"Set a reminder for tomorrow" should just work. That's the promise of an AI partnership platform—you talk to it like you'd talk to a person, and it understands.
Except when it doesn't. We discovered that Sabine's reminder skill was rejecting natural language date inputs. If you said "remind me tomorrow" or "set a reminder for next week," the skill would fail silently or ask you to reformat your request. The very thing that should feel most natural—conversational date references—was broken.
What We Fixed
The reminder skill now properly parses natural language date inputs. "Tomorrow," "next Tuesday," "in three days"—all of these work as you'd expect. We overhauled the date validation logic to accept conversational input and translate it into the structured format our backend requires.
This matters because trust in AI products is fragile. Every time a user has to adjust their language to accommodate our system's limitations, we're asking them to meet us halfway. That's backwards. The system should adapt to human communication patterns, not the other way around.
Why This Happened
The root cause was overly strict validation in the reminder skill's parameter parsing. We were checking date formats before passing them to our natural language processor, which meant casual inputs never made it to the layer that could understand them. A classic case of defensive programming gone too far—we were protecting against bad input so aggressively that we blocked good input too.
What's Next
This fix is live now, but it's part of a broader effort to improve how Sabine handles time and dates across all skills. We're auditing other skills that accept temporal input to make sure they're equally flexible. We're also building better testing infrastructure around natural language parsing so we catch these issues before they reach production.
The goal isn't just to fix bugs—it's to build systems that feel effortless. When you tell Sabine to remind you about something tomorrow, it should feel like telling a friend, not programming a computer. We're not there yet, but we're getting closer.