Engineering management after the cost of code collapsed
As large language models make it dramatically cheaper and faster to generate code, many engineers argue that the real bottlenecks in software development are now system design, organizational structure, and understanding complex codebases—not typing out implementations. Commenters debate whether AI-written code is maintainable or simply accelerates technical debt, with some reporting major productivity and quality gains when humans focus on architecture and review, and others seeing a decline in reliability and clarity. Underneath this is a broader question of how engineering management, team composition, and metrics should evolve when “plumbing” work is nearly free but context, coordination, and long-term accountability remain costly.
Impact of LLMs on Coding Costs and Bottlenecks
- Many agree the mechanical act of writing code is now cheaper and faster.
- Several argue code-writing was rarely the true bottleneck; coordination, requirements, and understanding complex systems still dominate timelines.
- Others counter that implementation was in fact a major drag, and LLMs remove large amounts of “busywork,” enabling more architectural thinking and iteration.
Code Quality, Maintainability, and Technical Debt
- Strong split: some report LLMs now generate production-quality, well-tested code in many languages; others say outputs are always rewritten and not maintainable.
- Concern that LLMs accelerate accumulation of unreviewed, low-quality “slop” and massive codebases, deepening technical debt.
- Debate over whether LLM-written code can be safely “self-maintained” by future LLMs vs. becoming opaque, un-understandable black boxes.
- Worries that prototypes become de facto designs, with little upfront design work and entrenched poor abstractions.
Best Uses of LLMs in Development
- Broad support for LLMs as code reviewers, refactoring aids, security checkers, and for small scripts or migrations.
- Disagreement on planning: some advocate spec-driven development with LLMs doing plans and code; others find LLM-generated plans vague and unreliable.
- Mixed reports on LLMs’ ability to evaluate architecture or predict change cost; attempts described as project-specific and not comparable across repos.
Organizational and Management Implications
- Many note real problems are team structure, prioritization, and decision-making, not typing speed.
- Some fear management will use AI to justify treating engineering purely as a cost center, ignoring maintenance and long-term quality.
- Others argue good management fundamentals (impact, context, ownership) don’t change; token usage and LOC remain bad metrics.
- Discussion on whether management itself is more automatable than engineering, versus the need for humans to own consequences and shape abstractions and incentives.
Writing, Documentation, and “AI Slop”
- Complaints that AI has collapsed the cost of long-form text, flooding the web with verbose, generic content.
- Several readers found the linked article stylistically “AI-ish,” debated its authenticity, and criticized punchy headings and padded prose.
- Some call for shorter, denser writing and better tools or training to avoid AI-style verbosity.