AI in 2026 (so far): The Agents Are Here, but Are Businesses Ready?
2026 may be remembered as the year AI agents moved from impressive demos to real business interactions. The technology is advancing quickly; the business models, operating practices, and expectations around it are struggling to keep up.
In a recent BCStrategies podcast, five of our analysts and consultants, Tom Brannen, Blair Pleasant, Kevin Kieller, Chuck Vondra, and Steve Leaden, compared notes on what stood out most this year across customer experience, unified communications, enterprise strategy, and hands-on consulting. The common thread: AI's potential is enormous, but realizing it takes far more than deploying another agent.
Tom Brannen: When AI Agents Become the Customers
Personal AI agents are starting to research, book, buy, and request support on behalf of consumers. This is no longer theoretical: a major hotel brand is already fielding contact center calls initiated by AI agents. That raises questions traditional CX models were never built to answer. How do you authenticate an agent acting for a customer? Does it get the same service priority as a human caller? How do upselling, loyalty, and sentiment measurement work when the immediate participant is software?
The takeaway: businesses will need to design experiences for both human customers and the AI agents representing them.
Blair Pleasant: CX Is About Outcomes, Not Interactions
The focus is shifting from handling and routing interactions to the complete customer journey and relationship, and from answering questions to completing workflows and resolving problems. Blair pointed to Zoom's "conversation to completion" message and Cisco's emphasis on relationship management as examples. If AI agents are digital workers, they need to be supervised, governed, and measured like workers. And with personal agents on one side and business agents on the other, agent-to-agent interaction looks set to be one of the defining CX challenges of 2027.
The takeaway: the differentiator is moving from efficiently managing interactions to delivering meaningful outcomes across the whole customer relationship.
Kevin Kieller: AI Has Thrown Enterprise Pricing Into Disarray
Subscription licensing was predictable; consumption pricing, AI credits, and usage-based models are not, and customers often have no reliable way to forecast usage before they deploy. Kevin described the market moving from "token maxing" to "token axing" as organizations shift from encouraging experimentation to controlling costs. If an agent is positioned as a digital employee, what happens when its monthly allowance runs out mid-month? Which department pays when one agent serves several? Outcome-based pricing sounds appealing, but defining a successful outcome (was the issue really resolved if the customer calls back?) is far from simple.
The takeaway: AI pricing must become more predictable, transparent, and outcome-aligned before organizations can build critical workflows on it.
Chuck Vondra: The Problem Isn't Hype; It's Unrealistic Expectations
Many organizations are enthusiastic about AI but haven't defined what they want it to accomplish. Chuck described a credit union that bought an intelligent virtual assistant and expected it designed and live within two months, overlooking the substantial requirements, integration, and testing work involved. As Kevin put it during the discussion: "No one wants AI. They want the magic that they think AI brings." AI isn't necessarily overhyped; it is often misapplied.
The takeaway: successful adoption starts with the business problem and measurable success criteria, not with selecting the technology.
Steve Leaden: Test AI to Fail, Not Just to Succeed
In a healthcare deployment covering prescription refills and appointment confirmations, planned across 120 groups, pilot testing exposed long silences before transfers and repeated "do you want to be transferred?" prompts. Minor in a demo; damaging in a real patient interaction. The team recommended more testing rather than launching, including scenarios outside the scripted workflows, such as a patient describing a medication informally. The economics can still be compelling: one model compared roughly $1.2 million in annual staffing cost against about $200,000 in annual AI usage, though real results depend on resolution rates, implementation costs, and workforce decisions.
The takeaway: try to break your AI before your customers do. ROI depends on disciplined implementation and testing against real-world scenarios.
Five Lessons Heading Into 2027
Prepare for agent-to-agent interactions. Engagement models, security, authentication, and service policies will need to evolve.
Measure business outcomes, not AI activity. Completed workflows and resolved problems matter more than interaction counts.
Understand the economics before scaling. Consumption pricing introduces real budgeting risk.
Start with clearly defined, achievable use cases. Executive enthusiasm cannot substitute for requirements, realistic timelines, and readiness.
Test aggressively before deployment. A successful demo is not proof of production readiness, especially in high-stakes environments.
Looking ahead, the panel expects context and orchestration to get much more attention as agents multiply across applications and departments. The winners won't be the organizations deploying the most agents; they will be the ones that understand where AI creates value, manage its costs, and implement it in ways that reliably deliver better outcomes.
This post summarizes the original BCStrategies article. Read the full piece and join the discussion in the comments on LinkedIn.
