AI’s 2026 bottleneck will not be the model. It will be the data layer.

Confluent’s 2026 Predictions Report makes a sharp argument about the next phase of enterprise AI: the model race is no longer the whole story.

After a year where agentic AI dominated data conversations, Confluent says 2026 will be about solving the barriers that made AI hard to move into production. The report opens with a direct lesson: “all AI problems are data problems.” It also notes that 68% of IT leaders recently cited data silos as a major roadblock for AI success.

That framing matters because many companies treated 2025 as the year to experiment with AI agents. The next phase is less glamorous but more decisive: making sure those agents have the right data, context, governance, infrastructure and resilience to operate inside real businesses.

The rise of machine customers

One of Confluent’s strongest predictions is the rise of agentic commerce. The report argues that AI agents will increasingly act as customers, handling routine purchases and optimizing across price, quality and convenience.

This is not limited to consumer shopping. Confluent expects enterprises to see similar changes across supply chains and B2B relationships. The warning is blunt: machine customers will have no patience for latency, no brand loyalty and the ability to switch vendors mid-transaction when a better offer appears.

For businesses, that changes the definition of customer experience. It is no longer only about human-facing websites, brand campaigns or sales funnels. Companies will need agent-safe APIs, tokenized payment protocols and real-time product data so autonomous systems can search, decide and transact instantly.

MCP becomes the connective tissue for AI applications

Confluent also predicts that leading platforms will offer Model Context Protocol. The report describes MCP as an open source standard for connecting AI applications to external systems and says it gained rapid momentum in 2025.

The reason is practical. AI applications need to access the right tools and data without becoming trapped inside one model provider’s ecosystem. Confluent argues that MCP will become table stakes for platforms that want to participate seriously in the AI ecosystem, even as other standards such as Agent2Agent and Agent Communication Protocol compete in agent-to-agent communication.

The market signal is clear: enterprise AI needs portability. If companies cannot switch between models, tools and providers, their AI strategy becomes another lock-in problem.

Context engineering becomes the next AI discipline

Confluent’s third prediction is that 2026 will shift attention from retrieval-augmented generation and agentic AI toward context engineering.

The report defines the problem clearly. Even when companies can access the right data, they still need to decide what context enters the model, how much context is too much, how to avoid slowing the system down and how to ensure important information is not lost inside a growing context window.

Confluent’s larger point is that building with foundation models changes how software is improved. Traditional software is about iterating on code. Context engineering is about iterating on data, prompts, rules, feedback loops and live context.

That may become one of the most important enterprise AI operating skills in 2026.

Databases will feel the pressure from AI agents

The report also warns that agentic AI will increase pressure on existing databases. At first, agents may look like replacements for human tasks, but Confluent argues they will eventually perform work at scales humans would not attempt.

That means more queries, more real-time demands and more stress on operational systems of record. Confluent says enterprises should prioritize change data capture pipelines so operational changes can stream into real-time serving layers with minimal impact on core systems.

This is where the infrastructure argument becomes concrete. AI agents do not only need intelligence. They need fresh, accessible and reliable enterprise data.

Governance moves from compliance function to AI foundation

Another major prediction is increased investment in enterprise-wide data governance. Confluent says cross-system data lineage will become a top priority as organizations try to prove that the data feeding AI models is traceable and trustworthy.

The report notes that 84% of technical leaders recently cited data management and governance as a top-tier technology priority. It also highlights the core questions many organizations still struggle to answer: Where did the data come from? Can it be trusted? What data was used to make a decision?

AI raises the stakes because these questions no longer apply only to regulated sectors. Any AI system touching personal, financial or sensitive data will need stronger lineage, auditability and governance.

The data stack will need cold storage, neutral planes and durable execution

Confluent’s report also points to several less flashy but important infrastructure shifts.

Apache Iceberg is expected to become a standard for cost-effective cold data management, turning historical data into an “active archive” for auditing, security forensics, AI model training and long-term analysis.

The report also predicts that AI strategies will need an independent data plane to avoid overcommitting to a single LLM ecosystem. As model providers build broader platforms, Confluent warns that enterprise context and real-time operational data could become trapped by vendor-specific frameworks.

Durable execution engines are another theme. Confluent predicts that early adopters of durable execution engines such as Temporal and Restate will gain an edge as AI agents require more reliable multi-step workflows, retries, timers and state persistence across failures.

The real takeaway: data becomes the AI advantage

Confluent closes the report by arguing that 2026 will bring solutions to many of the barriers that made agentic AI difficult in 2025. It says businesses will need modern platforms and architectures that support streaming data, intelligent data flows and scalable AI use cases. The report also names platforms such as Confluent, Databricks and Snowflake, along with hyperscaler offerings, as part of the broader convergence around integration, processing, governance and storage.

The most important line is not about any single tool. It is the report’s conclusion that the opportunity lies in simplifying the stack so data, not tools, becomes the true driver of AI advantage.

That is the real 2026 enterprise AI story.

The winners will not be the companies with the most AI pilots. They will be the companies that can give AI systems clean data, live context, governed access, resilient execution and enough architectural freedom to avoid the next generation of lock-in.