Here's the dream scenario (as seen on TV). A busy knowledge worker tells her smartphone to find an Italian restaurant that is pet-friendly and to text the result to her dining companion. Jump cut to the couple sitting in front of a chic Italian restaurant with their dog. Message: your voice and your smartphone are all you need to search, shop, and complete transactions.
At least that is one of the scenarios that gets us from the opening scene to the coda. Explicit in the messaging: Google Gemini running on Samsung can understand spoken words, carry out a search, and text results to concerned parties. Undepicted are several intermediate steps that would culminate in gaining approvals from participating individuals (companion and dog), including agreeing on a time and completing a reservation.
In the interest of time (a 15 second spot), the producers have elected to streamline what can turn out to be rather complex workflows and processes. That's a shame. There are a lot of business opportunities and lessons to be learned before we can unleash the power of automated personal assistants on the world of mobile and online conversational commerce.
Our too-busy-to-do-this-myself TV protagonist represents the perfect demo (meaning demographic) for showing the power of personal agents. She packs her prompt (yes, it is a prompt) with all the criteria that represent her intent (dinner that night). Some form of artificial intelligence is successful in fulfilling her request for a nearby, dog-friendly, outdoor Italian restaurant, and generating a response that is shared with her dinner date. Those are the modest, first-order tasks that Gemini on Samsung performed.
Here on earth, the more complicated stuff happens after the obligatory form-filling and notification. AI agents bring the promise of acting on their end-users' behalf, starting with understanding their overall intent, then parsing it into a number of intermediate tasks that need to be carried out. Successful completion of those tasks, whether it is booking a table at a restaurant, booking travel, shopping for kitchenware, or returning merchandise to a retailer, relies on an intelligence that can carry out complex tasks, often autonomously on their end users' behalf. They are the machine customers that carry out transactions on behalf of their human counterparts.
To do so, they must have their human client's permission and, say it with me now, "agency."
Agency: A Blessing, a Curse, and A Necessity
Conversational AI resources like ChatGPT, Claude, and Gemini are fast becoming commodities. They are easy to sign up for and quick to provide results to their end users. Each of the major providers reports hundreds of millions of active users every month. Even a niche player, Replika, reported more than 60 million active users in August. Millions of people have a high comfort level using conversational AI at work, school, and at home.
Even though simple-sounding intents can be complicated to carry out, conversational AI agents carry them out with increasing frequency, efficiency, and proficiency. Back in December, I noted that companies and their solution providers need to get their ducks in a row with policies, procedures, and workflows to take on the growing number of calls they will be getting from bots. (See: https://www.smartcustomerservice.com/Columns/Expert-Advice/Companies-Need-to-Set-AI-Agent-Interaction-Policies-and-Procedures-167164.aspx).
Some progress is being made. At this year's Enterprise Connect there was a lot of talk from contact center solutions providers surrounding machine customers. There is growing recognition among the likes of Five9, Genesys, and NICE that the latest renditions of Amazon's Alexa, Google Gemini, and (perhaps in a couple of years) Apple's Siri are poised to become machine customers. For years they have been working, primarily with Google and OpenAI, to incorporate generative AI into automated agents and automated assistants to live representatives. They are aware of but have not yet addressed contingency plans when interactions with autonomous agents go sideways.
Dan Miller is founder of Opus Research.