As usual, though, we'll let others deal with the surface-level analysis. We at 3PO-Labs are interested in diving a little bit deeper, especially with regard to the long-term play Amazon is making here. Here's a few ideas we had about why Amazon might be doing this...
"Putting a wall around the state"
In what has become something of a rallying cry for the Alexa dev community, the "Snowcone Dilemma" is a shorthand way of describing the lack of monetization options on the platform. The idea goes, essentially, "Why would I invest time in building quality Alexa skills when I can make more money selling snow cones?".
For one segment of skill devs, at least, Amazon may have just melted the Snowcone Dilemma.
If "data is king", games must be the royal family
Amazon regularly touts Alexa as a system that is "always learning" or "constantly improving". It does that with data. Lots and lots of data. The more times people repeat phrases, the better Alexa's models get. If this change facilitates even a 1% increase in daily engagement with third party skills by existing skill users, that's a lot of data - millions or billions of extra utterances per year, feeding right into Alexa's brain.
There's a good chance that this new data is statistically more valuable, as well. While it's great to have thousands of people say "Alexa, open One Bus Away" every single day, there's certainly an aspect of diminishing returns with such a constrained set of utterances. Games, on the other hand, are likely to have very broad intent models that can take all kinds of distinct and interesting input (proper nouns, emotionally-charged speech, etc), which makes this data more valuable in many ways.
A secondary boon from long-form interactions
The thing is, Alexa and the other voice assistants are still existing in a nascent environment when it comes to user experience for voice. We are starting to see some best practices emerge, but it's not entirely clear yet what form skeumorphism, affordances, and other visual UX concepts might take in a voice-first world (or if they should even exist at all). With a proliferation of game skills and the accompanying move towards increased quality, Amazon can treat the game ecosystem as a giant experiment in voice UX, essentially comparing skills in an almost apples-to-apples or A/B test to see what users like and what they don't.
Speaking of which...
Quid pro quo
Take the idea of prompts during an open session - which is notorious to skill developers as "Checklist item 4.1". Many of us in the development community believe that ending every Alexa speech with a question or prompt is a huge mistake in terms of the experience for our customers. For most users it is readily apparent when Alexa is still active and whether or not you're expected or able to add input of your own. Amazon, however, seems to be optimizing certification for the least savvy set of users - those who may actually require a prompt every single time. Many of these users probably aren't interested in actively engaging in a training program to be better users of the platform, but if those users could be passively trained through rote repetition by the games they enjoy, it would be a boon to all of us.
Likewise, there are many features that even expert users likely take for granted - the ability to ask for help, for example. Further, there are sure to be more advanced features added later (either platform-wide, or by skill devs) that will require training even for those already familiar with the platform. Games provide a perfect avenue for user training.
There's an auxiliary benefit for users here too, which is that longer engagement with a single skill means less skill hopping, ergo less skill names to remember. Skill name fatigue is a very real consumer complaint, and until Amazon solves top-level invocations it's something they'll need to work around. Depending on how they've optimized the payouts, it may be the case that we see skill developers migrate away from individual micro-skills, and towards composite macro-skills. An example here might be a "3PO Card Table" skill replacing both "3PO Blackjack" and "3PO Hold'em". Amazon has the ability to influence this ever so slightly with their payouts.
Throwing a bone to the guys behind the scenes
This is an obvious windfall for folks like VoiceLabs, but we at 3PO-Labs expect even our humble utilities (ASKResponder and our as-yet-unreleased-and-super-secret tool) to see renewed interest.