A developer has spent months letting AI write their code, and the habit has turned them into a reviewer of work they barely understand. The writer, who builds the LibreWeddingPlanner project, describes a slow spiral that began with small requests and ended with months of no original coding at all.
The story, titled “One Month Without AI,” is a warning about what happens when developers lean too far on code-writing tools.
The Spiral Begins
The writer’s friend urged them to try AI coding agents. “You feel so powerful,” the friend said. Once started, the writer found themselves asking AI to implement functions, write tests, and handle entire Jira tickets. The tasks that once took days were done in hours.
The writer was a TDD practitioner, used to writing tests before code. But the AI changed that habit. The writer said they no longer wrote the code themselves. Half of the purpose of TDD — not biasing tests by how the code was implemented — was gone.
Control slipped early. The writer saw the agent signing commits alongside them and rushed to opt out. They wanted to pretend the code was their own.
Losing Control
The writer pasted full Jira descriptions into the AI and let it implement the work. On a codebase they barely knew, they trusted the AI’s judgment over their own. They admitted they did not know what the code they pushed to production actually did. “Fucking scary,” they wrote.
Small tasks became PRs in 30 minutes. The writer set up multiple agents across worktrees, connected them to Jira via ACLI, and asked them to work on tickets in separate worktrees. The writer felt ecstatic.
But the joy did not last. Larger tasks required decomposition, and the AI produced PRs with changes the writer could not explain. Code reviews became a problem. Peers who knew the codebase better might ask why something was done, and the writer could not answer. “How was I going to admit it was all AI, and I was nothing but an AI shepherd?” they asked.
The writer forced themselves to understand the code to avoid humiliation. That effort did not stop the spiral.
Exhaustion Sets In
With too many agents running, the writer found themselves exhausted. Small tasks that should take 20 minutes took 5 minutes of AI work and two days of review. The writer switched between code, tests, style checks, and CI status, and the constant context switching wore them down.
No single task went in without change. The AI wrote commits and PR descriptions, often seven paragraphs long. The writer stopped reading them. AI sounded professional enough that the writer relaxed and let it handle the details.
The writer had not written a single line of code themselves in months. They had not committed anything themselves either.
What We Make of It
The writer’s core argument is simple: AI productivity is an illusion. The exhaustion of managing multiple agents, reviewing code you do not fully grasp, and answering for decisions you did not make is real.
Most AI developers will deny they have lost control, the writer suggests. But deep down, they know it is true.
The story is a cautionary tale about dependency. It is easy to let a tool do the work and harder to take it back.
| Stage | What Happened |
|---|---|
| Start | Friend urges AI use; writer tries enhanced autocomplete in VSCode |
| Middle | AI handles Jira tickets, PRs in 30 minutes, multiple agents |
| End | Months without original coding; exhaustion sets in |
The writer admits the addiction comparison is apt. “Not without a degree of shame,” they wrote about using AI at work. The shame comes from admitting the tool took over a skill they once owned.
The story ends with a confession of dependence. Whether the writer can break it is not stated.
Source material: “One Month Without AI,” bustikiller.com.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.