The software industry has spent two years selling enterprises on AI agents. Now it is discovering that agents are only as good as the data pipelines feeding them, and those pipelines are where most AI budgets quietly go to die. 

[For more news, click here]

For all the money enterprises have poured into artificial intelligence over the past two years, the technology’s most persistent obstacle has turned out to be one of the least glamorous parts of the stack: data engineering. Executives can buy a large language model in an afternoon. Building the pipelines, quality checks, and governance structures that make that model trustworthy inside a real business can take a year or more, and it is where most AI initiatives quietly stall. 

That gap between ambition and infrastructure is the backdrop for Qlik’s announcement that its agentic data engineering capabilities are now generally available across Qlik Cloud. The release, which moves tools first previewed at Qlik Connect 2026 into production, gives data teams a set of purpose-built AI agents designed to find trusted data, define what it means in business terms, check its quality, and package it into governed products that can be reused across analytics, automation, and AI systems. Qlik, based in Philadelphia and used by roughly three quarters of the Fortune 500, is aiming the release squarely at the large enterprises where that infrastructure gap tends to be widest. 

It is a narrower and more technical announcement than the flashier consumer-facing AI news of the moment, but it points to a shift that matters more for how enterprises function. The industry’s early agentic AI wave focused on interfaces: chatbots that could answer questions, copilots that could draft code. Qlik’s release, along with similar moves from data platform rivals over the past year, reflects a second wave aimed at the unglamorous machinery behind those interfaces, the pipelines, catalogs, and stewardship processes that determine whether an AI system’s output can really be trusted. 

The bottleneck nobody budgeted for 

Qlik’s own research, released earlier this year, found that while 97 percent of enterprises have committed budget to agentic AI, only 18 percent have fully deployed it, with data quality, integration, and governance cited as the leading obstacles. Deloitte’s 2026 State of AI in the Enterprise survey found a similar pattern. Confidence in AI strategy is rising, but confidence in the underlying infrastructure, data management, and governance needed to support it is not keeping pace. Only one in five companies, Deloitte found, has a mature governance model for the autonomous agents they are increasingly deploying. 

The reason is structural, not a matter of vendors moving too slowly. Generative and agentic AI systems can draft code, recommend architectures, and generate entire pipelines in seconds, far faster than human engineers can review, govern, or maintain that output. That speed differential creates a new kind of bottleneck: not a shortage of ideas about what AI should do, but a shortage of trusted, well-documented, quality-checked data to point it at. Analysts have described 2026 as the year the metadata layer, rather than the model layer, became the real battleground for enterprise AI leadership, because without consistent definitions, lineage, and access controls, an AI agent’s confident-sounding output is only as reliable as the mess of legacy systems underneath it. Enterprises are, in effect, accumulating data debt faster than they can pay it down, and every new agent bolted onto the stack raises the interest.

Agents for the plumbing, not just the interface 

Qlik’s answer is to embed specialized agents directly into the data engineering workflow rather than layering a general-purpose assistant on top of it. The release includes a data quality agent that surfaces trust scores and anomaly detection, a data products capability for packaging curated datasets for reuse, a catalog glossary meant to standardize business terminology so both humans and AI systems interpret data consistently, and declarative pipeline tools that let engineers describe what they want in natural language while working inside development environments and coding agents they already use. Underpinning all of it is an expanded set of Model Context Protocol, or MCP, integrations, the increasingly common open standard that lets AI assistants access enterprise systems and data with governance and permissions intact. 

“Organizations are using many AI tools, it isn’t just one assistant or one model or a single data platform,” said Drew Clarke, Executive Vice President, Product and Technology, Qlik. “Our approach is to bring governed Qlik context into the tools data teams already use, so they can accelerate engineering work with agents while preserving choice, transparency, and control.” 

What it looks like in practice 

For Valpak, the UK-based packaging compliance business owned by Reconomy Group that manages regulatory data for more than a hundred enterprise customers across the UK, Europe, and the US, the appeal is less about speed for its own sake and more about not sacrificing control to get it.

“Qlik’s agentic data engineering capabilities will help us find the right assets, understand quality, and move trusted data products into use faster, while keeping our governance process intact,” said Robin Astle, Head of Qlik Analytics at Valpak. “That balance of speed and control is what will make AI practical for us.” 

Industry analysts see the same tension defining the next phase of enterprise AI adoption broadly.

“Enterprises are under pressure to operationalize AI faster, but many are discovering that data engineering and governance remain major bottlenecks,” said Stephen Catanzano, Principal Analyst, Data & AI, at Omdia. “What’s notable about Qlik’s approach is the focus on embedding agentic capabilities directly into governed data workflows, helping organizations accelerate delivery of AI-ready data products without separating speed from oversight.” 

Whether that balance holds at scale is the open question facing every vendor making similar bets this year. Agentic tools that write and revise pipelines faster than humans can review them raise a governance problem of their own, which is: who is accountable when an autonomous agent quietly changes a data definition that downstream systems depend on. Qlik’s answer, keeping humans in the loop for the decisions that matter while automating the repetitive work around them, is also the answer most of its competitors are converging on, which suggests the industry has not yet solved the problem so much as agreed on where the fault line sits. 

The capabilities are generally available now in Qlik Talend Cloud and Qlik Cloud Analytics, with availability varying by region, entitlement, and deployment configuration. The release follows Qlik’s introduction of its Predict and Automate agents in June, with an Analytics Agent planned for the third quarter of 2026, part of a broader push across the data and analytics industry to treat governance not as a compliance afterthought but as the infrastructure AI itself depends on.