Physical AI Layers 5 & 6 (Software / Planning & Control System) + My Latest Investment

Over the past few months we have been working our way through the different Physical AI investments layers from the ground up.

I took a little break from writing about the layers and have been focusing on finding the best investments and building out the portfolio.

But now it’s time to dive back in.

So at this point our machine has been trained, it’s practiced, it can see and sense the world, it has enough local compute to understand what’s happening.

And now we need to turn that understanding into action.

Which brings us to the next two layers!

Layer 5 - Software & Planning

Layer 6 - Control Systems

I am covering these two layers together for a few reasons which we will get to. But they naturally flow into each other. Layer 5 decides what the machine should do and Layer 6 makes sure it does it correctly.

In this report I am going to break down both layers in a (hopefully) digestible way, explain where I think the durable opportunities sit, introduce a handful of public ways to have exposure here, and then lay out the one stock that I purchased today.

Let’s dig in!

I spent a lot of time talking about, and investing in, companies that help these machines understand their surroundings.

And it can be overwhelming understanding just how much input these machines take in that they have to constantly synthesize and organize.

But what is also significant is just how many decisions the robot has to make based on all of these inputs.

And I think this is where we can really start to appreciate how amazingly built humans are. The tasks that we do without having to think about them may seem trivial when we just go about our day. But when you think about a robot having to do the same thing…not so much.

For example, if you were at the grocery store and wanted to go get some cereal, you probably wouldn’t have to do too much thinking to achieve the task. Sure, you might need to have an idea of what kind of cereal you want and what aisle it’s in. But for the most part, it’s quite an easy task that doesn’t require too much planning.

But if a robot needed to do it, well, it would need to know:

• All the same things, like what the item is and where it is

• What happens if a human or another robot is blocking the fastest route? What are the alternate routes? What if those are blocked too?

• What is the best angle to approach the cereal box?

• What if it’s out of reach?

• What if the cereal is missing?

• What if the robot fails to grab it on the first attempt?

You get the idea.

A human can just work through most of this because we have spent our entire lives learning how to be a human and operate in the physical world.

A robot needs specific software and planning to make sure it can complete even rudimentary tasks like these.

The way I think about it, planning happens at different levels.

At the highest level, the machine needs to decide what steps are required to complete a task. Then it needs to figure out how to physically move through the environment to execute those steps.

This is generally separated into ‘task planning’ and ‘motion planning’.

So “go get the cereal” might turn into something like: Find the cereal aisle. Navigate to it. Find the correct box. Position the robot. Reach for the box. Grab it. Then continue to the next part of the task.

But the plan also can’t be static. The real world changes constantly, so we also need to plan around what happens when something goes wrong.

Because good planning software not only needs Plan A, it also needs Plan B, C, D, E etc. All the edge cases.

Language starts to become very relevant in this layer as well.

For example, if I tell a robot, “Pick up the box of Cheerios and put it next to the Frosted Flakes,” it needs to connect language to perception, planning, and eventually physical action.

It needs to understand what I said, identify the correct objects, figure out where they are, decide what steps it needs to take, and then execute the plan.

So to sum it all up, this layer is really where understanding starts turning into planning for action.

This software is necessary. No doubt.

But necessary to me does not always mean investible.

And this is something I have gone back and forth on for this portfolio. And this is really where individual readers need to form their own opinion.

Let’s backpedal for a sec. When I first started this portfolio a few months ago I initiated with a handful of software picks that I thought were well positioned in this space and were durable. But at that time, software sentiment was absolutely garbage. And these stocks were just getting absolutely punished. So I ended up trimming a couple of my stocks here. And those would have (so far) ended up being some of my top performers, especially U (Unity). But even still, I have a major concern about investing in software, and I could be wrong so please work to decide where you stand on this.

A lot of the capabilities we just talked about can be directly built by the companies making the robots themselves. If you are designing an AV, a robot, a drone, a humanoid, or whatever it is, the planning software is pretty core to the product. So why outsource it?

Then we also have open source robotics software and increasingly capable AI models making some of these tools easier to build. Things like basic navigation, path planning, task planning etc.

So all in all I don’t think that Layer 5 is uninvestible in general, but it does really raise the bar for what I would want to own. And I personally am not prioritizing it at this time.

But nonetheless, when I do invest here I would want to invest in a company when I can understand why the customer can’t eventually build it themselves (or doesn’t want to), replace it, or use something open source instead.

So the companies that interest me most are the ones where the software has become much harder to separate from the broader system. Maybe it’s deeply embedded into the customer’s workload and data or maybe it owns the actual operating layer that everything else connects into.

This leads me to the first two companies I want to discuss:

Manhattan Assosciates (MANH) and Symbotic (SYM).

Before we get into all the companies I am going to present (these 2 + another 3) I want you to understand that just because I am not investing in them today, that doesnt mean the stocks won’t go up. Don’t take these brief summaries of the companies and form your entire investment decision upon them. Use them as a means to do some more research and fill in gaps and stress test my thinking. And then find your own price where you deem the risk:reward in your favor. That’s how I would really like for you to leverage my content!