I Tried to Put AI Into the Race. Then I Realized It Didn't Need AI.

In the previous post, I started working out what EvoWild Run is actually supposed to be as a game.

Raise Creatures. Breed them. Continue their lineages across generations.

Race them on Courses with different conditions, then use the results to influence the next breeding decisions.

Eventually, players may also be able to create Courses themselves.

But one major question was still unresolved:

What does the player actually do after the race starts?

I left that question open in the previous post.

What Is EvoWild Run Actually Supposed to Be?

So that is where I started this time.

I don't want the player directly controlling the Creature during the race

The first decision was that EvoWild Run should not become a game where the player directly controls the Creature after the start.

Move left and right.

Press an acceleration button.

Time your inputs through corners.

If I make that the core of the race, the player's action-game skill starts becoming more important than the Creature they actually bred.

What I want to matter more is:

• what kind of Creature you raised
• which Creature you chose to enter
• which Course you selected
• what strategy you prepared
• what you learned from the result
• how you changed the next generation

So the rough structure became:

Breed → choose a Creature → choose a strategy → watch the race → analyse the result → feed it back into the next decision

Up to this point, the idea is still fairly close to a horse-racing game.

Then I wondered: what if AI made the decisions during the race?

That led to an AI idea.

If the human player is not directly controlling the Creature, perhaps an AI prepared by the player could observe the race and make decisions instead.

For example:

• the path ahead is blocked
• the remaining distance is getting short
• the Creature is heavily fatigued
• an outside path is available

The AI could respond with commands such as:

CONSERVE
MOVE OUTSIDE
ATTACK

`

The AI would not directly move the Creature.

The structure I had in mind was:

text
AI / Agent
↓
evaluate the situation
↓
issue a command
↓
Creature attempts to execute it
↓
Race Engine determines the actual result


That is where the idea of a Race Agent started.

But it felt strange for only AI users to have an Agent

The next problem was obvious.

What happens to players who do not use AI?

If AI users have a Race Agent while everyone else uses a completely different system, then I effectively have two different race systems inside the same game.

So I changed the idea.

Every racing Creature should have a Race Agent.

A normal player would not need to write an Agent File.

They could simply choose understandable options in the game UI, such as:

• push forward / standard / hold back at the start
• conserve / standard / aggressive base pace
• prefer inside / natural / prefer outside
• wait / look for an opening / overtake aggressively in traffic
• early / standard / late spurt timing

Beginners could use simpler presets.

Internally, those choices would be converted into a common Agent File.

An advanced player could instead write the Agent File directly.

Or they could ask ChatGPT, Claude, Gemini, a local model, or another AI to create one.

The important part is:

an Agent created through the normal UI and an Agent created with AI eventually become the same kind of Agent File.

AI users do not get access to special commands.

From the Race Engine's point of view, they are the same thing.

And then I realized the Agent didn't actually need to be AI

This was the point where the idea changed.

Suppose the Agent File contains a rule like this:

`text
remaining distance <= 400m
AND
fatigue < 70%

↓

ATTACK
`

That is not AI.

It is just checking conditions and selecting a predefined command.

Games have been doing variations of this for a long time: instead of directly controlling a character, the player defines tactics, priorities, commands, or rules in advance.

The name Race Agent sounds AI-related, but the runtime system itself does not actually need AI.

So I had started with:

What if I put an AI Agent into the race?

And after breaking the idea down, I ended up with:

The race does not need AI running inside it at all.

What remained was essentially a tactical command system.

I don't think Live AI is necessary at the beginning either

It would still be possible to connect an external AI during the race and let it make decisions in real time.

But then a different set of problems appears:

• API latency
• connection failures
• differences between models
• differences between free and paid AI services
• API costs
• dependency on external services
• competitive fairness

I would be designing a racing game, but suddenly the price and performance of external AI services would also become part of the competitive environment.

So for the initial version, I think this is enough:

Create an Agent File before the race, then let the Race Engine execute it.

Live AI can be reconsidered later if there is a reason to add it.

Just because AI is involved somewhere does not mean an AI model needs to be running throughout the race.

AI can still help create the Agent instead of running it

I did not remove AI completely.

I moved it to a different part of the process.

For example:

`text
Creature abilities
Past race results
Current Course conditions
Existing Agent File

↓

External AI

↓

Improved Agent File

↓

Validator

↓

Race Engine
`

A player could ask an AI to analyse a Creature's previous races and produce an improved strategy.

A player who does not want to use AI could create the same kind of Agent through the normal game UI.

So:

AI is optional.

It can help with strategy creation and analysis.

But the game rules themselves can still work without it.

At least for now, that seems like a much better fit for EvoWild Run.

This feels a little like the blockchain question again

EvoWild Run originally started from me thinking about blockchain games.

The question was not:

If this is a blockchain game, where do I put the token and NFTs?

It was:

Does this game actually need blockchain at all?

I seem to have ended up doing something similar with AI.

I started with:

Could AI make the race more interesting?

Then I separated the idea into its actual components and found that the Race Agent did not need AI in the runtime.

But the Race Agent itself survived.

So the next question became:

If this is now just a tactical command system, how do I make it interesting enough to become part of the game rather than just another settings screen?

That is what I started thinking about next.