The cover says it in one image, so I won’t describe what you can already see. Your customer meets an agent at the table; everything that decides how that conversation ends sits below the waterline. This article dives straight down — layer by layer, with the numbers, and with the companies already living it.

First, why now. The world will spend $2.52 trillion on AI in 2026 — a 44% jump in a single year, according to Gartner. Yet Forrester predicts three in ten firms will harm their customer experience this same year by deploying AI self-service prematurely, in contexts where it cannot succeed. Record investment; eroding experiences. The gap between the two is not a model problem, a vendor problem, or a talent problem. It is an infrastructure problem.

1. The Surface Illusion: Record Spending, Eroding Experiences

Gartner’s forecast does not just report a spending record; it places AI squarely in the Trough of Disillusionment for 2026 and argues that predictable returns — not enthusiasm — are now the gate to enterprise scale. Forrester completes the picture from the customer’s side: organizations pressured to cut costs will push generative chatbots and virtual agents into service before the foundations exist, and their Total Experience Scores will stay flat or decline for it.

Read those findings together and the conclusion is uncomfortable for every executive who has approved an AI line item: money is not the constraint, and it never was. Within its $2.52 trillion forecast, Gartner counts $401 billion flowing into AI infrastructure in 2026 alone. Companies are simultaneously outspending every technology cycle in history and disappointing customers with the result. What separates the two outcomes is the submerged stack. Every failed chatbot, every wrong recommendation, every agent that confidently invents an answer traces down the iceberg, not up the org chart.

2. The Connective Tissue: Where Agents Get Their Truth

Descend to the enterprise data and knowledge layers and you find the most decisive fact in AI-powered CX today: an agent is only as good as the data it can reach and trust.

MuleSoft’s 2026 Connectivity Benchmark, surveying more than 1,000 IT leaders, found that 88% of organizations are already on the path to partial or full agentic transformation — yet 50% of their AI agents operate in isolated silos, cut off from the systems where customer truth actually lives. The consequences are not subtle: 95% of organizations report integration challenges, 86% of IT leaders warn that unintegrated agents add more complexity than value, and 96% agree that agent success depends on seamless data integration. The result is a confidence crisis in plain numbers: 64% of IT leaders doubt their ability to meet their own near-term AI goals.

For customer experience, translate that gap into what it produces at the table above. An agent disconnected from entitlements data promises what the contract doesn’t cover. An agent blind to service history apologizes for the wrong problem. An agent without the knowledge layer — connected context and semantics — retrieves facts but misses the why behind them, which is precisely what makes an interaction feel understood rather than processed. This is not artificial intelligence failing. It is artificial intelligence performing exactly as well as its access allows — confidently uninformed, at scale. Readers who followed my argument in From Systems to Signals will recognize this as the destination that road was always pointing toward.

3. The Capital Layer: Someone Is Buying Your Ceiling

IDC’s tracker projects AI infrastructure spending will reach $487 billion in 2026 — roughly 53% growth — and exceed $1 trillion by 2029. IDC recorded nearly $90 billion in the final quarter of 2025 alone, and its analysts’ reading is explicit: enterprises and hyperscalers are building not for today’s workloads but for AI architectures still being defined. Within Gartner’s forecast, AI foundations alone drive a 49% increase in AI-optimized server spending this year.

This is the physical substrate of every future customer interaction: it determines the latency customers feel as responsiveness or friction, the cost per interaction that decides which experiences are economically viable at all, and the capabilities your teams can reach — or watch competitors reach first. These purchase decisions are experience decisions wearing procurement clothing. Which cloud, which data platform, which AI platform — each choice quietly fixes the ceiling on what your customer experience can become for five years. In most enterprises, the people accountable for experience outcomes have no line of sight into those decisions until they surface as constraints. By then, the concrete has set.

4. The Trust Layers: Governance, Control, and Access

The bottom of the iceberg — governance, the AI control tower, security and access — is where theory has already turned into corporate reality, so let real B2B companies tell it.

The scale problem first: MuleSoft’s benchmark finds only 54% of organizations have a centralized governance framework for AI agents, and reporting on the same research notes that more than a quarter of enterprise APIs remain ungoverned while the average organization now runs 957 applications with only 27% integrated. That reporting includes a case every executive should memorize: Alcon, the global eye-care company, found itself with more than 900 agents built in silos in under a year — described internally as a security risk first and foremost. Its response is the template: a formal AI governance board, enforced human-in-the-loop controls for customer-facing agents, and standardized agent access through the API architecture it had invested in years earlier. Nine hundred agents in twelve months is not an anomaly; it is a preview of what happens everywhere agent creation outpaces governance.

