🔍 Read the full analysis: How AI Agents Helped Deliver Gewerkton’s New Look In Two Days on ThorstenMeyerAI.com

Prime Big Deal Days · Oct 6–7Offer from Amazon

Get monitors, keyboards and dev gear delivered free — and shop member deals

• Fast, free delivery on millions of items
• Access to Prime Big Deal Days deals on October 6–7
• Prime Video, Amazon Music and more included

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

Construction-site documentation tool Gewerkton says, in the original analysis, it replaced the look of its web and phone apps, Studio, 26 icons and 40-language website with a design called HORIZON during Oct. 2-3, 2026, using AI coding agents to implement an approved plan. Founder Thorsten Meyer, who built the product in Iffeldorf, Bavaria, says he approved the plan, inspected results and authorized releases; the account is published by Gewerkton’s publisher, which discloses that it built the product.

The scope included moving 281 hard-coded web-app colour values into design tokens, restyling 47 cards and releasing phone-app version 0.5.0 for iPhone and Android. Gewerkton says Studio received light and dark versions of the new look, and all 26 icons were rendered in Blender in cobalt and navy. The website’s 529 pages were updated across 40 languages, with 1,583 files uploaded without an error, according to the company.

The company also reports 228 AI film runs for five regions, 222 of them with regional versions, and eight help videos. The two-day description refers to execution from an approved plan through the live website; an interim design draft from Sept. 26 was already available. Gewerkton says the work began with a written plan listing open decisions, options and recommendations. Agents worked on a separate code branch, and Meyer released each part after reviewing it.

Some work required human approval or intervention. Films showing the redesigned products were held until Meyer had reviewed the interface on Oct. 3, and paid film reruns required his explicit approval. Apple’s developer agreement also had to be accepted by Meyer before an iPhone build could proceed. For accessibility, the original cobalt colour measured 4.38:1 against a light background, below Gewerkton’s stated 4.5:1 working threshold under WCAG 2.1 AA; the team darkened it to reach 4.80:1.

Human Reviews Shaped the Release

The account offers a specific example of AI agents carrying out a coordinated software and design implementation across several products, rather than a single coding task. Gewerkton says the agents worked from written plans and completed changes across web, mobile, Studio, graphics and website content within two days. That makes the case relevant to companies evaluating whether agent-assisted work can shorten delivery timelines across a product suite.

But the company’s own description also shows the limits of treating automated completion as proof of readiness. A visual review caught icons whose white tiles obscured nearby labels in dark mode, while colour-masking tools missed red details in film frames. Automated tests did not flag the icon issue. The release depended on human approval, visual inspection and corrective work as well as agent execution; the account does not establish that the same timeline or results would apply to other teams or projects.

The rollout included measurable checks, but those figures are Gewerkton’s project records, not an independent audit. The distinction matters because test counts, visual reviews and successful uploads show what the company says it checked, while they do not by themselves establish long-term reliability, usability or accessibility across every real-world use.

A Two-Day Execution Window

Gewerkton is described by its publisher as a construction-site documentation tool built in Iffeldorf, Bavaria, by Thorsten Meyer. The publisher’s disclosure says it builds Gewerkton itself and writes about what it learns. That relationship is material: the detailed figures and assessments in this account come from the company responsible for the product, rather than an outside evaluator.

The work was directed through coding agents identified as Codex and Claude. Gewerkton says the agents implemented written plans on a separate copy of the code and their own branch. Meyer approved the plan as recommended, reviewed the resulting work and released each part. The design itself was not created from scratch during the two-day window: an interim draft from Sept. 26 was carried forward.

Testing covered different screen sizes, languages and display modes. Gewerkton reports inspecting the web app in 200 views across English, German, Japanese and Hindi, at widths of 390, 834 and 1,440 pixels, in light and dark modes. It says the phone app passed 853 of 853 tests and Studio passed 2,158 of 2,158. The website passed seven test suites and was checked in five regions at three widths, including after upload. The company also disclosed eight failing web-app tests in seven files: six files reportedly failed the same way on the main branch, and one failure depended on test order.

“Gewerkton is built by our publisher — we build it ourselves and write down what we learn.”

— Gewerkton’s publisher, in its disclosure

What the Project Records Do Not Show

The figures and outcomes are reported by Gewerkton; the supplied account does not include an independent assessment of the code, tests, accessibility or finished design. It also does not provide a comparison with the time, staffing or cost required for a non-agent-led redesign, so the project cannot establish how much the agents reduced effort.

The account gives limited cost information. It says project records put the actual cost of the films at about $134, but the supplied material ends before explaining what that figure includes. It is unclear whether it covers only generation charges or also staff time, review, failed jobs and other production costs. The source says 76 failed film jobs were not billed, but does not provide a full project budget.

Other open questions include how the redesign performs with users over time and whether the scheduled mobile fixes have been released. Gewerkton says five major findings from an Android test were not blockers and were planned for version 0.5.1; the supplied account does not confirm that release or report user feedback after launch.

Mobile Fixes and Film Rework

Gewerkton said five findings from its Android test, none classed as a blocker, were scheduled for phone-app version 0.5.1. One involved camera-screen buttons wrapping mid-word on a Pixel 10a after the redesign increased text size and padding; the German word for “cancel” wrapped onto two lines. The company did not provide a release date in the supplied material.

The film work also continued after initial production problems. Gewerkton says nine regional runs with leftover red were later reshot with Meyer’s approval and passed visual review, after 10 regional versions were temporarily switched to the European cut. The account says each reworked film was viewed again at 10%, 50% and 90% of its length. Further costs, performance results and updates to the mobile app remain unreported in the source material.

Key Questions

What did Gewerkton change in two days?

Gewerkton says it implemented its HORIZON redesign across the web and phone apps, Studio, 26 icons and a website with 529 pages in 40 languages on Oct. 2-3, 2026. The website also received 228 AI film runs for five regions and eight help videos.

Did the design start from scratch on Oct. 2?

No. Gewerkton says an interim design draft from Sept. 26 was carried into the work. The two days refer to execution from an approved plan through the live website, not the entire design process.

What did the AI agents do, and what did Meyer approve?

Gewerkton says agents identified as Codex and Claude carried out implementation from written plans on a separate code branch. Thorsten Meyer approved the plan, reviewed results and released each part; he also approved paid film reruns and the product-screen films.

What problems did the team find?

A human visual check caught icon tiles that obscured labels in dark mode, and review found film recolouring errors that automatic masking missed. The company also reported eight web-app test failures, which it said were not caused by the redesign, and five non-blocking findings in an Android test.

How much did the film work cost?

The supplied account says project records show an actual film cost of about $134, and that 76 failed jobs were not billed. It does not explain which expenses the $134 includes or provide a full project budget.

Source: ThorstenMeyerAI.com