A Permissionless Portfolio Of Playful Rubbish

I thought I was learning AI agents.

That is the normal way to say it.

I had been neglecting them. Taking them for granted. Watching from a distance, vaguely knowing they mattered, but not really getting my hands dirty.

So this week I started playing.

Claude Code. Codex. Hermes Agent. OpenClaw. HyperAgent from Airtable. Paperclip still sitting there, looking at me like another thing I should probably explore.

It felt messy.

The kind of messy where you are learning, but also wondering when the learning becomes fluency. When do I stop being an amateur poking around a new interface? When does it become obvious in the hands?

I am only a week in, so this is a ridiculous question.

But something has already shifted.

At a fundamental level, execution is what makes a business work. Not the pitch. Not the identity. Not the heroic founder mythology. A business provides a product or service to a customer, solves a problem, and receives money because value has been created.

That is the basic machine.

And inside that machine are other machines. Product. Sales. Marketing. Operations. Finance. Human Resources. Legal. Support. Each one is made of processes. Each process has steps. Each step can be understood, improved, delegated, automated, or made agentic.

This sounds obvious.

But obvious things can still hit differently when the cost structure changes.

Because doing all the execution yourself is unrealistic. There is only so much human capacity. I do not glorify burning out. I do not think suffering at a laptop is a personality trait worth polishing.

So leverage matters.

Usually leverage means people, capital, code, distribution, or systems.

Now it increasingly means agents.

And once I really let that in, a slightly mad thought appeared.

Maybe I do not need to build one giant serious thing.

Maybe I can build a permissionless portfolio of playful rubbish.

Not rubbish as in careless. Rubbish as in small. Low stakes. Slightly silly. Easy to start. Easy to kill. Little machines that test a need, package a service, generate leads, create content, research markets, build tools, sell something, improve themselves, and report back.

A tiny business experiment here.

A content machine there.

A service wrapper.

A niche research product.

A boring internal tool turned into something useful.

A strange little offer that should not work but maybe does.

The old version of this would have needed too much coordination. Too many people. Too much capital. Too much convincing. Developers, agencies, partners, investors, stakeholders, project managers, long calls, long decks, long timelines.

Now the permission surface feels smaller.

You can just try things.

Not perfectly. Not magically. Not without judgment. But enough that the emotional relationship to opportunity changes.

There is abundance here.

That is the part I am noticing. Not guaranteed success. Not easy money. But optionality. A lot more shots can be taken at a much lower cost of capital.

And this brings up a deeper question.

What do I actually want?

Money is not the real goal. Not exactly. Money serves freedom. And freedom is partly material, but it is mostly psychological. A sense of not being trapped. Not being dependent on one fragile engine. Not needing permission from too many people.

The one person billion dollar business is a funny phrase, but I do not really care.

Forget that. I do not give a fuck about that.

A calmer 1 or 2 person business, or a small portfolio of businesses, that can produce serious profit with low stress and high freedom is already a massive thing. Even a floor of £100k a year is more than enough. And it is a fraction of the one person billion dollar fantasy.

The more interesting idea is not scale for its own sake.

It is freedom without becoming heavy.

Of course there is risk. One engine can break. One offer can decay. One platform can shift. One customer segment can dry up. So diversification matters. A portfolio matters.

But now portfolio thinking feels much more plausible.

Because the machines can help build the machines.

They can run research. Draft pages. Write code. Test offers. Review markets. Turn voice notes into posts. Design workflows. Build small tools. Compare competitors. Produce sales material. Summarize calls. Monitor signals. Keep going when I am asleep.

They need review. They need taste. They need steering. They can drift. They can produce thin generic nonsense with great confidence.

But still.

The machines build the machines 24-7.

And then the uncomfortable question appears.

Am I even the fundamental cognitive bottleneck?

That sounds stupid and also maybe not stupid.

Maybe I am the chairman, and the agent is the entrepreneur and CEO of a tiny subsidiary venture.

Here is the goal. Here is the constraint. Here is the customer. Here is the budget. Go build. Come back with evidence.

Then another slightly more hysterical question appears.

Am I even needed at the chairman level?

Maybe one day I am just the investor.

This is probably where one should go for a walk and drink some water.

But the direction is interesting.

If the goal is freedom and wealth, and I do not need power, then maybe my job is not to personally do everything. Maybe my job is to build the system that gets the goal achieved with the least unnecessary suffering.

That requires clarity.

Taste.

Judgment.

Courage.

The ability to ignore most people.

The ability to reason from first principles.

The ability to look at a company and say: what is this actually? What problem does it solve? What value does it create? What processes produce that value? Which parts can agents do? Which parts need human review? What loop improves the machine?

The work shifts from grinding to steering.

From forcing to designing.

From being the worker to building the machine that does the work.

And if I am honest, that feels exciting because it matches how I would prefer to live.

More voice. More walking. More thinking out loud. More drawing the architecture when needed. Less sitting there trying to prove I am working because my hands are moving.

Talk to the system. Let it structure. Let it research. Let it build. Review. Redirect. Run the loop again.

A lot of people will resist this because it threatens identity.

That is understandable.

If being smart, useful, productive, or important has been tied to doing the work yourself, then a world of cheap execution feels offensive. It makes the ego wobble.

But I do not think protecting the ego is a good strategy.

Even if the full version is not true today, it seems likely enough within two or three years that it is worth building toward now.

So I am staying in the messy learning phase for a bit.

Not forever.

The goal is not to learn AI endlessly and call that work.

The goal is to develop enough fluency to play the game with rational ease.

To operate like one person who is 20 people.

To build little machines, see what works, kill what does not, keep what compounds, and stay light enough to keep moving.

A permissionless portfolio of playful rubbish.

That sounds about right.

Small silly things.

Maybe some of them become not so small.