Why AI Coding Tools Are Making Developers More Exhausted: Reflections on Autonomy from Claude Code to GLM-5.2

A Developer's Declaration of Freedom

In a Reddit developer community, a user posted an emotionally charged thread with a blunt, powerful title: "I'm FREE! No more babysitting… GLM-5.2." What looked like a casual share actually struck at a widely overlooked yet very real pain point in today's AI coding tool landscape — the "babysitting" dependency developers develop on AI programming assistants.

The developer openly admitted that he started with Claude Code and once loved OpenCode, but after extended use, found himself feeling "trapped" by both tools. Claude Code in particular (which he bluntly called "the worst") actually drove his productivity down instead of up.

Why AI Coding Tools Can Actually Kill Productivity

This is a counterintuitive observation. AI coding assistants are designed to boost efficiency — so why do some developers end up feeling less productive? The author offered two very honest, psychology-rooted explanations.

The Anxiety of Never Being Able to Step Away

The first reason is the constant supervision burden that comes from a lack of trust. The author mentioned that he couldn't bring himself to walk away from the computer because he was afraid the AI would "go off the rails." When AI autonomously executes code changes and file operations during long-running tasks, developers are forced into the role of a full-time babysitter, eyes glued to the screen. Even without writing a single line of code themselves, they remain in a state of high alert — the mental load never actually lightens.

The Dopamine Trap and Attention Drain

The second reason is more subtle. The author mentioned that the AI's instant feedback loop "fed his ADHD brain." Every AI response, every auto-generated snippet, functions like a small dopamine hit — keeping you involuntarily glued to the screen in an addictive cycle not unlike doom-scrolling. For users who already struggle with sustained attention, this "stickiness" baked into the interaction design becomes an enemy of productivity rather than an ally.

What Changed with GLM-5.2

The author described his experience with GLM-5.2 as a kind of liberation. While the post doesn't dive deep into GLM-5.2's specific technical advantages, reading between the lines, what actually set him free wasn't some leap in model parameters — it was a shift at the level of workflow and interaction experience.

He specifically highlighted one key design feature: ETA (Estimated Time of Arrival). His exact words: "It's not perfect, but what progress bar is?" Behind that quip lies a profound product insight: AI tools need to give users a sense of predictable control. When a developer knows roughly how long a task will take, they feel safe stepping away to do something else — instead of being tethered to the screen.

The Overlooked UX Problem in AI Coding Tools

This developer raised a point worth serious reflection across the entire industry: "I'm going to bring solid UX principles back to AI, because the stuff out there is broken."

Balancing Autonomy and Control

Today's mainstream AI coding tools (like Claude Code) optimize for maximum autonomous execution — letting the AI power through complex tasks in one go. But raw autonomy without transparency and predictability translates directly into user anxiety. The ideal AI coding assistant should strike the right balance between letting the AI run freely and keeping the user in control.

Why Progress Visualization Matters More Than Ever

ETA indicators and progress bars — these seemingly old-fashioned UX elements — are actually more valuable than ever in the age of AI. Humans have a low tolerance for uncertainty, especially in contexts that require trust. Even an imprecise progress indicator beats a total black box. This is decades of accumulated UX wisdom from software engineering — wisdom that many product teams have quietly abandoned in the race to showcase AI capabilities.

A Sentiment That Resonates Widely

At the end of his post, the author asked: "I'm not alone in feeling this way, right?" That rhetorical question voices what many heavy AI tool users feel but rarely articulate.

As AI agents grow increasingly capable of executing longer, more complex autonomous tasks, the question of human-AI trust will only become more pressing. Developers are slowly realizing that the measure of a good AI coding tool shouldn't be limited to how smart the model is — it should also include whether it makes you feel safe and in control while using it.

Worth noting: this is a subjective experience shared by a single Reddit user. His praise for GLM-5.2 and criticism of Claude Code carry strong personal bias and should not be taken as a rigorous comparative benchmark. Whether GLM-5.2 objectively outperforms its competitors still requires broader, independent testing. But the real value of this post isn't the product comparison — it's that it precisely captures a real and important direction in the evolution of AI coding tools.

The Next Battleground for AI Coding Tools

While vendors are still competing on model scale, context window length, and coding accuracy, this ordinary developer's "declaration of freedom" reminds us: the competition in AI products is shifting from a pure capability race to a race of experience and trust.

Whoever can let a user confidently walk away to grab a coffee while the AI works has truly solved the core productivity problem for developers. Transparency, predictability, a sense of control — these solid UX principles shouldn't be discarded just because AI has arrived. If anything, they deserve to be championed more loudly than ever. That may well be the only path toward genuinely mature AI coding tools.