Bring Humans Back Into Agentic AI Discussions

Too many genAI stories lack a human component. It feels a bit like Kevin Costner’s 1995 movie Field of Dreams. “Build it, and they will come.” First, just because you can do something doesn't mean you should. An HR professional on a panel yesterday suggested monitoring everyone's Slack conversations 24x7 to identify potential defectors or offer constant feedback. (#silentscream) Second, what is tedious for one person might be enjoyable to another. Listening to a handful of Silicon Valley folks discuss what they find inconvenient in life (i.e., the tasks they’d like agentive AI systems to automate for them) can feel a bit tone deaf. 

Announcements early this year have focused on using more assistive or agentive services to enhance or replace the web browsing experience (e.g., Anthropic, Brave’s Leo, Google’s Project Mariner, OpenAI’s Operator, Opera’s Aria, and more). One to two years ago the frenzy was around personal AI devices (e.g., Humane’s AI pin, Rabbit’s r1) automating similar services through a voice interface. 

Keep in mind:

• Consumers can access digital services through hundreds of touchpoints. A touchpoint is any combination of a device (e.g., smartphone, smartwatch), platform (e.g., Android, iOS, Instagram), channel (e.g., browser, app, message), and interface (e.g., touch, voice, XR). Combinations number into the hundreds. Consumers regularly use dozens of them. Conversational interfaces are one branch of interfaces and platforms. In fact while many consumers feel comfortable texting friends or fetching answers in these interfaces, they don’t feel comfortable transacting in them.

• Conversational interfaces won’t be the most convenient way to do many things. Ever remember your parents saying, “Forget it. I’ll just do it myself.” Sometimes it is just too hard to explain what you want someone else to do. Moreover, we all have different expectations of what a good job is. More often than not, I  run out of patience with these tools (e.g., image generators) that do 90% of what I need, but getting to 100% seems impossible. Moreover, processes for setting up permissions, links, etc. to enable agentive services have friction today as well as limitations based on their access.

• Consumers will choose touchpoints based on their needs. Consumers often want convenience, but not always. Consumers want convenience when they need to complete tasks (e.g., order cat food) or hit goals (e.g., track calories). Convenience is best delivered through invisible experiences i.e., ones that minimize consumer engagement. Consumers will want more immersive experiences (i.e., interactive, extended reality) when they seek entertainment or education. Keep in mind grocery shopping may be a chore for some and entertainment for others.

• Assistants and agents depend on context or memory - both short and long term. Compare two processes for ordering a pizza. In one process, you ask your spouse to pick up your favorite pizzas. Your spouse knows what vendor and what pizza type. In another process you go online and navigate several pages of pull down menus - unless you are ordering from Domino’s. The notion of context or memory demands either observing a consumer’s behavior (e.g., see, touch, smell,  or ingesting their preferences over time. Few if any entities outside of Apple, Google, or Meta have holistic profiles of their users.

• Context or memory depends on consumer trust. Trust is earned over time and often demands explicit value exchanges. Think about the last time you downloaded a mobile app that immediately asked you for permission to use your location or send you notifications. If you understood the circumstances or value  (e.g., we’ll let you know when your pizza is ready for pick up), you likely agreed. If not, you likely didn’t. Brands build trust over time as they deliver value. Using progressive disclosure tactics (i.e., asking consumers to share a little bit more over time) will deliver best results if executed well. 

• Websites are more than thousands of linked pages with tedious navigation schemes. I do not believe that websites are a core touchpoint in our future digital experiences, but they are a component. Websites give brands and consumers personalized and interactive ways to engage with one another in ways that third party platforms do not. They can help consumers make product decisions or get support. They also offer a long tail of content that apps and other interfaces are less likely to support. 

I appreciate the capability of the technologies that automate web-browsing. I have yet to see a truly compelling use case for consumers that doesn't require significant set-up. Consumers do want more invisible or automated experiences when they don't want to engage with brands. The ultimate invisible experience is when a brand (entity) can anticipate the need of a customer and serve them proactively.