What An AI Coding Agent Can Teach You Through A Five-Day Build

🔍 Read the full analysis: What An AI Coding Agent Can Teach You Through A Five-Day Build on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on office and shipping supplies

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

Developer Thorsten Meyer says he built and released a free web app for people counting down to partial retirement in Germany over five days, working with the coding agent Claude Opus 5.5. His account, published October 13, describes a project that grew from a one-minute animated film into a site with a workday countdown, narrated film and 3D games — while Meyer retained responsibility for the concept, design choices and release approval.

Meyer began the project on Monday, September 28, 2026, asking the agent to make a short film about the working days remaining in an Altersteilzeit arrangement. The term refers to a German partial-retirement arrangement; in the block model described by Meyer, a person completes an active work phase followed by a passive phase away from work. A 4:4 model, for example, means four years in each phase.

By Friday, October 2, the resulting app included a daily countdown, milestones, a narrated film, games and tools such as a bridge-day finder. Meyer says users enter a date and select a model from 1:1 to 5:5; the app then calculates working days, accounting for weekdays, state public holidays and entered leave. The project also included a team countdown, badges, calendar functions and legal pages.

Meyer reports that the build involved 24 commits co-authored by Claude Opus 5.5. He says the agent handled implementation, testing, translations, voice clips, sound effects and deployment. The site is described as having no accounts, cookies or tracking, with calculations performed in the browser. These details come from the developer’s own account; the article does not report independent testing of the app or the agent’s work.

Where the Agent’s Work Ended

The account offers a practical example of how a coding agent can contribute across more than writing code. Meyer describes using it for feature development, testing, localization and production tasks, with work divided among parallel agents for larger translation jobs. That range helps explain why a small initial request could expand into a multilingual public website in a short period.

But the build also shows why speed alone does not describe the work. Meyer says he selected the project’s tone and visual idea, chose voices and dialects, made decisions about narration and contact details, and approved the release. Those are reported decisions from this project, not evidence that all coding-agent workflows require the same division of responsibility. Readers evaluating similar tools should distinguish the agent’s reported output from the human choices and checks that shaped it.

The app’s subject also gives the project a defined practical use: helping people track a long work-to-retirement timeline. Its calendar calculations and public-holiday handling could affect the countdown users see, making testing and accurate configuration relevant beyond the novelty of the games. Meyer says the app accounts for state holidays and leave, but provides no independent evaluation of calculation accuracy.

From Film to Countdown App

The project began with a narrow creative brief: a short animated countdown for a person in the active phase of an Altersteilzeit block model. Meyer says he wanted the countdown to make a long period of work feel more engaging. The film’s imagery contrasts an office setting in the active phase with the sea, sky and sunrise in the later phase.

According to Meyer, the first day brought the film, date calculations and two game prototypes; later days added the daily view, more games, language support, narration and sound. He describes sixteen versions across five days, as well as a fix for an iPhone scrolling problem. His account says the app supports seven languages and five German dialects, and refers elsewhere to twelve languages for narrator production and translated text. Those figures appear to describe different parts of the project, but their exact relationship is not fully explained.

The game design connects calendar time to gameplay: Meyer says months, weeks, working days and public holidays determine elements such as levels, rows or obstacles. He also says the names were chosen to avoid other companies’ trademarks. This is relevant to how the app translates its countdown data into play, rather than a separate claim about the agent’s capabilities.

“The interesting part is not how much an AI agent can build, but what was still left for me to do.”

— Thorsten Meyer, describing the build

What the Account Does Not Verify

The development timeline, feature list and agent contributions are reported by Meyer, the project’s creator. The source material does not include an independent review, repository audit, test results or confirmation from Anthropic about the agent’s role. It is not clear how much time Meyer spent prompting, checking outputs or correcting defects during the five days.

Some production figures also need qualification. Meyer cites twelve languages for voice production and text, but says the finished app has seven languages plus five German dialects. He also refers to nineteen 3D games in the overview, while the day-by-day account describes fifteen game modes and a changing number of games. The source does not explain whether these counts refer to different categories or stages of development.

The account says the app calculates working days using state holidays and entered leave, but it does not identify the holiday data source, explain how edge cases are handled or provide an accuracy assessment. Its privacy claims and the app’s current availability likewise have not been independently checked here.

Testing the Published App

Meyer’s account ends with his approval of the build on Friday, October 2, followed by additions including five German dialects. The source does not name a future release, product milestone or independent evaluation. The next useful step for readers interested in the project is to check the published app and its documentation for current features, calculation rules and privacy details.

Further reporting could compare the app’s countdown results against official holiday calendars, examine its code and test behavior across supported languages and devices. Until such checks are available, the five-day timeline and production figures should be understood as the developer’s account of a specific collaboration, not a general benchmark for how quickly an AI coding agent can build and release software.

Key Questions

What did the AI coding agent help build?

According to Thorsten Meyer, Claude Opus 5.5 helped build a free web app for tracking working days in a German Altersteilzeit arrangement. Features included a countdown, a narrated film, games and calendar tools.

How long did the project take?

Meyer says work began on September 28, 2026, and the app was approved and released by October 2, a five-day period.

Did the agent make all the project decisions?

No. Meyer says the agent implemented and tested features, while he set the concept and tone, chose voices and dialects, made other design decisions and approved releases.

Has the app’s accuracy or privacy been independently verified?

The source account provides Meyer’s claims that calculations happen in the browser and that the site has no accounts, cookies or tracking. It does not include an independent privacy audit or a separate assessment of the workday calculations.

Why do the language and game counts differ in the account?

Meyer describes seven languages plus five German dialects in one section, and twelve languages for voice production and text elsewhere. The account also uses different game and mode counts. It does not fully explain how those figures relate.

Source: ThorstenMeyerAI.com