API.AI is really great but I really don't like this UX.

From using and developing for the Amazon Echo for the last year or so this "Conversational" setup just sounds super unnatural.

You want bus times: Amazon: "Alexa, ask [nextbus] [for the next bus]" Google: "Hey Google, let me talk to [nextbus] <wait> when is the next bus?" Perfect: "[keyword], when is the next bus?" (context set in advance for the specific bus you care about)

You want to control the lights: Amazon: "Alexa, turn off the [kitchen lights]" Google: "Hey Google, let me talk to [the house] <wait> [turn off] the [kitchen lights]" Perfect: "[keyword], turn off the lights" (contextually turn off the lights immediately nearest the source of the voice. optionally "turn off the [named] light"

You want to activate a scene: Amazon: "Alexa, [turn on|activate|start] [scene name]" Google: "Hey Google, let me talk to [the house] <wait> activate [scene name] Perfect: "[keyword], [time for a movie|it's bedtime|good morning]"

I haven't seen or heard of a single "Conversational" app for the Echo that has real usage. I just don't think that's the killer app (at least until they get smarter). The current killer apps are being able to: - Set timers/alarms - Ask quick questions (measurements, weather, time) - Control home automation

I just don't think the experience of talking back and forth to a bot is that enjoyable (again, yet, maybe in a Her future).

From the documentation, they also support Amazon Echo style actions such as "Tell Personal Chef to find me chicken recipes" or "Ask $name to $action_phrase".

See https://developers.google.com/actions/distribute/invocation-...

Spot on! At the core of the bot trend is the fact that users do not want to navigate through menus anymore. They are tired of searching for that one icon on screen after screen so they can enable a function. So they just want an interaction at the root that feels natural and gets them to the goal quickly.

I was pretty impressed the other day by what hobby roboticists are now able to build leveraging stuff like api.ai:

http://imgur.com/a/ue4Ax

I don't think that the newly released Actions always expect a conversation. In scenarios where it makes sense to just reply with an answer you can still do that. I am guessing you just have to use something like 'Hey Google, can you check when is the next bus to SFO on <app name>"

I think that, over time, both paradigms are required. For the quick lookup/command, "ask" is best. For longer interactions, like driving directions, or cooking instructions, etc... a "conversation" makes more sense. Given that we're at the early stages of conversational UIs, I think the 'ask' case is more helpful, but I'm more excited to see what will evolve through conversations.

This is being made a platform for external developers, unlike Google search for instance. So what makes more sense for users is only important on large averages and to the extent users don't get mad enough to abandon it.

What matters for action developers ?

Having people buy shit

Keeping people occupied longer (for commercials)

(for companies that do other things) Preventing people from trying to communicate with a real person

IIRC, the perfect examples where you neither say "Alexa" or "Hey Google" aren't going to happen any time soon - those recognition phrases are hard-coded for performance reasons, and adding custom ones will be difficult.

That's not true. Most conversational platforms do use a separate "trigger word" subsystem which has a much smaller vocabulary and is optimized for the low power, always on scenario. But the trigger words aren't hard coded - for example on the Motorola X series you could (can?) customize the trigger phrase: https://motorola-global-portal.custhelp.com/app/answers/inde...

I think "[keyword]" is meant as a placeholder for either "Alexa" or "Hey Google".

Privacy reasons as well. Every ounce of what you say after "Alexa" or "OK, Google" is submitted to Amazon/Google.

I don't see them changing this ever, so having a recognition phrase is perfectly fine by me.

Performance may be a small part of it, but it's more about branding. The trigger word is the brand, the functionality is the product.

It's a nod to Siri requiring the app name, Siri API was punted from iOS 9 to add that, it was considered a privacy risk.

I don't agree FWIW.

You may be looking for https://godoc.org/golang.org/x/net/context which is likely to make it into stdlib in the 1.7 timeframe from what I've seen in various channels. Doesn't quite solve the global logging problem though.

Yep, that's the context we (I and the grandparent comment) are using.