Can AI Build a Retro macOS Web App in 10 Minutes? A Hands‑On Test of Manus 1.5

After more than six months of low‑profile development, Manus jumped from version 1.0 to 1.5, claiming a four‑times speed increase (average task time reduced from 15 minutes to under 4 minutes) and a 15% quality gain with 6% higher user satisfaction.

The headline feature of Manus 1.5 is its full‑stack web‑application capability: through conversational prompts it can generate and deploy production‑grade back‑ends, databases, authentication, custom domains, analytics, and even version control without leaving the platform. New collaboration tools and a Library also let teams share sessions, versions, and assets.

To test these claims, the author prompted Manus to create a retro macOS‑style web desktop. The prompt specified a classic Mac OS 9 look, three built‑in apps (a notepad, an AI‑powered painter, and a calculator), and an account system.

Prompt: "Help me build a web‑based retro Mac OS desktop." 1. Appearance: classic Mac OS 9 with top menu bar and draggable windows. 2. Built‑in software: notepad, AI painter, calculator. 3. Account system required.

Manus initialized a full‑stack project, automatically setting up the server, database, and authentication layer. It then generated a design document outlining the tech stack, database schema, UI details, and potential challenges.

During development, Manus installed dependencies, restarted services, and fixed type errors autonomously, completing the entire project in roughly ten minutes.

The resulting UI captured the iconic cyan desktop background, top menu bar, and three small app icons reminiscent of classic Mac OS 9, while also echoing some retro Windows aesthetics.

Functional testing of the three apps showed:

Notepad: creating, editing, and saving documents worked flawlessly.

Calculator: basic arithmetic operations performed correctly.

AI Painter: generated a cat on the moon using Manus’s built‑in image‑generation API, despite a typo in the prompt.

The account system functioned as expected: users could log out, log in with a new email (receiving a verification code), and their data remained isolated. All history persisted across sessions.

After passing functional tests, the project was published, generating a public link (https://retromacos-cnt7elud.Manus.space) for anyone to try.

Manus also promotes an “unlimited context” capability, achieved through reversible compression, structured summarization, file‑system offloading, and hierarchical action spaces, allowing agents to run long‑term without context degradation.

In an interview, founder Xiao Hong explained that Manus focuses on building a robust general‑purpose kernel first, letting various scenarios emerge naturally, positioning the startup as an engineering‑driven AI agent rather than a pure model provider.

Overall, while Manus 1.5 still has rough edges (e.g., fast token consumption), it demonstrates a significant step beyond toy‑level AI agents, delivering a near‑complete web app from a single natural‑language prompt.