The proactive version exists too. AstraZeneca, per Salesforce’s official announcement, is deploying agentic AI for customer engagement with healthcare professionals globally — and doing it by design: extending a composable architecture with an agent-orchestration fabric so that internal and external agents act in coordination across field engagement, commercial operations, brands, and regions. The difference between Alcon’s emergency and AstraZeneca’s architecture is not talent or budget. It is whether the trust layers were built before the agents arrived.

5. When the Buyer Is Also an Agent: Prepare for Machine (Agent) Purchasing

Everything above matters twice over, because increasingly the entity across the table will be an agent too — buying, not just serving. Gartner’s research on machine customers reports that CEOs expect up to 20% or more of company revenue from these non-human buyers by 2030, with at least 15 billion connected products already able to behave as customers. Forbes, drawing on Gartner’s modelling, sizes the stake at $30 trillion in machine-influenced purchases by 2030, with nine billion connected B2B products able to buy by 2028 — and over 50% of CEOs already planning a strategy for machines in the buying or selling process within two years.

Preparation is concrete, and it maps onto the same iceberg. First, make your commercial truth machine-readable: catalogue, pricing, availability, and terms exposed through governed APIs — a buying agent cannot purchase what it cannot parse, and every gap is a deal your competitor’s cleaner data wins. Second, convert trust into verifiable data: certified specifications, delivery performance records, consistent cross-channel pricing. Marketing language is invisible to an algorithm; auditability is persuasion. Third, govern your own selling agents as commercial policy: an agent designed to obscure fees or stall will be deselected by buying agents silently and permanently — no complaint ticket, no churn interview, just revenue that stops arriving. The companies that treat machine buyers as an infrastructure requirement now will own the earliest and stickiest agent-to-agent relationships in their industries.

6. The Iceberg on One Page

The Depth Test

Here is the conclusion this deserves, stated without hedging. The AI conversation in most boardrooms still starts with the wrong question — what can we automate? The right one is a depth test with two parts: Is our iceberg strong enough for the agents we deploy — and attractive enough for the agents that will buy from us? Alcon’s 900 silent agents answer the first part; Gartner’s 20%-of-revenue projection answers the second. By 2030, one euro in five may arrive from a buyer that never saw your logo, never felt your brand, and evaluated nothing but the nine-tenths of your company that customers never see.

Experience-led growth was always earned; the agentic era just moved the mine. The surface is now won in the depths — funded there, integrated there, governed there, and lost there. The companies that build below the waterline this year will be the ones still visible above it in five.

👉 Stay ahead of CX, AI, and innovation — Subscribe to my weekly LinkedIn Newsletter “CX Insights by Ricardo S. Gulko.

Share it if it’s useful — and let’s connect for more. Ricardo Saltz Gulko

My columns in several respected CX publications.

• My recent articles on Eglobalis: https://www.eglobalis.com/blog/
• My recent articles on the European CX Organisations ECXO.org : https://ecxo.org/blog/
• My recent articles on CMSWire: https://www.cmswire.com/author/ricardo-saltz-gulko/
• My German articles on CMM360: https://www.cmm360.ch/author/ricardo/
• My old articles on CustomerThink: https://customerthink.com/author/rgulko/

Primary Sources

Every statistic in this article was verified directly against the live pages below on July 21, 2026.

• Gartner — Worldwide AI Spending Will Total $2.5 Trillion in 2026 ($2.52 trillion, +44%; $401 billion AI infrastructure; Trough of Disillusionment; +49% AI-optimized servers)
• Forrester — Predictions 2026: Customer Experience (three in 10 firms will harm total-experience growth with frustrating AI self-service; flat or declining Total Experience Scores)
• MuleSoft — 2026 Connectivity Benchmark Report (1,000+ IT leaders; 88% agentic transformation; 50% of agents in silos; 95% integration challenges; 86% complexity warning; 96% integration dependency; 64% goal-confidence gap; 54% governance gap)
• IDC — AI Infrastructure Spending: 2029 to Eclipse $1 Trillion ($487 billion in 2026, ~53% growth; ~$90 billion in Q4 2025; $1 trillion+ by 2029)
• Eglobalis — From Systems to Signals: How Real-Time Data Are Rewriting B2B CX
• net — The 2026 Connectivity Benchmark: Agentic Enterprise Infrastructure (957 apps, 27% integrated; 25%+ ungoverned APIs; Alcon case: 900+ agents, governance board, human-in-the-loop)
• Salesforce — 2026 Connectivity Report Announcement (AstraZeneca: agentic customer engagement orchestrated across field and commercial operations)
• Gartner — When Machines Become Customers (CEOs expect up to 20%+ of revenue from machine customers by 2030; 15 billion connected products)
• Forbes — Machine Customers: AI Buyers to Control $30 Trillion in Purchases by 2030 (Gartner modelling: $30 trillion; nine billion connected B2B products by 2028; 50%+ of CEOs planning a strategy)