Three instruments, a few minutes

Three interactive instrument figures, a grand action, a Rhodes and a Hammond, made in minutes with an AI agent by someone who knows the instruments and never learned to draw them. On what is scarce now that the making is no longer the hard part, and what that does to learning.

I played jazz piano for a living once, and I have owned and worked on Rhodes pianos since. I know these actions by feel, the way you know a bicycle you have fixed yourself.

Yesterday I typed a few sentences to an AI agent and got the three working models on this page back: a grand action that escapes under your finger, a Rhodes that rings when the hammer hits the tine, a Hammond whose drawbars change the wave on a small scope. Minutes of my time, no manual work, and a lot of fun.

The grand piano hammer escapes

A grand piano key is a lever. Press it and the wippen rises, and the jack standing on it pushes the hammer up toward the string. About two millimetres before the hammer arrives, the jack's toe meets the let-off button and tips out from under the hammer, and the hammer flies the rest of the way on its own. On the way back the backcheck catches it, and the repetition lever holds it up while the jack slips back under, so the next note can start before the key is all the way up.

Play the grand action and see it working.

An idea and the making of it

For most of my life, an idea and the ability to make it were two different things. You could know exactly how a piano action works and still have no way to show it to anyone, unless you learned to draw or to program, or paid someone who had. Explanations like these stayed in the heads of the people who could have given them.

That gap has closed. I know because I crossed it in a few minutes, by typing what I wanted to see. An agent can draw, program, measure, test and publish, faster than I can read what it did. What it cannot do is know what a Rhodes feels like under the finger. Its first version had the hammer resting against the tine, the steel rod that rings, after the strike. A real hammer settles back a few millimetres, so the tine has room to swing. I knew that from working on these pianos, and one sentence fixed it. The rest of the figure, down to the real measurements, the agent found and built on its own.

The Rhodes hammer strikes a tine

A Rhodes has no strings. Each key throws a small hammer with a neoprene tip against a tine, a stiff steel rod that forms a tuning fork together with the tone bar above it, and a pickup two millimetres from the tip turns the swing into the signal that goes to the amplifier. The curved underside of the hammer rides on the key's pedestal, and with the key held it settles back a few millimetres behind the crest, which leaves the tine free to swing. Let the key go and the damper's wool felt comes back up against the tine and stops it.

Play the Rhodes and see it working.

The knowledge in your hands

What is scarce now is the knowledge that sits in your hands: what a correct one looks like, and what a wrong one looks like. The agent built the whole organ from its table of gears, from the motor to the drawbars, and every frequency on the page comes from that table. It could not see that its tonewheels had come out egg-shaped. Anyone who has seen a picture of one sees that in a second.

The Hammond spins a toothed wheel for every pitch

A Hammond makes its tones with spinning steel. A motor turns ninety-one wheels with teeth around the rim, each in front of a magnet with a coil, and every tooth that passes changes the field in the coil and makes one cycle of a near-sine wave: a wheel with sixteen teeth, turning a little more than sixteen times a second, gives middle C. The wheels turn all the time. A key connects nine of them to the output, from an octave below the note to three octaves above it, and the nine drawbars set how much of each goes into the mix.

Play the organ and see it working.

So if you are the person who knows a thing, a trade, a craft, a sport, a machine, a body, a kitchen, you can now make the interactive explanation you always wished you had. You still have to say what you want to see, and you still have to look at the result yourself, because an agent is confident when it is wrong, and it argues well. Your knowledge is what makes the model true.

What this does to learning

Most explaining is still describing. A teacher puts a mechanism into words, a book adds a drawing that does not move, and the learner has to build the moving picture in their own head, or never does. The figures on this page are the other way round. The moving picture exists, with the real numbers inside it, and the words only say where to look. That is what I think learning is going to look like: whoever knows a thing spins up the animation or the short video of it in an instant, instead of describing it, and the learner plays with it until it makes sense.

This is early. These models have rough edges, the agent argues when it is wrong, and a reader still has to check what they see against what they know. It will get better in no time.

What you could make

Think of the best teacher you ever had, the one who could explain a thing with their hands and never had a way to put that explanation in front of more than the people in the room. That teacher could make one of these now: an explanation people press, play and watch, with the real numbers inside it.

The three figures were drawn with anatomy, Ryan's open-source skill that teaches an agent to draw mechanisms as isometric figures. Ryan is @wheresryan22 on X.