Every few months, the same question circles through the AI industry: as frontier models keep improving, what's left to build? What survives a technology that gets cheaper and better on a schedule?
Julie Yoo of Andreessen Horowitz recently framed an answer for healthcare. The durable businesses, she argues, will be both AI-Native — built to take full advantage of frontier models — and AI-Proof — unlikely to be disrupted by them. And in healthcare, the layer that holds those two together is accountability: licensed care delivery, bearing financial risk, regulated products. The model layer gets commoditized; the layer where someone must stand behind an outcome doesn't.
The same logic applies to trading software. It just looks different — and it explains a positioning decision LY Bots made before the AI conversation was fashionable.
In trading products, the part that "AI" would naturally improve is the intelligence layer: the analysis, the signal, the forecast. The headline. The reason a trader is told to buy.
That layer is precisely the one being commoditized. If a frontier model can summarize markets, generate setups, or mimic "analysis" at near-zero cost, any product whose value is its intelligence loses that value on the model's release schedule. The moat evaporates quarterly.
LY Bots does not build that layer. Not because it can't be built — because it's the least durable part of the product.
A prediction is a promise about the future. No software can keep that promise reliably, and any product that sells the promise inherits the liability when the future disagrees. That is not a moat. That is a lease on a commodity.
In healthcare, accountability is a licence. In trading software, it's something quieter — but it does the same job: it puts the product's weight behind something the model layer can't touch.
Refusing to promise what can't be kept. No performance claims. No "consistent returns." No "AI predicts markets." The absence of hype is not modesty — it's the product declining to take on a liability it cannot honour.
Standing behind what is delivered. The 100-Day Test Guarantee exists for a reason: run the software for a real period, and if it isn't what was claimed, a full refund is available between day 100 and day 115 — after the test, not before. That is accountability expressed as a policy.
Doing the integration work the model never touches. Compiled MT5 files. Broker compatibility. Session-aware logic tuned per instrument — because an index doesn't behave like a currency pair, and gold has its own rhythm. Setup guides. Clear legal documentation. None of this improves when the model improves. All of it is what makes the software actually usable.
Here is the counterintuitive part: in trading software, the less a product promises, the more defensible it becomes.
Products that compete on intelligence compete in a race with a fixed endpoint — the moment the model catches up, which the model's own roadmap guarantees. Products that compete on trust and integration compete in a race with no endpoint, because every instrument, broker, and market condition adds work that a model cannot abstract away.
The value sits where the promise is kept, not where the promise is made.
A trading bot is discipline, expressed as software. It observes. It checks conditions. It acts — or waits. No opinion, no discretion, no "just this once."
That mechanism is the accountability layer in its smallest, most honest form: the software commits to the rules it was given, and the company behind it commits to the claims it made. As models improve, that layer does not weaken. It becomes the whole product.
Discipline isn't a feature. It's the mechanism.
