What you take away

#The last half hour

At a quarter to five on the third day, with the game live and more than four thousand matches played, Roshan asks the question that organises this whole course: if you started another business tomorrow, what would you do differently?

The three answers do not overlap. This is the lesson that collects them.

#The hard part is not building it

The first answer is Matt's, and it starts with the predictable: once they agreed on an idea and started shipping, everything went much faster than deliberating.

But it continues with something he did not have to say, and that is what gives the rest of the livestream its credibility.

Day 3 · 7:42:31Would anyone have played this without the livestream?

They had distribution handed to them. Thousands of people watching, so getting players was trivial. He asks out loud whether anyone would have played that game without the stream behind it, and answers that they would not.

Building is the easy part. Getting anyone to care is the hard one. A course about agents that does not say this is selling a mirage, because what agents make cheap is precisely the part that was already easy.

#Restraint when capability is abundant

Lauren supplies the counterweight.

Day 3 · 7:45:31With AI you can build anything, and that is the danger

With these tools you can build literally whatever occurs to you. Which makes it very easy to end up with a product that has a thousand features and is no good. You need discipline exactly when capability is abundant, which is when you least feel like having it.

She mentions that internally they talk about what they are going to unship, not only what to add. They applied it here: they threw out the complicated stat blocks on each card, threw out the separate battlefields, threw out an entire mechanic. They kept one kind of combat. That is what let them launch.

So the useful question is not only what do I add this week. It is what do I remove.

#Where to start

The two answers to "what would you tell someone building something with this" are different and complementary.

#Build something you already know how to do

Day 3 · 7:47:24Build a business around something you already know how to do

Matt illustrates it with what happened on day one. A guest rattled off ten things about running an event that they had never considered: permits, catering, whether the food is hot or cold, whether you need a kitchen. Things you do not know until you have done it.

His conclusion is that doing something consists, unintuitively, of doing it. And he finishes without dressing it up: you do not make money without doing anything.

#And create Bots for what you do not know

Day 3 · 7:48:37Create Bots that cover what you don't know

Roshan adds the other half. Their game design BotsBot An AI teammate with a name, a job and its own conversation. It works on a computer in the cloud, even when you're away. See in the glossary → learned by studying the best games. He admits he used to undervalue how much context you can hand a Bot, and links it to what a guest had said that same afternoon: send them off to take courses in the craft they lack.

His recommendation: assemble a varied team and think of it as hiring people.

Together the two answers give you the criterion. You bring the craft you already have; the Bots cover the gaps. It does not work the other way round, because if you know nothing about the subject you also cannot tell whether what comes back is any good.

#And then Lauren says the unexpected thing

Day 3 · 7:49:02The most valuable thing in the era of agents is people

On a three-day livestream launching an agent product, the final answer is that the most valuable thing is still the people you work with.

And she makes it concrete. A team needs two profiles: someone with a clear idea of what should be built and good judgement about what goes into the product, and someone relentless about making things happen. Better still if it is the same person.

She adds that this role will stay human no matter how capable the models get. Coming from someone who has spent three days teaching people to delegate to machines, it is the heaviest line in the whole stream.

#The method, in one sentence

Roshan sums it up before closing, and it works as a definition of everything in this course.

Day 3 · 7:44:12You discover the problems as a human and turn them into Bots

You go out, as a person, and discover the problems on the ground. And once you know how to solve them, you turn that into a Bot and leave it running forever.

That is why the order matters. First you understand the problem, then you automate it. The other way round you end up with a very efficient system solving the wrong thing.

#The real product was never the game

Day 3 · 7:51:48This was never really about building a game

He says it at the close: this was never really about building a game, nor even about building a studio.

What they take from three days is not the product. It is the factory: a chief Bot coordinating a team, others playtesting the game, others fixing pull requestsPull request A proposed change to the code, packaged so someone can review it before it's accepted. See in the glossary → and approving them on their own once the bug is gone. That carries over to the next game, and to whatever comes after.

It is also what you can take from this course. The mechanics of a card game are no use to you. The loop that produced them is.

#And the ending, unvarnished

Two things happened in that last half hour that were not in the script.

The first: while Lauren was talking about restraint, her software factory took production down with a bad SQL query. Signups closed for a while, live, on exactly the wrong sentence.

Day 3 · 7:51:15Production goes down while they talk about restraint

The second: they tried the advertising at the end and it half worked. They uploaded a placeholder logo, bid a dollar, and admitted moderation was broken.

Day 3 · 7:57:31Our first theoretical sponsorship dollar

And that is how they closed: "we made our first theoretical sponsorship dollar, and I think that's a great way to end it."

Those two details are not anecdotes. They are the proof the rest was real. Three people built a company in seventy-two hours with a team of Bots, and still took the database down and never collected a euro. Both things are true at once.

#What you take from this course

• Start with something you already know how to do, and create Bots for what you do not.
• Discover the problem yourself; automate it afterwards, never before.
• Ask what you are removing, not only what you are adding.
• What you build matters less than the loop that built it.
• Getting anyone to care is still the hard part, and no agent solves that.
• The judgement, and the people you work with, are still yours.

That is the end of the course. If any of this was useful, the best thing you can do now is close it and record the fifteen-minute voice memo from lesson 07. That is how she started, and it is how anyone starts.