March 16, 2026

Two out of three agent deployments are failing. PwC data cited by Ekfrazo shows that 79% of enterprises are already running AI agents, but only 34% of those deployments reach production. The remaining 66% stall somewhere between pilot and scale. Ekfrazo is blunt about why. "The deployment window for competitive advantage is not opening; it's already closing for organizations still in strategy phases." Gartner, as cited in the same Ekfrazo analysis, projects that 40% of enterprise applications will embed task-specific agents by the end of 2026. The ambition is not the problem. Infrastructure, governance, and auditability resolved before deployment rather than after separate the 34% that ship from the majority that do not.

The stalling pattern is visible across both B2C and B2B, but the causes diverge. In retail, the bottleneck is consumer readiness and data quality. Invisible Technologies published a comprehensive breakdown of agentic commerce architecture this month that draws a hard line between chatbots and agents. Chatbots handle conversation and basic routing. Agents are goal-oriented systems that autonomously research, compare, and complete purchases across platforms without human intervention. Invisible Technologies warns that retailers who do not optimize their APIs, metadata, and checkout flows for agents risk being invisible to ChatGPT, Gemini, and similar assistants entirely. The homepage is no longer the front door. Backend systems that agents can query through APIs now determine visibility.

Mastroke makes the Shopify case for what "agent-ready" means in practice. In a video on the rise of agentic commerce, the firm argues that the primary entry point for many shoppers is already a chat interface in ChatGPT or Gemini, not a landing page. The Universal Commerce Protocol becomes a visibility gate. If an AI agent cannot read a merchant's inventory, pricing, and catalog metadata through UCP, the merchant is effectively invisible in AI-mediated conversations. Traditional optimization tactics like landing pages, pop-ups, and retargeting flows get bypassed when agents transact via APIs and protocols. Clean, structured product data now beats large ad budgets because agents do not see banners. They see data schemas.

Carbon6, reporting from NRF 2026, confirms the scale of the operational shift underway at the largest retailers. Walmart is deploying AI agents across shopping and operations through its Google Gemini partnership while expanding AI tools in Walmart Connect, its retail media arm. Carbon6 frames this as a direct challenge to Amazon as assistants become the next battleground for e-commerce and advertising. But the same report surfaces an underappreciated nuance. DICK'S Sporting Goods is building "House of Sport" locations that prioritize activities, coaching, and community engagement over shelf density. AI handles staffing optimization, inventory planning, and personalized recommendations behind the scenes. The experiential layer remains human. Not every category collapses into an agent-mediated transaction. Physical retail is being reimagined as social infrastructure, with agents optimizing operations rather than replacing the experience.

Where agents have delivered the clearest ROI is in enterprise operations. Ekfrazo documents 4-to-7x conversion rate improvements in agentic Salesforce deployments versus manual sales operations, citing Landbase data from January 2026. The same Ekfrazo analysis reports that AI agents processing invoice matching, purchase order approvals, and vendor onboarding show cost reductions of up to 70%. The numbers come from production deployments, not pilots or vendor projections. Leverage AI's survey of procurement automation platforms for industrial manufacturers confirms that most procurement executives now use generative AI weekly and name it as a top strategic priority. Procurement AI has moved from reactive task automation to proactive orchestration, compressing cycle times and improving on-time-in-full outcomes. The gating factor is no longer proving value. It is scaling without losing governance.

Keith Kirkpatrick at The Futurum Group frames the vendor dependency that makes scaling difficult. Agentic AI is shifting from simple assistance to complex workflow orchestration, embedding agents inside platforms rather than layering them on top. Whether growth materializes depends on SaaS vendors building more complex, policy-aware agents and helping customers move from pilots to production quickly. Human-in-the-loop checkpoints remain necessary for complex agents. The pattern emerging in both retail and enterprise is controlled autonomy, where agents act within policy boundaries and escalate exceptions to humans. Getting those boundaries right is the governance challenge that stalls two-thirds of deployments.

The Agentic Commerce Summit in London on April 13-15 will test whether the industry can close this gap. Banks, card networks, and retailers are convening behind closed doors to negotiate agent payment protocols, multi-agent orchestration standards, and governance frameworks. The timing matters. Defaults set in the next 90 days will likely persist for years, and two-thirds of current deployments are stalling on exactly the infrastructure and governance questions this summit intends to resolve. The execution gap is not a technology problem. It is an organizational one. And organizations that cannot ship agent deployments past the pilot stage will not get a second window to compete on terms they helped define.