If your data engineering team spends more time maintaining pipelines than building new products, you’re not alone. Industry surveys consistently show that 40 to 60 percent of data engineering effort goes into integration work — connecting systems, transforming formats, and fixing broken connections when upstream sources change.
A new architectural pattern called “declarative data services” aims to change that equation. Instead of writing custom code to connect each data source, teams describe what data they need and let AI agents figure out how to get it. The approach is gaining traction in research circles and starting to appear in vendor roadmaps.
What Declarative Data Services Actually Means
Traditional data integration is imperative — engineers write specific instructions for each connection. Pull this field from Salesforce, transform it this way, load it into Snowflake at this time. When anything changes, someone rewrites the code.
Declarative data services flip this model. Teams define the outcome they want — “I need daily customer revenue by region” — and intelligent agents handle discovery, connection, and composition. The agents understand data schemas (the structure of how data is organized) and can automatically find relevant sources across an organization’s systems.
Think of it as the difference between giving someone turn-by-turn directions versus telling them the destination and letting GPS figure out the route. The destination stays the same even when roads close or traffic patterns shift.
Why This Is Gaining Momentum Now
Three factors are converging. First, large language models have become good enough at understanding data structures and business context to make automatic discovery practical. Second, the explosion of SaaS tools has made integration complexity unmanageable — the average mid-size company now uses over 100 SaaS applications, each with its own data format and API.
Third, the economics of data teams are hitting a wall. Hiring data engineers in India has become competitive, with senior talent commanding salaries that would have seemed absurd five years ago. Companies need to get more output from smaller teams.
Major cloud data platforms including Databricks and Snowflake have been adding AI capabilities that hint at this direction. Startups in the data orchestration space are moving faster — though most aren’t yet using the “declarative data services” terminology explicitly.
The Governance Problem Nobody Wants to Talk About
Here’s where excitement meets reality. When agents automatically discover and compose data sources, you lose visibility into exactly what’s happening. Which systems did the agent query? What transformations did it apply? Did it access data it shouldn’t have?
For companies in regulated industries — banking, healthcare, insurance — this creates serious compliance questions. Even for others, it raises basic quality control issues. If an agent decides to pull customer data from an outdated CRM sync instead of the authoritative source, your revenue reports will be wrong in ways that are hard to debug.
The research community acknowledges this. Recent work on declarative data services explicitly calls out the need for new observability tooling — systems that track what agents do and why. But these tools don’t exist yet at enterprise scale. Early adopters will need to build their own monitoring layers.
What Vendors Are Actually Shipping
Today, no major vendor offers a complete declarative data services platform. What you’ll find instead are partial implementations. Some tools offer natural language interfaces for querying data warehouses. Others provide AI-assisted schema mapping for integration projects.
The pattern to watch for: vendors adding “agent” capabilities to existing data catalogs and integration platforms. When your data catalog can not only describe what data exists but also autonomously connect and deliver it, you’re seeing declarative data services in practice.
Informatica, Talend, and the major cloud providers all have AI initiatives that could evolve in this direction. Smaller players like Airbyte and Fivetran are also adding intelligence to their connectors. The question isn’t whether this shift happens, but which vendors get the implementation right.
What This Means For You
Don’t reorganize your data team tomorrow. But do start a conversation with your vendors about their agentic data roadmaps. Ask specifically about observability and governance — any vendor that can’t answer those questions isn’t ready for enterprise deployment.
For new projects, consider whether declarative approaches might work in low-risk areas first. Internal analytics dashboards are safer experiments than customer-facing data products.
Budget for governance tooling. If you plan to adopt agentic data services in the next 18 months, assume you’ll spend 20 to 30 percent of that investment on monitoring, testing, and access control systems that don’t exist in your stack today.
The integration tax your team pays today isn’t sustainable. Declarative data services offer a genuine path to reducing it. But like any architectural shift, the benefits come with new risks that need managing from day one.