manager later pressured him to use AI more, noting that while his code delivery speed was good, the time spent on documentation was excessive.

He realized that the knowledge he had accumulated—trade‑off decisions, settlement mechanisms, idempotency for preventing duplicate charges—was being rendered less valuable because LLMs could synthesize these concepts from publicly available articles and documentation.

Second Pillar: Debugging and Distributed Systems

He notes that LLMs began excelling at writing code after the Claude Code surge in late 2025, followed by tools like Codex. Although initially skeptical about LLMs completing full implementations, he started integrating more AI into the coding process.

He observes that while LLMs improve at coding, they still cannot reliably debug the chaos they introduce. However, newer models (Claude 4.5, 4.6, 4.7, GPT‑5.5, Opus 4.8, DataDog MCP) dramatically increased defect‑resolution rates: Claude 4.5 solved about 60 % of defects given a Sentry link, and later tools allowed him to resolve roughly 90 % of cross‑system defects—including race conditions, edge cases, third‑party integration issues, and undocumented API boundaries—without manual intervention.

Despite this, he still performs code reviews and operates the “robots,” but his unique domain expertise no longer provides a competitive edge; any senior engineer using LLMs can match his performance.

Third Pillar: Code Quality and Architecture

He continues to value clean code, DDD, hexagonal and clean architecture, and enjoys discussing trade‑offs. However, he notes that AI‑generated code often violates SOLID principles, introduces circular dependencies, repeats code, and adds irrelevant comments, leading to a decline in overall code “taste.”

He argues that the industry is shifting toward code organized for machines rather than humans, accepting C‑ or D‑grade code because it is primarily consumed by LLMs.

Future Outlook

He acknowledges that his current role remains secure for now, but the long‑term outlook is uncertain. The erosion of domain knowledge value, combined with AI’s rapid improvement, suggests that software engineering may become commoditized. He compares this to the collapse of copywriting jobs after AI tools saturated the market, emphasizing that demand for many knowledge‑work roles is finite.

He concludes that the only viable path may be to focus on areas where LLMs struggle—creative problem‑solving, novel research, or domains requiring deep human intuition—while recognizing that even these may eventually be encroached upon.