Why the AI Era Makes Us More 'Efficient' Yet More Exhausted
> "In the AI era, the reason you feel more 'efficient' yet more exhausted is that you are using a set of pre-AI-era, outdated agile work practices to ride the 'speed horse' of AI."
With AI coding assistants like GitHub Copilot and Amazon CodeWhisperer promising to free developers from repetitive coding work, a puzzling trend is emerging: instead of ease and efficiency, many engineers, product managers, and teams feel unprecedented fatigue. This "more efficient, more exhausted" paradox is the starting point of this analysis.
Key points
1. The AI productivity paradox: individual speed-up, overall stagnation
- A McKinsey survey of 300 global enterprises found that despite widespread adoption of Copilot, agents, and LLMs, actual productivity gains concentrate in only 5–15%, far below expectations (McKinsey via daily.dev).
- Root causes:
- Organizational workflows have not evolved in sync with AI efficiency.
- Traditional agile processes have become new efficiency shackles.
- A surging "collaboration tax" (more complex code reviews, more frequent alignment meetings, more rework on AI-generated code) cancels out coding gains.
- Quality risks: AI code may be syntactically correct but weak in logic, boundary handling, and system compatibility — all requiring costly verification and fixes.
- Security risks: AI can generate code vulnerable to SQL injection or XSS, or use insecure third-party libraries, increasing security-review burden.
- Cost overruns: inefficient database queries, unnecessary loops/recursion, and memory-hungry structures inflate cloud bills.
- Code review dilemma: AI floods reviewers with high-volume, "black-box" code whose intent and context cannot be inferred the way a human author's can.
- Fixed cadence fails: when AI completes a user story in hours, fitting it into a two-week sprint is pure management overhead — "iteration for iteration's sake" distorts agile's original intent (Tencent Cloud article).
- Role silos break down: clinging to "developers write code, testers find bugs" creates duplicated work and communication friction.
- Ceremonies become bottlenecks: PR reviews are overwhelmed; daily standups become hollow when AI does most of the work.
- Collaboration costs explode:
- "Review-as-refactoring": reviewers must rework AI drafts toward long-term goals, turning review into a second development pass.
- Context-switching costs grow as AI tools complicate an already complex toolchain.
- Tacit project knowledge is invisible to AI, forcing developers to constantly explain context or fix context-blind mistakes.
- "Writing code fast" is no longer scarce or a core measure of value — AI generates in seconds what once took hours or days.
- Differentiation now lies in the What and Why: understanding business needs, translating vague requirements into clear specs, asking the right questions, and challenging bad assumptions.
- Systems thinking matters more than local features: AI can generate parts, but assembling them into stable, scalable, high-performance systems is systems engineering. As one source puts it, the next-generation developer will be an architect, not a programmer — the future of software is design, not writing (Juejin post).
2. The AI speed paradox: faster code generation, more downstream problems
3. Agile: from remedy to shackle
4. The competency model flip: from executor to thinker and orchestrator
5. The way out
The article argues that exhaustion comes from driving AI speed with pre-AI agile practices. Breaking the deadlock requires redesigning workflows around AI's dynamic output: slimming down ceremony-heavy processes, redefining roles, treating developers as problem definers and system orchestrators, and letting organizations — not just individuals — capture AI's efficiency gains.
*Note: the source page was truncated; sections follow the article's stated structure (AI productivity paradox, agile's evolution, competency model shift, and the way forward).*