Blog GenAI This week’s AI Bite: Can AI build an application from scratch? Our front-end developer tests the capabilities of Opus 4.5This week’s AI Bite: Can AI build an application from scratch? Our front-end developer tests the capabilities of Opus 4.5
Weekly AI Bites is a series that gives you a direct look into our day-to-day AI work. Every post shares insights, experiments, and experiences straight from our team’s meetings and Slack, highlighting what models we’re testing, which challenges we’re tackling, and what’s really working in real products. If you want to know what’s buzzing in AI, check Boldare’s channels every Monday for the latest bite.
At Boldare, we regularly explore AI solutions that can improve product development. Recently, our front-end developer decided to test how far Opus 4.5 could go by combining code generation, reasoning, and direct browser interaction. What seemed like a simple experiment turned into an exploration of the ways AI could fundamentally change how we build and test digital products.
The game and repository are publicly available:\ Demo: crypto-game-opus-4-5.netlify.app\ Repository: github.com/jankepinski/crypto-game
Table of contents
First steps: Code generation and UI setup
From the start, Opus 4.5 generated code with surprising confidence. In just a few prompts, it created a working HTML structure, layout, and transformation logic. The result was minimalistic yet functional, ready for testing without unnecessary visual frills.
Challenges in game level configuration
The mini-game focused on text transformations and encoding (Base64, ROT13, Caesar Cipher, Atbash). These fine-grained operations posed a challenge for the LLM, revealing mistakes: some transformations were incorrect, the order of operations failed in some cases, and a few levels were initially impossible to complete. Importantly, this was intentional, the goal was to see how the model would identify and fix its own errors. Through iterative testing, Opus made mistakes, evaluated results, and corrected them, demonstrating self-directed debugging.
The model as user, tester, and developer
The experiment took a new dimension when Opus was asked to “click through” the application in the browser. The model ran through the levels, verified operations, returned to the code, and fixed errors. In effect, Opus created a full iterative feedback loop, acting simultaneously as a tester and a developer addressing regressions.
Multimodality in action: UI improvements
Another insight emerged: Opus handles front-end tasks more effectively when it can see the rendered UI. It suggested improvements in spacing, alignment, and readability based on how the interface actually appeared. This ability to analyze visual output allowed Opus to combine the roles of front-end developer and UI designer intuitively. By “seeing” its own work, it could deliver responsive design and replicate the intended interface accurately, a true game-changer for front-end development.
What this means for Opus 4.5
Even though the experiment began as a small mini-game, it revealed a critical insight: Opus 4.5 is not just a code generator. It can understand application behavior, run tests, identify errors, and fix them independently. This positions the model as an active participant in development, rather than a passive assistant.
Summary: Small project, big insights
The experiment highlights the potential of multimodal AI in digital product development. Models like Opus 4.5 can understand and interact with their environment, respond to errors iteratively, and collaborate with humans at every stage of development. This doesn’t replace developers, it augments their workflow, enabling faster, smarter, and more reliable product creation.
Share this article:
