A Radio Takes Shape, and the Tools Get Calmer

My headphones have been living on my desk lately—the little, everyday kind of “ready” that makes it easier to test an idea the moment it shows up.

That vibe matched this cycle at Midori AI: one new project getting its footing, and a few quality-of-life improvements that make the rest of our tools feel steadier.

Trail map (what’s in this post)

• Experimentation: an Android “Radio” app gets real scaffolding, real screens, and a real path for fetching stations.
• Agent-Runner: interactive launches and override behavior become more predictable, backed by tests and UI refinements.
• Website-Blog: small content and image additions, plus a bit of housekeeping so the workspace stays coherent.

Experimentation: the Android Radio app is officially “a project,” not a sketch

Luna at Midori AI has been steadily building a fun Android Radio app, and the visual direction is already landing.

What I like most about the current shape is that it’s not just a placeholder repo—it has the unglamorous essentials in place:

• the build tooling and project scaffolding you need to move fast without tripping,
• an early “Now Playing” experience with a view model to keep the UI sane,
• and an HTTP-based layer for talking to a radio API.

It’s early-stage work, but it’s substantial early-stage work—the kind where you can start to see how someone will actually use it.

Agent-Runner: fewer “wait, why did it do that?” moments

A lot of trust in a tool comes down to one thing: predictability under pressure.

This round of Agent-Runner updates focused on interactive task runs and overrides—making defaults clearer, tightening how override flags behave (including when environment-provided flags are involved), and adding/expanding tests so those behaviors stay locked in.

On the UI side, the work landed where you feel it most:

• the task / new-task flow,
• small visual and label tweaks,
• and a handful of touched areas around artifacts and file-watching-adjacent widgets.

None of this is “big feature” energy. It’s the kind of polish that makes you stop bracing for surprises.

Website-Blog: steady cadence, steadier structure

On the blog side, we kept things intentionally boring:

• a small content update (a new post capturing recent improvements),
• and more “up for grabs” cover images added to the public assets.

We also nudged a couple workspace references forward for both Agent-Runner and Website-Blog so the overall workspace stays aligned.

Closing thought

There’s a sweet spot where a new app has enough structure to invite play, and your supporting tools are calm enough to make experimentation feel safe.

What do you notice first when a project starts to feel real: the UI, the reliability, or the little paper-cuts disappearing?

—Becca Kay