Brad Shimmin is Vice President Practice Lead, Data Intelligence, Analytics, & Infrastructure at Futurum, where he provides strategic direction and market analysis to help organizations maximize their investments in data and analytics. Currently, Brad is focused on helping companies establish an AI-first data strategy. With over 30 years of experience in enterprise IT and emerging technologies, Brad is a distinguished thought leader specializing in data, analytics, artificial intelligence, and enterprise software development. Consulting with Fortune 100 vendors, Brad specializes in industry thought leadership, worldwide market analysis, client development, and strategic advisory services. Brad earned his Bachelor of Arts from Utah State University, where he graduated Magna Cum Laude. Currently, Brad lives in Longmeadow, MA, with his beautiful wife and far too many LEGO sets.
Predictions
Data Control Planes will Shift Focus from Passive Insight Generation to Governed, Autonomous Execution of Business Actions
In the second half of 2026, the focus for data and AI leaders will shift from generating trusted insights to executing governed actions. We spent the first half of the year building semantic layers to help models understand business context. Now, organizations face a harder problem: letting agents act on that data in a safe, governed, and performant manner.
Why This Is Trending
The transition from generating insights to executing governed actions requires a significant investment in the underlying data estate, breaking down data silos, surfacing critical meaning through metadata, and overcoming architectural barriers that divide operational and analytical engines. However, success is currently constrained less by technology or budget than by a critical shortage of professionals capable of governing autonomous software and managing the technical debt of rapid agentic AI adoption. Amid these shifts, enterprises also face new risks of technical debt and inescapable inertia via vendor lock-in as providers attempt to control both the foundational database layer and the intelligence stored at the agent and ontology level.
Bridging the Read/Write Gap: Organizations have largely figured out the read-path for AI. Nearly 60% of enterprises are directing budget toward semantic layers, and standards like the Open Semantic Interchange (OSI) are helping ground models in shared business meaning, with MCP standardizing how agents reach that data. However, almost a quarter of organizations report that their agents’ inability to write back to systems of record is their main architectural bottleneck. If the data an agent reasons with and the data it alters live in different environments, the system stalls.
Governing Software like Human Workers: Letting software take action is a governance issue before it is a technical one. With 93% of organizations struggling to establish production governance, the emerging baseline is a “read informs, write requires approval” model. Agents are now distinct, trackable identities. We need the ability to authorize them, monitor their actions, and roll them back with a full audit trail if they make a mistake.
The Ceiling is Human, not Financial: The shortage of data professionals has climbed to 10.4% over the past year, overtaking budget constraints as the primary hurdle to scaling AI. This labor gap is the real forcing function behind the push for self-managing databases and simpler governance controls. The idea of the “AI Shepherd” is no longer just a forecasted job title; it is a hard limit on how much autonomy a company can safely deploy.
Vendor Impact
The Governed Write-Back Loop: Instead of an agent suggesting an answer for a human to manually type into an ERP, the platform handles the loop on a single substrate. An agent reviews the reference data, drafts a reallocation, and the database simulates the outcome using copy-on-write before committing. Once a human approves, it writes back using the agent’s identity, respecting existing access controls, and creates a reversible audit trail. Vendors who keep context and transactional data in separate silos, connected by pipelines, will struggle to compete with those that bring them together.
Agent Identity as a Baseline Requirement: Agents require their own identities, much like an employee badge. This allows platforms to manage specific permissions, track risk, and maintain audit trails. If an agent goes off track, the governance plane limits the damage and reverts the system to a known-good state. The feature that unlocks enterprise budget right now is the ability to discover and govern shadow agents across multiple clouds before they access and alter production data.
The “Refuse to Guess” Toggle: Basic, naive RAG is giving way to curated context engines that sit between agents and data. The most practical feature of these engines is a strict governance toggle: if a trusted definition is missing, the agent declines to answer rather than guessing. If asked for a metric without a governed definition, it returns a refusal rather than a confident fabrication. This approach increases accuracy and lowers token costs, as precise context requires less compute.
The Ontology-Portability Clause: Procurement teams are starting to catch on to the shell game being played among vendors. Expect the standard RFP question to move beyond whether a vendor supports open formats like Iceberg: buyers want to know if they can export their ontology, entity relationships, and agent memory in a reusable format. Vendors who offer open storage but lock down the intelligence layer will face heavy pushback from buyers who have learned this lesson the hard way.