There's no such thing as a small software team anymore
Claims that AI coding agents will turn every engineering group into a “large team” by enabling hundreds of parallel changes are meeting heavy skepticism. Commenters argue that splitting systems into thousands of microservices to feed agents mainly shifts complexity into operations, makes coordination harder, and risks floods of low‑quality code optimized for PR counts rather than user value. Many see LLMs as powerful tools for individual developers and small, well‑architected monoliths, but warn that overreliance on agents threatens maintainability, product focus, and the long‑term role of human software engineers.
Agents, “small teams,” and productivity metrics
- Several commenters doubt that “small teams” doing 100 PRs/day via agents is meaningful progress.
- High PR/commit counts are seen as easy-to-game vanity metrics, further detached from user value in the AI era.
- Some orgs reportedly now pressure “laggards” to use AI harder, even when AI-generated work has to be quietly redone.
Microservices vs monoliths with AI
- Many argue microservices add operational complexity (deployments, networking, retries, consistency, versioning) without removing inherent product complexity.
- Several say well-structured monoliths with modular code and clear boundaries remain superior, especially because an LLM can hold more context.
- Others claim fine-grained services (even “nano-services”) align better with LLM context limits and agent parallelism, but this is contested.
Quality, correctness, and coordination
- Strong skepticism that agents can safely modify thousands of services: agents lack full-system context and can introduce subtle cross-service breakage.
- Parallel “agent swarms” risk semantic conflicts instead of simple merge conflicts, making integration and release coordination harder.
- Some suggest using AI for maintenance: enforcing invariants, simplifying code, or systematically paying down tech debt – but others report models often ignore invariants as context grows.
Developer experience and job impact
- Experiences diverge: some solo/small-team devs claim dramatic productivity and even entire app rewrites enabled by LLMs; others are effective without them.
- Debate over future roles: from “you’re a prompt engineer or fired” to the view that system design, understanding, and craftsmanship remain essential.
- Concern that seniors’ architectural knowledge being replaced by opaque AI-generated changes could increase outages once experienced staff churn.
Ethical, economic, and societal concerns
- Worries about dependence on external LLM vendors, unpredictable costs, privacy risks, copyright issues, and skill atrophy.
- Some see AI-driven productivity as misaligned with societal needs (“more crud apps,” not “better” software), and driven by investor FOMO and “tokenmaxxing.”
- Others argue AI is a permanent, transformative technology wave; there may be a capex “winter,” but not a return to pre-AI software development.