- Published on
Kākāpō Party
- Authors
- Name
- DeepHorus
At the closing keynote of the WeAreDevelopers World Congress North America, a presenter showcased an unusual experiment: turning a conservation story about kākāpō parrots into a celebratory pixel-art animation. In an account dated September 26, 2026, the presenter said the “Kākāpō Party” project was created for the final slide of a STAR moment and was built around the species’ record-breaking 2026 breeding season. The demo connected a live conservation update with two emerging AI capabilities: multimodal content generation and agentic browser automation. It also provided a concrete example of how a speaker can assemble a polished multimedia asset during an event itself.
The conservation backdrop gave the technical demonstration a timely and emotional focus. Kākāpō are rare New Zealand parrots, and the presenter wanted the closing visual to celebrate their successful breeding rather than serve as a generic AI animation. For the creative stage, three kākāpō photographs collected through Google Image Search were supplied to Claude Opus 5.5 as visual references. The prompt asked for an HTML5 canvas animation featuring at least 20 pixel-art birds jumping, dancing, and celebrating with confetti. Grounding the design in real photographs helped establish the species’ recognizable appearance while still leaving the model freedom to invent a playful scene.
Claude Opus 5.5 generated the interactive page, which the presenter described as a particularly strong result for pixel-art animation. The HTML5 canvas implementation was then downloaded so it could be incorporated into a Keynote presentation. A local Claude Code session received the file location and instructions to record the page in a video lasting roughly 15 seconds, interact with it in a browser, and distribute clicks across the canvas. The second prompt also specified that clicking should not begin until three seconds had elapsed, preventing the confetti from appearing before the animation had settled into its opening sequence.
To carry out the request, Claude Code used Playwright to launch Chromium with a 1280-by-720 viewport and video recording enabled. The short Python script opened the local HTML file, calculated delays from a start timestamp, and performed ten scheduled mouse clicks. Those interactions moved from the center to the corners, edges, lower regions, and finale position between seconds three and 13.2, triggering different parts of the confetti effect. Playwright then closed the browser context after the capture window, producing a video that could be placed directly on the keynote’s closing slide.
The notable achievement was not any single generated image, but the end-to-end production chain that linked natural-language direction, generated code, a local browser, automated interaction, and screen recording. Claude created the animated experience, while Claude Code operated that experience as an automated user rather than merely proposing instructions for someone else to execute. This distinction illustrates a shift from AI as an assistant producing suggestions toward AI acting as a practical production pipeline. For conference speakers, such workflows could reduce the time required to create demos, slides, tutorials, and event visuals without requiring every tool in the sequence to be operated manually.
The presenter’s reaction was enthusiastically positive, calling the generated page “pretty great” and the resulting recording “exactly what I needed.” The short Playwright program was also described as “pleasingly short,” suggesting that the workflow became easier to inspect and reproduce as well as faster to run. Sharing the prompt transcript, resulting page, and automation script allowed readers to examine the process instead of taking the final animation on trust. The account did not report independent audience reaction or a formal evaluation, so the strongest documented response came from the presenter’s own successful use of the tools.
Looking ahead, the experiment points toward broader use of coding agents in live content production. As browser automation becomes more reliable, AI systems may be able to generate an artifact, test its behavior, record demonstrations, and package deliverables with limited supervision. Teams adopting this approach will still need clear safeguards around copyrighted source images, browser security, timing accuracy, and visual quality control. They will also need to decide how much of the process should remain transparent and reproducible. Even with those caveats, the Kākāpō Party demo shows how multimodal generation and agentic execution can combine into a compact but surprisingly complete creative workflow.
