FocusNow AI Club, meet-up #3 – 30 June 2026.
The third AI Club session was a hands-on account of building a production interface animation entirely by prompt – what worked, what it cost, and where the tooling still falls short.
The problem
A conversational agent needs to show users what it is doing: listening, ready to speak, or speaking. The design concept was a set of rings that shift between three states – drifting slowly while listening, arranging themselves when ready, resolving into waveforms while speaking.
Static images of each state were straightforward to generate. Animating them, and moving smoothly between states rather than cutting, was not. An attempt with a general coding assistant did not get there.
What worked
Claude Design did. Working from the static images and a long written prompt, it produced the three animated states, then – on a follow-up instruction, still with no code – the continuous transitions between them.
The output format is the part worth noting. Claude Design produces interactive HTML rather than video, so colours, ring sizes and timing all remain adjustable afterwards. A parallel attempt using video generation looked arguably better, glassier and more three-dimensional, but was effectively frozen: changing a single colour or viewpoint meant regenerating the whole clip and hoping. For anything still being iterated, an editable artefact beats a polished one.
The process is visibly iterative. The tool builds a version, evaluates it against the brief, notices where the motion does not match the description, and reruns. That costs tokens – roughly $50 across three days, with pauses waiting for limits to reset. The free tier was described as fine for playing and not much else.
The result is live in production, wired in by a full-stack developer.
The economics
The alternative was a freelance designer: several hundred euros, plus the time to explain the concept well enough to brief someone. At $50 and three days of intermittent attention, the comparison is not close for work of this size.
The same tool takes a company style guide as input and applies it across wireframes, clickable prototypes and decks – which is where it stops being a one-off and starts affecting how quickly client-facing material can be produced.
Key takeaways
- Interactive output beats video whenever the design still needs iteration.
- The paid tier is the real product. Free is a demo; serious output consumes tokens.
- Budget for the loop, not the render. Cost comes from the tool critiquing and redoing its own work.
- Briefing time is part of the comparison. Prompting can be faster than explaining the same idea to a person.
Join the next one
The FocusNow AI Club meets monthly, with working demonstrations rather than slideware. If you are exploring where AI fits into your own delivery work, get in touch.