Why Do We Feel More Exhausted the More 'Efficient' We Get in the AI Era?

Forum topic · ✨步子哥 · 2025-12-23

Summary

This article from zhichai.net explores the productivity paradox of the AI era: although AI coding tools like GitHub Copilot have dramatically accelerated code writing, real-world efficiency gains remain modest. Citing a McKinsey study of over 300 companies, it reports that actual productivity improvements after adopting AI tools are only 5–15%, far below the promised multiples. The root cause, the author argues, is not the weakness of AI itself but the stagnation of the 'container' around it—workflows, team structures, and operating models still based on decade-old Agile practices. Bottlenecks have shifted from coding speed to decision latency, architecture friction, team design, and evolving quality requirements, while collaboration, rework, alignment, and explanation costs have surged. The article describes a paradigm shift in which engineers and product managers evolve from executors into orchestrators and supervisors of AI—defining problems, directing AI work, and ensuring quality. It concludes that organizations must redesign their ways of working for rapid learning and end-to-end accountability to translate AI's raw capability into genuine productivity.

Why Do We Feel More Exhausted the More "Efficient" We Get in the AI Era?

*A deep dive into ways of working and organizational models in the AI era.*

> "Perhaps the problem is not that you aren't working hard enough, but that you are working with an outdated 'Agile way of working.'"

With the rapid development of AI—especially large language models (LLMs) and code generation tools like GitHub Copilot and Amazon CodeWhisperer—software development is undergoing an unprecedented transformation. These tools promise to free developers from tedious, repetitive coding tasks. Yet a puzzling, widespread phenomenon has emerged: although AI has genuinely made "writing code" faster, many engineers, product managers, and whole teams feel more exhausted, entangled in more complex collaboration, and less fulfilled.

The Efficiency Paradox

In a recent McKinsey talk, "Moving away from Agile: What's Next", a survey of 300 companies revealed a stark reality: even after introducing Copilot, agents, and large models, most organizations' real efficiency gains are only 5%–15%—a sharp contrast to the multiples promised. The reason: teams are still rushing stories in two-week sprints, writing PRs, and waiting for reviews.

The root cause is not that AI tools lack power, but that the "container" carrying them—our ways of working, team design, and operating models—has barely evolved in the past decade. We still use Agile processes designed for a previous era, and every stage of that old pipeline becomes a new bottleneck once AI multiplies coding speed.

Amplified Bottlenecks in the Traditional Process

Surging collaboration costs — AI-generated code requires more cross-role coordination
Soaring rework costs — AI code may deviate from business logic
Increased alignment costs — unified AI code review standards are needed
Rising explanation costs — AI code logic must be explained to teammates

The Agile Dilemma

> "An uncomfortable truth about traditional Agile: we built 'Agile' primarily for 'predictable delivery,' not for 'rapid learning.'" — Dr. Thomas R. Glück

Bottlenecks Have Shifted

| Traditional bottleneck (pre-AI) | New bottleneck (post-AI) | Core challenge | |---|---|---| | Coding speed | Decision latency | Defining problems and priorities quickly | | Human execution | Architecture & platform friction | Integrating AI code into existing systems | | Task management | Team design | Establishing end-to-end ownership | | Delivery speed | Evolving quality definition | Ensuring reliability, security, auditability |

Limits of traditional Agile: designed for predictable delivery rather than rapid learning; stagnant ways of working and team design; no coverage of AI-era bottlenecks.

New requirements in the AI era: capacity for rapid experimentation and learning; team design with end-to-end ownership; a rethinking of what "quality" means.

Paradigm Shift: From Executor to Orchestrator and Supervisor of AI

As AI evolves from an assistive tool into a core productive force, engineers and product managers are undergoing a profound role transformation—from traditional "executors" to higher-level "conductors" and "supervisors."

Traditional role (executor):

Hand-writes every line of code
Focuses on implementation details
Measured by output volume
Follows given requirements and designs

New role (orchestrator & supervisor):

Defines problems and direction
Orchestrates AI work
Ensures output quality
Applies systems thinking and insight

Driving forces of the shift: AI automating coding tasks; a productivity paradigm shift; the move from "how" to "what"; reshaped standards of value.

Conclusion

The exhaustion of the AI era stems not from insufficient effort but from applying last-generation workflows to next-generation productivity. Realizing AI's promise requires redesigning organizations for rapid learning, end-to-end accountability, and a redefinition of quality—while individuals transition from writing code to directing and supervising AI.

*Source: McKinsey QuantumBlack — Moving away from Agile: What's Next*