Thought provoking paragraph buried in this article about how the bot/agent interest is a reaction to the decline of apps: "More than anything else, the rise of the bots signals the death of the mobile app ... Facebook isn't the only company to notice that the whole app thing didn't really work out. There are a few wildly popular apps (Facebook among them), but generally people don't like downloading a zillion apps to their mobiles. Why have separate doorways to every single thing you do online when you could just have one doorway called Messenger?"
6 plus ones
Shared publicly•View activity
- A similar question might be, "Why not have one doorway called the web browser?"
Sometimes, I wonder if the app trend and this new bot trend might just be caused by clutter and excessive, annoying, and irrelevant advertising on the mobile web.
As I think we've seen with apps, the clutter and ad-free initial early adopter experience on these new surfaces quickly becomes the same cluttered and ad-filled experience of the Web as the surface matures.
Also, I can't help but point out how old this bot idea is and how many attempts there have been at it. Here's some great work from 1994 on it:
Bots are still a good (but hard) idea, but learning from history is worthwhile on this one, unless we just want to repeat it.50w
- Most of the bot hype is ill-conceived.
Bots won't kill apps but they will cause people to think more carefully about if something should be an app or a bot or a web app.
I also find it more enlightening to think of bots as agents instead. It moves people away from thinking about gimmicks, IRC-style chat bots, and pipe dreams like conversational commerce which are little more than glorified phone trees.
Instead people start to think about personalisation, persistence, context, identity and proactivity.
Imagine an agent that can act on your behalf (because it has your OAuth token) without having to be told but which reports its actions and results via a messaging app you're already using. Agents that react to explicit commands are a tiny fraction of the interesting agents. The more interesting ones hook into the user's data and perform actions where it's hard for the user to even formulate their intent but where the agent can infer and execute it.50w
- I kind of think of it like IM versus email. With an IM system on my phone, I can have a high volume of trivial interactions, but it's hard to have any in-depth interactions. With both email and the web, it's easy to find specific high-value interactions which can be better done in other ways, but it's really hard to also accommodate the long tail without building in your own set of annoying problems.
Something which often bothers me about this kind of thing is that the problem being solved is relatively trivial. Dialing the phone to order a pizza really isn't so bad, and it's really robust in the face of minor problems that come up. If I wanted to do it by computer, I wouldn't want some algorithm guessing what I want based on free text, I'd want to click around on a menu, because the pizza place doesn't have a free-form kitchen, they have specific ingredients. Web pages are great at presenting lots of options and letting you make selections.
Another frustration I have with this kind of "advancement" is that the examples often involve removing a bunch of friction which exists in other systems for a reason. For instance, if I could engage a chat bot to order something and it all just keys off my saved credentials to do payment and everything ... well, then why couldn't the same thing happen with an order online or by phone? And if someone is capable of programming a bot to effectively predict and respond to my goals, why couldn't the same thing happen online? Hint: Because these are damned hard problems.50w
- Grr, just hitting a perfect example. Want a longer charging cable for my bedside table. I query Amazon with "usb-c to usb-c 6-foot cable" and it turns up a motley collection, including 6-foot USB-A-to-Micro-USB, a parallel printer adapter, a 10-foot USB-A-to-USB-C, etc. There are like 3 plausible candidates in the 20 results it gave me, #1 is a reasonable match, #2 is not.49w