The Plan for 2026

Disclaimer

This post contains forward-looking, aspirational, statements. While these forward-looking statements represent our current plans, they are subject to risks and uncertainties that could cause actual results to differ materially. You are cautioned not to place undue reliance on these forward-looking statements, which reflect our opinions only as of the date of this post.

Of my plans for 2025, I failed many and abandoned many. Surely, 2026 will be no different.

Intro

At the beginning of 2025 I made plans. And I did things I had planned on doing, and failed to do things I had planned on doing, and did things I had not planned on doing.

And the rest of the things which I did or did not do in 2025, how I trained LLMs with trinary weights, and how I melted them into logic DAGs, and made complex cellular automata run very fast, behold, are they not written in the 2025 retrospective.

Last year I became very good at moving bits to and fro in the registers. In this year, I wish to stick atoms together and make things which go to and fro on the face of the earth.

We are having a Cambrian explosion of weird VTOL aircraft.

I feel that the existing players are not doing a good enough job of things. <sarcasm>Surely it can’t be all that hard to out compete the big defence contractors and the dozens of new aviation startups.</sarcasm> Someone has to do it and not enough people are trying!

We will probably not build the next Antonov Mriya, but let’s see if we can win the base tier of the DARPA lift challenge, lift 110lb of payload with a <55lb VTOL craft, and carry it 4 miles.

Airplanes are for making money.

Airplanes are tools for war.

But airplanes are also beautiful dreams.

To make airplanes, we need good materials. We also need to transfer the forces correctly. Any part of our frame which does not break when stressed slightly beyond its design tolerances needs to be made thinner.

Alternatively, perhaps we should build a kayak. If we had a very lightweight kayak, we could carry it up to high alpine lakes and paddle about. That would be truly wonderful!

The Yuri is a unit of tensile strength to weight. (We do not much care about strength to volume ratio, just strength to weight. (We do however care about compressive strength to weight as well as tensile. Let’s ignore that for the time being. We can add compressive members later as needed.))

At ~2.5 mega Yuri, Aramid is not good enough for a space elevator, but is still good stuff.

Also, unlike carbon fiber, aramid does not produce evil lung cancer dust when cut or ground. (Although one does not simply grind aramid.)

Alternatively, UHMWPE fiber is 3 or perhaps even 4 mega Yuri. Perhaps we should use it instead.

Aramid degrades under UV. UHMWPE melts and creeps under heat. Aramid absorbs moisture. UHMWPE is hard to bond with epoxy. What fiber should we use?

How shall we arrange the fiber? How shall we bind it?

We could lay down fiber cloth and vacuum bag, but this limits our ability to do optimal fiber arrangement, and requires skilled human labor. Humans are slow and expensive. Humans make mistakes. Humans get sick when they have too much contact with evil chemicals.

I am a factorio player. (Specifically, I am not a factorio player because it is dangerously addictive for me.) I believe that one does not truly understand how to make a thing unless one can build a machine to make it, parametrically, lights out. I also hate getting epoxy in my skin. We need to do as close to full automation as possible. We also need to have good control over fiber angle. Vacuum bagged cloth does not grant us this. Therefore, we need continuous fiber winding.

Some say that the path of royalty is a continuous cutting motion. This does not apply to aramid composite. Aramid hates being cut. It fuzzes and rips out and jams ones tools. Cuts are weakness. True strength is a continuous fiber path with no cuts.

Therefore, the Plan for 2026:

A

Pump and accurately dispense, two, differently viscous, corrosive, fluids, in the correct 3:1 ratio. West systems 207 epoxy hardener dissolves silicone, it dissolves Tygon Pharmed. What else can we use in a peristaltic pump? Are there truly no materials both flexible enough and capable of surviving the hardener? We must use glass, 316 stainless steel, or fluoropolymer for everything. Fluoropolymers are fairly stiff. We must build a pump anyway, with no flexible parts. It must be fully sealed, both because I hate leaks, and because any air getting in will form carbamate deposits.

B

Mix the two sides of the epoxy. They must be well mixed or the epoxy will not cure well. The mixing system must be self cleaning, it must run for weeks on end without human intervention.

C

Impregnate the epoxy into the aramid tow. It must be well integrated into the fibers. If the fibers are dry, it will be weak.

D

Wrap the wet fiber onto the mandril. The mandril is complex. It has bolt holes. Each bolt hole must have fiber properly wrapped around it. We need good tension control, we must wrap it tightly, to squeeze out any excess epoxy, but if we pull too tightly, we will crush our mandril or snap our tow. We need to rotate the mandril along many axes to wrap the shape correctly. This is a complex bit of CNC control.

E

Figure out how to make a mandril. It needs to be light weight while also strong enough to withstand the forces of winding. It will be wrapped by thousands of strands of tow under tension. This will impose quite significant compressive forces. Any normal structure will tend to implode. Do we 3D print it and somehow remove it afterwards? Do we make it out of carbon fiber rod and leave the rod inside for compressive strength?

F

Fiber wind a 5 inch quadcopter frame. Quadcopters are not the most urgent use case for fancy fiber winding, but they are small and complex and if I can wrap such a complex shape, I can likely wrap other less complex shapes. It will have 4 arms, each of which has 4 bolt holes. It will be a good test case.

G

Once have the fiber composite quadcopter frame, we test it. Drop the frame. Crash the frame. Kick it with steel toed boots. Step on it. Does it break? How does it break? What are the failure modes of epoxy aramid composite? What of UHMWPE? If the aramid or UHMWPE snaps, add more winding, if it buckles, perhaps add a few carbon fiber rods for compressive strength? (Aramid is quite poor under compression.)

H

Build a larger winding machine capable of handling 2-3 meter large objects. In theory this will just be a scale up of D, but scale brings new problems.

I

Wind a kayak. It needs to be waterproof, it needs to displace more water than a human weighs, and it needs to not buckle when a human sits in the seat.

J

Paddle along the river and not sink.

K

Build a kayak which weights under 20lb. Making something strong is easy. Making something strong and light is hard. I must carefully structure the forces. I must carefully design the carbon fiber internal frame, and the fiber tension skin.

Stress test the kayak. Jump on it, drag it down the bumpy street. How does it fail? How does it not fail? The parts which are still intact after stressing it, use less material there. Understand how aramid on a carbon fiber frame fails.

L

Carry the kayak up a mountain and paddle about in a high alpine lake. Feel superior to the people who don’t have such a light weight kayak.

M

Use the big winding machine to make a VTOL flying wing. It should be basically the same principle as a kayak, just a somewhat different shape.

N

Can I achieve a lift ratio of 2:1? Vertically lift a payload twice as heavy as the unladen weight of the craft? And travel 4 miles with it? What morphology do I use? I want a dumb stupid simple design. I want as few joints or moving parts as I can get away with. I don’t need speed, I just need good lift with little power. Joints are weakness and/or weight.

O

Scale it. How big can I make it? Can I scale up to carrying a 110LB payload? While weighing under 55lb?

P

Win the DARPA Lift challenge. For the money, and for the glory. (This is a stretch goal. It is unlikely that I will actually achieve it this year or ever.)

Q

Run cellular automata on GPU.

R

Draw hair on a GPU, at 4K 60fps.

S

Render sound on a GPU. Render 65536 sine waves and make beautiful sounds.

T

Build a game with GPU CA, vector drawn hair, and GPU render sound, and run it on a steamdeck at 4K 60FPS.

U

Learn to write Rust for the RPi Nano.

V

Finish transitioning all my devices to NixOS.