Declarative Data Services: Structured Agentic Discovery for Composing Data Systems

URL SCAN: Declarative Data Services: Structured Agentic Discovery for Composing Data Systems
FIRST LINE: Computer Science > Artificial Intelligence

THE DISSECTION

This is a systems-engineering paper from the AI/CS research community. It describes a framework called Declarative Data Services (DDS) — essentially a structured approach to getting AI agents to reliably compose multi-system data backends (e.g., trading systems) using typed contracts, bounded sub-searches, and inline skill citations with error routing.

The operative claim: unbounded agentic discovery (blind iteration) fails to converge on a working deployment stack, but structured discovery with typed knowledge channels does.

THE CORE FALLACY

The paper presents this as a systems design problem — finding the right architectural scaffolding to make agents reliable at a task that matters. This framing is internally coherent but politically/structurally blind.

The hidden assumption: The task itself is stable and worth doing. DDS makes AI-driven backend composition marginally more reliable. The paper assumes this is a good use of AI capability.

It is not. This is a perfect illustration of the DT's productive participation collapse.

THE ACTUAL SYSTEMIC FUNCTION

This paper is a technical validation of labor displacement mechanics dressed as systems research.

• Who builds and maintains data backends today? Skilled engineers: database administrators, backend developers, infrastructure architects, integration specialists.
• What does DDS do? Decomposes the search space so AI agents can own the discovery, composition, deployment, and debugging cycle — with "runtime failures becoming skill patches that the next deployment cites inline."
• The result: The entire stack composition lifecycle moves from human craft to AI execution. The human role collapses to writing declarative intent, not engineering the system.

The "bounded search" and "typed contracts" are not improvements to human tooling. They are automation primitives that encode the conditions under which human involvement becomes optional. The paper even admits the proof-of-life is "a trading-backend workload" — financial infrastructure. High-stakes. High-complexity. The domain where "it runs" is the verifier.

THE KILL MECHANISM (DT LOGIC)

Under the Discontinuity Thesis:

1. P1 (Cognitive Automation Dominance): DDS is a direct instantiation of durable AI cost/performance superiority in a cognitive domain: system composition, debugging, integration. The paper is literally documenting that AI can own this now.
2. P3 (Productive Participation Collapse): Backend engineering is not an edge case. It is infrastructure labor. If DDS converges reliably (even as a prototype), the labor category "data system architect/engineer" enters terminal obsolescence. The work doesn't disappear — it migrates to AI, and the human role shrinks to intent-setter.
3. The verification is the point: "Verifier is whether a deployed stack actually runs." This is a production-grade test, not a benchmark abstraction. When AI passes it consistently, the human credential barrier collapses.

LAG-WEIGHTED TIMELINE

• Mechanical Death: 3-7 years for composable AI deployment in backend engineering roles. The research path is clear; scaling from "trading backend prototype" to enterprise deployment is a matter of engineering investment.
• Social Death: 10-20 years, because legacy codebases, institutional inertia, and credentialed gatekeeping will slow adoption in regulated sectors. But the direction is fixed.

THE VERDICT

This paper is a proof-of-concept for infrastructure labor obsolescence. It is technically precise, well-argued within its own frame, and completely indifferent to the human displacement it describes. The "skill patches" framing is especially telling: AI failures are converted into citations for the next AI deployment — human feedback loops are not in the loop, they are the input signal that gets routed backward as typed errors. The human is a sensor, not an agent.

Classification: Technical acceleration, not policy paper. Function: Documenting the advance. No mitigation, no ethics, no acknowledgment of what this means for the humans currently employed in the tasks being automated.

The paper will be cited by AI infrastructure companies as evidence that AI can own complex system composition. It will be used to justify further investment in agentic pipelines that replace backend engineers. The academic framing creates distance from the human cost, which is exactly the ideological function this paper serves.