This was the week the enterprise AI conversation stopped being aspirational and got specific about what it actually takes to run a modern, customer-facing operation when intelligence is woven into the underlying systems.
Salesforce’s Nitin Mangtani reframes on-site search as a continuous conversation rather than a one-shot query, turning the dead-end “no results” page into a discovery engine. From PegaWorld, enGen’s Richard Rutkowski walks through what it really means to move agentic AI from slide decks into production inside a highly regulated healthcare environment, while Unum’s Shelia Anderson tells the unglamorous but essential story of modernizing COBOL-era infrastructure so an enterprise can become genuinely AI-ready. Twilio’s Vanessa Thompson connects the data layer to the front line, arguing the traditional funnel obscures intent and that a unified customer view is what makes intelligent engagement possible.
And in the bonus pick, The Artificial Intelligence Show grapples with the economic and governance reality lurking beneath all that ambition — token costs, model availability, and the human-versus-machine balance every leader now has to price out.
Salesforce’s Nitin Mangtani on how AI is evolving on-site search
Nitin Mangtani, GM and EVP of Agentforce Commerce at Salesforce, makes the case that on-site search is undergoing a step-function shift on the order of the jump from web search to ChatGPT, moving from rigid keyword matching to a conversational discovery experience that anticipates intent. He frames the old model as “fire and forget” — a shopper types two words, gets a result, and the session resets — versus a continuous in-store-style dialogue where a customer can keep refining (“show me something dressier,” “anything in blue,” “in the $100 to $200 range”) without starting over. Drawing on Salesforce’s acquisition of Simulate, Mangtani illustrates the leap with a “Taylor Swift wine” query that would historically return zero results but, with advanced semantics, can infer the shopper actually wants a New Zealand Sauvignon Blanc. He ties the payoff to hard commerce metrics — conversion, average order value, and notably the reduction of apparel return rates that now run as high as 18 to 20 percent — and extends the same agentic interface to store associates and to syndicated channels like OpenAI and Gemini, where referral traffic is already showing up.
From PegaWorld: enGen’s Richard Rutkowski on moving agentic AI from theoretical to practical
Recorded at PegaWorld 2026, this conversation features Richard Rutkowski, Director of Product and Technology at enGen, the technology and services subsidiary of Highmark Health, who defines agentic AI as the system stepping into the determination a human used to make rather than simply surfacing analytics for a person to act on. His argument for why healthcare has no choice but to adopt it comes down to scale — aging populations, rising chronic illness, clinician burnout, and tightening CMS service-level requirements that traditional, fragmented care management simply can’t keep pace with. Rutkowski breaks down the architectural “chassis” built on Pega using a concrete discharge-event example: an agent reads the EMR signal, gauges the criticality, updates the authorization in one system, opens and schedules a case with a skilled case manager in another, and nudges the member through their preferred digital channel. He’s candid that most of this keeps a human in or on the loop, describing a “shift left” approach to utilization management where agentic determinations run in parallel to a medical director until trust and accuracy are proven, all underpinned by governance, auditing, and traceability.
Twilio’s Vanessa Thompson on moving to a unified view of your customers
Vanessa Thompson, VP of Revenue & Growth Marketing at Twilio, opens with a provocation — the traditional marketing funnel is “lying to you” — and argues that linear, click-to-conversion thinking tells you what happened without revealing what a customer actually needs in the moment. Her fix is to prioritize behavioral signals over funnel stages and to stitch first- and third-party data together in near-real time so teams stop guessing. She grounds it in a vivid example from Twilio’s product-led growth motion, where signup drop-off is treated like cart abandonment: the platform captures the event, links anonymous activity to a known account, and within fifteen minutes triggers a personalized email nudge alongside a synced retargeting audience — an orchestration that has driven roughly 11,500 additional signups, a 54 percent lift in spenders, and more than $4 million in closed-won business. Thompson also details Isa, Twilio’s email-based AI agent that holds full conversations to help users succeed rather than just sell, connecting to the unified customer profile, making users 3x more likely to upgrade, retiring the no-reply email, and lifting digital sales productivity by 200 percent while freeing teams for more strategic, curiosity-driven work.
From PegaWorld: Unum CIO & CDO Shelia Anderson on AI-augmented enterprise modernization
Also from PegaWorld 2026, Shelia Anderson, EVP and Chief Information & Digital Officer at Unum, shares the story behind her “COBOL Meets Cloud” keynote and the work of modernizing a 175-year-old benefits and insurance company without halting the business mid-flight — an effort she likens to changing the interior of a plane you still have to land. She emphasizes that real modernization starts with business-aligned value creation rather than technology, prioritizing the claims experience first, aligning investment to value streams, and measuring genuine business outcomes like customer satisfaction and average handle time instead of internal IT metrics. Anderson splits “AI-ready” into two deliberate tracks: everyday AI for the broad workforce, supported by training curriculums, Copilot, and governed citizen development, and customer-embedded AI requiring deeper engineering skill, including a goal that 100 percent of her engineering staff be trained to use AI in their daily work. She’s frank about adoption friction — early “build it and they will come” training fell flat, prompting an AI champion and advocacy program — and details Pega’s specific role using AWS Transform to interrogate 1.5 million lines of COBOL and Pega Blueprint to reimagine the resulting workflows, closing with advice to simply get moving before you’re certain you can swim.
Bonus Pick: [The AI Show Episode 221]: Anthropic vs. the White House, Microsoft CEO on the Future of Firms & AI’s Token Crisis — The Artificial Intelligence Show
Hosts Paul Roetzer, founder and CEO of SmarterX and the Marketing AI Institute, and Mike Kaput, SmarterX’s Chief Content Officer, spend this episode on the messy economic and governance realities sitting underneath every enterprise AI ambition. They unpack the still-unresolved export controls keeping Anthropic’s Fable 5 and Mythos 5 offline and the administration’s effectively impossible demand that model guardrails be guaranteed unbreakable, then turn to Satya Nadella’s widely circulated “future of the firm” essay and its framing of human capital versus token capital. The centerpiece for marketers and operators is an extended, refreshingly honest breakdown of the AI pricing problem no one has solved — why token-based billing, prompt caching, pooled versus per-seat usage, and agentic workloads make costs so hard to predict that planning multi-year AI adoption becomes genuinely difficult. It pairs naturally with this week’s Agile Brand conversations: where Mangtani, Rutkowski, Thompson, and Anderson describe what AI makes possible operationally, Roetzer and Kaput pressure-test what it costs and who controls it.
Taken together, these five conversations sketch the full arc of operationalizing AI: rethinking a customer interaction as a continuous dialogue, building the agentic and data architecture to support it, modernizing the legacy systems underneath, and then reckoning honestly with the economics and governance that decide what’s actually sustainable. The throughline is that ambition is no longer the hard part — disciplined prioritization, measurement, human-in-the-loop trust, and cost predictability are what separate the demos from the deployments. See you next week!
