When your AI coach promises a change it never made

Published 10 September 2026

During testing, a lifter spent eight back-and-forth messages with the coach settling on dumbbell shrugs for their next set. The coach agreed: "Perfect. Let me add that in. Accept it when you're ready and knock out those shrugs." Nothing happened. No card, no button, nothing to accept. The shrugs were never added — the coach had described doing it, in perfectly natural language, without actually doing it.

That's a real failure mode of AI coaching, and it's worth being straight about it rather than pretending it can't happen: sometimes the model writes the sentence that describes the right action instead of taking it.

Why this happens

Adding something to your plan isn't just talk — it's a specific action the coach has to actually take, separate from the words it writes back to you. Nearly every time, both happen together and you never think about the difference. But every so often, the coach can write the sentence without doing the thing — same as a person absent-mindedly saying "I'll send that over" and then not sending it.

What Yield does about it

Rather than just telling the coach not to do this again — an instruction that had already failed to hold — Yield checks for the specific pattern of the coach describing an action it didn't actually take, and surfaces a correction when it catches one. It's a safety net underneath the conversation, not a promise that the underlying mistake can never happen — it's the honest acknowledgment that it can, and a way to catch it when it does.

If a plan you agreed on with the coach doesn't show up as an actual card you can accept, that's worth telling us about directly — it's exactly the failure this exists to catch.

Yield Strength is in beta on TestFlight. Read more about what makes it different, or join the waitlist to try it yourself.