It is keynote season again, and this October the feeds are full of opinions about the latest big launch event, the slick demo, the standing ovation for a feature that turns out to be a checkbox. My timeline has already split into the people who think it changes everything and the people who think it changes nothing, and as usual both are slightly wrong.
I have stopped watching keynotes for the products and started watching them for what gets left out. The demos are always flawless, which is the first tell, because nothing I run is flawless. The interesting questions are the ones nobody on stage answers because answering them would slow the applause.
The question I always end up wanting answered is: what happens when this thing fails? Not in the marketing sense of "five nines", but concretely. What does the error look like? Can I see the failure before my users do? When the clever new managed service has a bad day, do I get a useful signal, or a spinner and a status page that goes green ten minutes after I already knew?
Because that is the gap between a keynote and a Tuesday. On stage everything is the happy path, narrated by someone who has never been paged by it at four in the morning. The product is real, the capability is often genuinely good, but the demo is a story about success and the job is mostly about managing failure gracefully.
There is also the small matter of cost, which keynotes treat the way menus treat market price. The capability is shown, the bill is implied, and the per-request economics that decide whether a thing is viable at your scale are left as an exercise for the reader and their next invoice. I have watched genuinely clever features turn out to be unaffordable the moment you multiply the demo's tidy example by real traffic, and that arithmetic never makes the slides.
None of this is to be cynical for sport. Some of what gets announced is excellent, and I have happily adopted things I first rolled my eyes at on a stage. The launch is just the trailer. Whether something earns a place in my stack is decided months later, quietly, by how it behaves when something upstream breaks and I am the one holding the pager.
If I had one ask of the people writing these talks, it would be a single honest failure slide. Show me the error message. Show me the dashboard during the bad minute, not the good one. Show me how I find out something is wrong before my users tell me. I would trust the product more for it, and I suspect most engineers in the room would too, because we have all been the one explaining to a manager why the thing from the keynote is currently on fire.
So I will read the takes, enjoy the better demos, and reserve my actual opinion for the first time the thing surprises me in production. That is the only review that counts, and it never makes the keynote.