PocketCoder-A1: how to make Claude work in three shifts

An article was published on Habr about PocketCoder-A1 — a homemade "Auto-Coder" that the author built to make the AI model Claude work according to a multi-shift schedule and extract maximum value from the existing LLM subscription.

The Idea: Let AI Work at Night

The starting point of the article is the author's observation that the emergence of powerful AI models in practice does not so much replace human labor as it increases their workload: instead of working less, specialists continue to work just as much, simply now in parallel with AI tools. The author considers a logical response to this as redistributing the load over time itself: if a person is busy during the day, let the model continue working at night, when the subscription access is not being used anyway.

Behind this idea lies an observation familiar to many who use subscription access to powerful LLMs: the actual load on the limit throughout the day is extremely uneven — peaks occur during working hours, while at night and on weekends access is essentially idle. From an economic perspective, this is a missed opportunity: the user has already paid for the limit but does not fully use it, while tasks that can be entrusted to the model without constant human supervision — refactoring, test generation, issue backlog processing — typically accumulate in abundance in any active codebase.

• Tool: PocketCoder-A1 ("Auto-Coder")
• Model: Claude
• Idea: "three shifts" — round-the-clock AI work instead of single use during work hours
• Motivation: extract maximum value from an LLM subscription, not from individual requests
• Source: Habr

How Does the Economics of LLM Subscriptions Work?

The subscription model of access to LLMs, unlike per-token payment through API, creates an economic incentive to use the allocated limit as fully as possible: the user pays a fixed amount regardless of how much the model actually "worked" during that time. Tools like Claude Code, operating on top of such a subscription, allow you to run the model to complete tasks without constant human participation — that is, theoretically you can organize a queue of tasks that the model will process in the background, including at night, when the person is not using the limit themselves.

This approach differs from classical API usage with per-token billing, where each additional request directly increases the bill: there the user, on the contrary, has an incentive to minimize the number of requests to the model. A subscription reverses this logic — and this is precisely why tools for automating background, round-the-clock model launches become particularly attractive to those who already pay for a subscription with a fixed limit, rather than being billed based on actual usage.

What Does This Mean for Developers?

The author describes his tool through the metaphor of "three shifts" — a mode in which the model works not episodically, by request, but practically continuously, like an employee on a shift schedule. For a solo developer or small team, this is a way to increase actual productivity without increasing subscription costs: if the limit is already paid for, idle hours become wasted computational power that can be occupied by background tasks — refactoring, test generation, backlog processing.

Similar homemade "auto-coders" reflect a broader shift in how developers use commercial LLM subscriptions: not as a one-time chat assistant, but as a constantly running background process embedded in the development cycle. How sustainable such an approach is in terms of the terms of use of specific subscriptions and restrictions on automated requests — a question that in such homemade solutions is usually left to the discretion and risk of the author themselves.

The author's own formulation — "AI does not replace people, people just work more" — captures an observation that is increasingly heard in discussions about the effect of AI implementation in development: the expectation that automation will free up specialists' time often turns out in practice to mean that they are expected to do a larger volume of work in the same amount of time, just now with the help of AI tools. PocketCoder-A1 in this sense is not so much a technological solution as a personal strategy of one developer to compensate for this growing workload by shifting at least part of the work to the model during the hours when he himself cannot physically work.