Agents, and Architecture, and Access - oh my!

Who is going to figure out how all this works together?

Couple recent articles have me thinking about the role of Chief AI Officer - although I also think that’s a transitional role that will either get rolled up into the CIO or CTO. Anyway, the first article, From selling access to selling work (and what it means for you) by Kyle Poyar, really clarifies some thoughts I’d been rolling around in my head. He lays out the shift from “access to software and toward a model of charging for the work delivered.” I’m trying to think of it as you wouldn’t pay for access to MSFT Word but pay by the documents created. This makes sense for AI systems since the more requests are made to the system, the more computer is being used and the more complex, multi-service AIs multiply that. The second article, Why Agent Orchestration Is The New Enterprise Integration Backbone For The AI Era, by Janakiram MSV, lays out the case for thinking of agent integration in a way that’s different than traditional IT infrastructures. What’s really interesting is that both of these aren’t just incremental shifts - one is a challenge to the entire SaaS business model and the other just seeks to redefine enterprise integration. That’s all.

Here’s a scenario - You have an AI agent, say its a chatbot, its job is to talk to customers on your website. Now the customer can ask it questions that could fall into a sales category or maybe a customer service category. Now there is another agent just behind that one, that, while the first one is answering the question to the best of its ability, is deciding where to send that query from the customer. The 2nd agent decides the question has implications for Marketing so it routes it there but there is also an element of training so it also routes it to the L&D team. Now those messages don’t end up in someone’s inbox, not yet anyway. The Marketing agent (that’s our 3rd one now), picks up the info and working from a pre-built template, does market and competitor research on just that issue and then, and only then, forwards the work of that research to a Marketing director. That director looks at the AI-generated work, selects a content option, makes some adjustments and then clicks send to share that with the agent that will handle updating the Web site, the print materials and adding that copy into the Salesforce instance so that sales people in the field can now make use of it.

The L&D agent picks up the data, compares it to all content in the existing content library (that includes text, video, audio, etc) and it finds a large chunk of text that really dives deep on the question and has an excellent rating for quality of answer from live instructors. It also checks content usage metrics and sees that the podcast version of answers currently has the highest CSAT score from external users. So the L&D agent pings another agent to take the text content and convert it to a podcast, add its to the content library, and after a quick check by a human, uploads the podcast to the external-facing content library under the “What’s New” banner. From one customer query then, we’ve generated an answer to the original question, updated marketing materials and sales enablement content, created a new podcast and the only time-limiting factor would be how available humans are to review and approve the content.

Now all those agents cost, and I’m guessing it’ll take some fairly serious financial modeling to figure out how much it can cost and then put triggers and alarms in place when it looks to exceed that estimate. It also means that we a backend architecture not to mention data hygiene program, that will allow these interactions to take place and will work to give the most accurate, least hallucinatory answers. Back to our Chief AI Officer then. Someone will need the perspective and authority to design these systems and processes and the strategy that will guide the adjustments and pivots that will inevitably follow. Sounds like a peer of the CFO, the CTO/CIO, and the CISO. Talk about jobs that AI won’t take away - that one feels like it would be secure for a good decade at least.