skip to content
fdbck lps

Calendar AI hasn't solved my problem (yet)

/ 4 min read

updated:
View more blogs with the tag work , View more blogs with the tag systems
table of contents

I’ve been thinking a lot about AI tools in meetings, probably due to lingering discontent with the reinforcing feedback loop of platform decay (opens in new tab).1 For example, Apple was just announced to have acquired an AI calendar app (opens in new tab) that will surely reappear at tomorrow’s WWDC,2 Copilot offers lots of calendar integrations, and a quick Kagi search reveals dozens more AI-enabled assistants that promise to book meetings faster, handle conflicts automatically, and generally optimize my work day. But I don’t think these tools are solving the right problem for me!

First, I don’t want a digital assistant that makes it easier to manage meetings, I want one with sufficiently aligned autonomy to know when not to have one in the first place.3 And second, these AI applications make it easy to book meetings just in time (opens in new tab), so it looks like they solve a critical bottleneck for the people in 30+ hours of meetings every week. But The Principles of Product Development Flow offers a fun counterargument:

… it is important to go beyond the popular but simplistic idea that the capacity of the bottleneck controls system flow. In fact, flow through a bottleneck is affected strongly by the process that precedes the bottleneck. The upstream process determines the variation in the arrival rate at the bottleneck, and this affects the queue. Managing the process upstream of the bottleneck is a valuable tool for improving flow at the bottleneck.

So we shouldn’t merely stop at the meeting bottleneck, but rather look upstream for higher-leverage process improvements. And what’s upstream of a meeting request? Precisely all the unspoken, unwritten context that’s hard to feed an AI assistant! For instance, it might not capture the preference falsification (opens in new tab) inherent in business relationships or the adverse selection caused by the resulting social capital dynamics;4 how many “could’ve been an email” meetings you’ve been party to recently; situations where your counterparty hasn’t Read The Manual; or times where nobody’s really sure why there’s a meeting in the first place. This is all stuff I think about sometimes, but how do I get the AI assistant to recognize it?

While we wait for AI to catch up, what I’d really like is a Rothian (opens in new tab) clearinghouse that matches meeting requestors with subject-matter consultants.5 This would thicken the market so that I can delegate effectively; it would enable deferred-acceptance matching schemes, which are strategy-proof and therefore improve business transparency; and it’d even allow batch-accepting invites so that nobody has to reprioritize whenever someone higher on the org chart needs to squeeze in a new meeting.6 We’ve basically independently recreated the sprint, but for meetings!

If in some distant future you’ve vibe-coded your way into founding a B2B SaaS company based on this idea, all I ask is that provide a requisite shoutout on your About page!

Footnotes

  1. Or is it just Sunday night and do I have a bunch of meetings tomorrow… ↩

  2. Okay, maybe not! ↩

  3. As a leader, you can fact-find in many ways, each with effects of different magnitude on system equilibrium. Calling a meeting is an official act that signals a certain blunt-force finality, and these optics change how participants think about, talk about, and do the work you want to discuss. This is a useful foil to the “see risk, book meeting” strategy that you’ll often see deployed on autopilot. Plus, over-communication is a failure mode (opens in new tab). ↩

  4. If you’re honest with yourself, you probably don’t want to meet with everyone who wants to meet with you! And the converse holds too, if you’re being extra honest. ↩

  5. See “What Have We Learned From Market Design” (opens in new tab). ↩

  6. The convenience of the marginal meeting time slot is strictly negative. People pick the best times first, so last-minute stuff either lands on Fridays at 4:30 pm, or you’ve got to reshuffle your entire calendar to free up time earlier in the day. The high-order shock wave this sends through the org’s calendars actually is a legitimate use case for an AI calendar assistant, but even so it’d only treat the symptom. ↩