When most of the order flow is bots: what actually changes

Equities crossed this line years ago. Most order flow there is automated, and the market did not stop working — it changed shape. Crypto is on the same path with a shorter history and fewer rules, and the shape it is changing into is already visible if you know where to look.

What automation did to markets that got there first

Three effects show up every time, in the same order:

• Spreads compress. More automated liquidity means tighter quotes. Good for anyone taking liquidity; brutal for anyone whose strategy needed a wide spread to work.
• Simple edges die fast. A pattern anyone can code gets coded by everyone. The half-life of a publicly describable edge is measured in months, not years.
• The remaining edges get more expensive. What survives requires either infrastructure other people will not build, data other people do not have, or patience other people do not have.

None of that is a prediction. It is a description of what already happened in equities, futures and FX, and there is no structural reason crypto is exempt.

The part people get wrong

The usual framing is a race: better model beats worse model, faster bot beats slower bot. That framing survives right up until everyone has a good model, and then it stops explaining anything.

A trading edge is a claim about a difference between you and the rest of the market. It is inherently relative. If a capability becomes universal, the difference goes to zero — not because the capability stopped working, but because the price already contains it.

So what survives

Empirically, four things, and none of them is “a better model” in isolation.

The pattern: what survives is not intelligence, it is either infrastructure or a willingness to be somewhere other people are not.

What gets better for everyone

It is worth saying the optimistic part plainly, because it is also true. More automation means tighter spreads, deeper books, faster price discovery and fewer obviously exploitable retail traps. The floor rises.

For someone building a bot today, that floor is a gift. Reliable APIs, documented WebSocket streams, cheap historical data and open tooling did not exist a decade ago. The cost of building a competent system has collapsed. The cost of building a differentiated one has not.

What this means for your project

Three practical consequences we apply to our own client work:

1. Assume the obvious version is taken. If a strategy can be described in one sentence and coded in an afternoon, price that in. It may still work; it will not work exclusively.
2. Spend on execution before spending on signal. A mediocre signal executed well often beats a good signal executed badly, and execution quality decays slower because it is engineering rather than information.
3. Measure decay, not just performance. Track whether the edge is weakening over time. Most strategies die gradually and their owners notice late — see model drift.

The honest position

We build trading bots for a living, so treat this with appropriate suspicion: our view is that automation is becoming table stakes rather than an advantage, and that this is fine.

You do not automate to beat other bots. You automate because your rules are better executed by a machine than by you at 3am, because you want to cover seven hundred instruments instead of seven, and because a system that records every decision can be improved while a discretionary process cannot. Those reasons survive any amount of competition, because they are about the gap between you and your own execution rather than the gap between you and the market.