When Conversation Is the Interface, the Dashboard Is an Artifact

Anthropic recently released in beta the ability to create dashboards. I am seeing conversations about Power BI being "cooked". Even if I agree that business intelligence UIs are "cooked", I think the discussion should not be about the visualization. My social feed has recently been full of clips of people building their own version of Jarvis, an agent they can speak to about their business, emails, whatever they have going on. Even though many of them want you to comment "Jarvis" to get a course link, the capability to talk to an agent is appealing.

If you attend a standup every morning, I would bet that, like me, you have sat in highway traffic and wanted a voice interface that gives you a synopsis of your area of responsibility and your inbox, with the ability to draft replies. Being able to have that functionality through voice with the capability to drill into anything or ask questions that require research for an answer is invaluable. If everyone had access to this type of agent, I would absolutely question the need for the standard standup. Conversational intelligence is the answer.

I define conversational intelligence as an agent you can talk to that reads across your sources of information, answers your questions, does the work you assign, and reaches out when something needs your attention.

1. If you need to know how infrastructure and batch jobs are doing, the data is there in logs that need correlation.
2. If project delivery is the question, the data is there in emails, messages, timelines, design artifacts and code branches.
3. If sales performance is the question, the information is there in funnel depth, emails, calls, meetings and letters of intent.

Our conversations can be about efficiency, improvements, strategic direction or moonshot attempts rather than current state. Anthropic dashboards can be one step toward conversational intelligence. To get the rest of the way, you solve access permissions, data ontology, data taxonomy, and data delivery to models. Those models can then answer questions like:

1. I have a theory about X, does the data support my theory?
2. Is there a better way to achieve Y based on what you see?
3. What insights do you have about how we do X, based on our data?

The agent will also need to be architected to handle untrusted content, know which actions need human approval, and verify identity by voice in both directions.

Agentic systems that can work on assigned tasks, produce artifacts on demand (dashboards, reports) and communicate with you on a channel you prefer (SMS, message, email, voice) are the future. Whether you build the conversation channel or integrate it into a properly designed architecture that addresses the points above, it is worth a pilot to embrace conversational systems that you can call or that can call you.