Agents Over Frameworks: The Return of Real Software Engineering

Advances in coding agents have made automated programming practical, shifting developer effort from typing code to making architectural and product decisions. This undermines the rationale for many frameworks, which add overhead, lock‑in, and hidden design constraints while solving problems agents now handle better. The author urges engineers to ditch unnecessary abstractions, lean on simple tools like Bash and Makefiles, and build only what is truly needed.

Key Points

• Coding agents now effectively automate boilerplate and routine coding, letting engineers concentrate on architecture, trade‑offs, and product design.
• Frameworks mostly add unnecessary abstraction layers that create new problems, incur operational overhead, and constrain product decisions.
• The three supposed benefits of frameworks—simplification, automation, and labor cost—either mask intellectual surrender or are now obsolete due to agent capabilities.
• Agents pair best with simple, durable tools (e.g., Bash, Makefiles), enabling fast, purpose‑built tooling without complex framework ecosystems.
• Engineers should delay complexity until necessary, avoid lock‑in from hyperscaler‑backed frameworks, and reclaim true software engineering practice.

Sentiment

The community is predominantly skeptical of the article's central claim that frameworks are obsolete. While many agree AI is changing how developers work and acknowledge the genuine productivity benefits for individual developers, the majority push back hard on the idea of abandoning established frameworks in favor of bespoke AI-generated code. The most resonant counter-arguments center on frameworks providing correctness, shared maintenance, and community knowledge that cannot be replaced by individual AI-generated solutions. The tone is engaged but often dismissive of the article's sweeping conclusions.

In Agreement

• AI coding agents genuinely reduce the mental toll of programming and make developers more willing to start over with better approaches, effectively lowering activation energy for good engineering decisions
• Frameworks historically served as crutches that prevented engineers from thinking critically about architecture, and AI tools now restore the optionality to build purpose-fit solutions at low cost
• AI excels at automating low-value tasks that developers previously ignored, effectively changing the 'is it worth your time' calculus for custom tooling and small automation projects
• Deep engineering understanding remains essential even with AI — knowing architecture, trade-offs, and edge cases is the irreplaceable human contribution, which validates the article's emphasis on real engineering
• AI as a pair programmer that handles typing while the engineer drives design decisions represents a genuine return to engineering-first thinking over framework-first thinking

Opposed

• The article misses frameworks' most critical function: correctness — frameworks encode battle-tested solutions to security, edge cases, and subtle bugs that AI-generated bespoke code will inevitably reproduce
• Abandoning shared frameworks creates the 'LISP curse': you bear the entire maintenance burden alone, losing community-discovered bug fixes that propagate to all framework users
• Despite widespread AI adoption, companies show zero increase in the rate major projects ship, suggesting productivity gains are overstated and the real bottleneck is problem understanding, not code generation
• AI agents actually work better with frameworks because they're trained on large sets of conventions — vibe-coding in vanilla JS produces unstructured messes while frameworks at least produce structured ones
• The article itself appears written with LLM assistance, which undermines its credibility, and its central claim conflates personal side-project workflow with professional software engineering
• Building bespoke solutions with AI creates harder-to-maintain technical debt because there's no community, no documentation, and no shared understanding — the opposite of what good engineering produces