You Try Texting During a Dinner Rush
A series on the oldest problem in restaurants: getting the right information to the right person. (Post 2 of 3, view part 1 here)
In a restaurant, the thing that decides whether a shift works isn't the food cost or the labor line. It's whether the right information reached the right person in time.
Operators have always known this. What they lacked was a tool to address it. So they did the next best thing and bought software that treated the symptoms. Scheduling tools, analytics dashboards, and inventory systems are all useful, but none of them fix the root cause, so the problems never go away.
We talk about 'rhythms' a lot at Colette. The best operators set the right daily, weekly, and monthly rhythms to keep the team engine running. Pre-shift briefings, team huddles, expediting. All attempts to build that shared picture before and during service. Not every operator has the discipline, but these practices are as old as the industry and still work well enough that nobody has replaced them.
Part of the resistance to change is 'if it ain't broke don't fix it'. But part of it is also structural: two technical problems had to be solved together or not at all, and a tool that solves only one is not going to replace the status quo.
Nobody Has Time to Type
The first problem is input: how do you capture information without disrupting the work?
The team is on their feet, moving, using their hands. The best ones are reacting to what's in front of them and anticipating what the guest needs next. There's little time, often none, to stop and type mid-shift. SaaS tools were built for a manager at a computer. They never reached the team on the floor.
Voice capture has advanced in the past few years. Transcription accuracy crossed a threshold. Models now handle noisy environments, overlapping speakers, and heavy accents far better than they did even a couple of years ago. Language became solvable in real time. A kitchen crew speaking Spanish and a front of house speaking English have always had a coordination gap. But real-time transcription and translation is now commercially viable, so the whole team can speak in their own language and be understood by everyone.
Now a line cook can flag something mid-service without stopping what they're doing.
And Then Somebody Has to Do Something With It
The second problem is synthesis: once it's all captured, who pulls it together?
Paper logs, emails, and text messages have been the standard for decades, and thanks to those advances, voice notes are becoming more common as well. A GM recording a quick note at the end of service, a manager sending an audio log for the opener. Better than nothing. But you still had to remember to listen, read the transcript, pull out what mattered, and decide who else needed to know. The synthesis was entirely manual.
Now multiply that by a whole shift. A dozen people noticing different things, all generating context that matters, all at the same time. The information is out there, but it's fragmented across everyone who was on the floor that night, with no way to pull it together.
Even if you captured all of it, no human could read through it in real time and get the right picture to the right person before the moment passed. Synthesis means making sense of everything the whole team saw, said, and noticed across an entire service.
LLMs make it all possible now. Context windows expanded from a few thousand tokens to over a million, so a full shift (every voice note, every flag, every manager observation) now fits inside a single context. The model can reason across all of it at once: understanding that Jorge calling out, the fish running low, and the food writer at table 7 are connected, and surfacing that link to the right person while it still matters
For now, in the absence of anything better, these problems get solved the way they always have. The floor manager carries most of it in their head. Some things get texted, usually GM to manager, rarely team-wide. Some things get written on a whiteboard by the expo or in the kitchen. Some things make it into the manager log, written at midnight by someone who just worked a double. But a lot of it just disappears.
AI solves this by analyzing all of the information that was captured, understanding what matters and what connects, and putting the right picture in front of the right person without being asked. What the closing manager noticed last night reaches the opener before the doors open. The guest who left unhappy last night is met tonight by a manager who already knows the story. Every shift starts where the last one left off.
The handoffs happen out loud or on paper and then they're gone. AI can solve this communication gap. It can hear everything said across a shift, hold all of it at once, and reason about what matters and who needs to know. Input and synthesis, solved together.