@ItsKieranDrew What would you use to recognize progress before audience numbers become meaningful? Better conversations and improved writing seem like useful early signals.
@emilkowalski How do you prevent a correction that was appropriate for one interface from becoming a rule that's applied too broadly? Does the skill preserve the context behind each decision? Does it survive across agent sessions?
@zqkrq A shared machine makes the comparison easier to reproduce, but the ranking might change on lower-spec hardware or different operating systems. It would be interesting to see which results remain stable across environments.
@raunofreiberg I wonder whether the bigger risk is accepting the first generated direction rather than seeing it at all. Comparing a deliberately unusual human concept with several generated alternatives could still broaden exploration.
@joshpuckett@withgraphical When someone edits a theme directly in the application, how do you keep the changes consistent across component states that aren't currently visible, like validation errors or loading?
@shouchan099 How do you decide which element deserves motion when several values change at once? Is the choice mostly about visual hierarchy or what the user needs to notice?
“Biggest lesson so far using AI agents as a junior frontend dev:
Give them a way to verify their work.
No tests / no browser check / no visual feedback = confident garbage.
The verification step is where I’m actually learning judgment, not just accepting the output.”
2026 dev setup:
3 coding agents.
12 MCP servers.
A beautiful terminal theme.
The button still doesn't work.
Maybe I should open the browser before installing another tool.
Your attention might be your greatest competitive advantage.
But every new trend, opportunity, or shiny idea competes for it.
I'm learning that focus isn't just about doing more.
It's about saying no, even to things that interest you.
What helps you stay focused?
@ScottShapiroUXD Does your verification process actually attempt a test request, or does it primarily inspect documentation? An API can be documented and still be unavailable to a particular account or subscription plan.
@emilkowalski How would you recommend testing a gesture when someone interrupts a drag halfway through? Making the animation look smooth seems easier than ensuring the component always ends in the correct state.
@mattpocockuk Understanding React seems useful even when an agent writes the code. Someone still has to recognize incorrect state transitions, unnecessary effects, and components that are difficult to maintain.
@emilkowalski There's an interesting distinction between reducing actual latency and reducing perceived latency. The first improves performance; the second changes how users experience the same wait.
@alperortac Generating alternatives seems especially useful for developing design judgment. Comparing different solutions forces you to articulate why one interaction works better than another.
@eagledev_@emilkowalski How did you approach translating the design into working interactions? I'm curious which details required the most iteration after the first implementation.
@mattpocockuk Visual checks seem especially important here. A component might pass its tests while still overflowing on mobile, trapping focus, or behaving incorrectly with a keyboard.
@ryanlanciaux AI-assisted writing and automated engagement aren't necessarily the same thing. I'd distinguish someone refining a genuine observation with AI from a bot replying to hundreds of posts without understanding them.