Operationalizing Data Plumbing for the Era of Autonomous Reason

For the last few decades, the people in charge (the engineers, the managers, and the bureaucrats) have been very busy building digital filing cabinets. They called these “solutions.” They filled them with precious, and strangely expensive, digital lint (data) and then they locked the cabinets and threw the keys into different oceans. We called this “The Enterprise” and it was a world where information went to die in a silo, and the only people who could find anything were the high priests of the mainframe.

Now, everyone is very excited about the new intelligences. The Engines of Reason. The Agents. We want these digital ghosts to run our companies, buy our groceries, and perhaps, if they have a spare moment, explain why we are all so tired. But there is a large, expensive, very human snag. You see, an Agent is only as smart as the junk you feed it. If you feed a brilliant AI a diet of fragmented, siloed, and expensive garbage, you get a very fast, very confident hallucination.

The problem isn’t the brains. We have plenty of brains. The problem is the plumbing. In a world where 40% of business-critical data sits in a silo, the “Engine of Reason” is effectively a second-rate machine trying to navigate a first-rate mess.

The Evolution of Enterprise Data Storage Paradigms

Now with 100% more "The Cloud is just a computer in Ohio."

Why Infrastructure Dictates Intelligence

In the old days, if a machine made a mistake, it just stopped. Now, when a model makes a mistake, it does so with the confidence of a man who has never been told he’s wrong. In this scenario, the agent is looking at the world through a keyhole and assuming the hallway is the whole universe.

The financial cost of this confidence is quite staggering. Poor-quality data is estimated to cost organizations an average of $12.9 million every single year. When these agents start making decisions without a “Unified Memory,” they start hallucinating at rates as high as 33% to 35% in complex tasks. They aren’t lying; they just don’t have the “plumbing” to reach the truth.

POV: You just asked a Data Scientist how their week is going.

The Great Data Partition and the Synthesis of the Lakehouse

Long ago (about three years ago) you had to choose. You could put your data in a “Lake,” which was cheap but became a “swamp” because nobody knew where the bodies were buried. Or you could put it in a “Warehouse,” which was tidy but so expensive it made the accountants weep.

The “Lakehouse” is a silly word, like “Spork,” but it is a vital one. It means the data stays where it is, in its natural, low-cost habitat, but we treat it with the respect and structure of a library. By using open formats like Apache Iceberg, we ensure that the AI never reads a partial or inconsistent truth. It brings ACID (Atomicity, Consistency, Isolation, Durability) transactions to the mess, which is a fancy way of saying the machine won’t forget what it was doing in the middle of a sentence.

What is Apache Iceberg? Apache Iceberg is an open-source, high-performance table format designed for massive analytical datasets. If a “Data Lake” is just a messy pile of files in a folder, Iceberg is the intelligent management layer that makes that pile act like a professional SQL database.

It was originally developed at Netflix to solve the “Small File Problem” and the performance bottlenecks of older systems like Apache Hive.

The Rise of the Non-Human Customer: Agentic Commerce

We are entering the era of “Agentic Commerce” and this is a fancy way of saying that machines are going to start talking to other machines to get things done. Your refrigerator will argue with a warehouse while your portfolio negotiates with a ledger.

For this to work, these agents need to be able to reach into the basement, the attic, and the cloud at the same time without needing a map and a flashlight. If you lock your data in a proprietary cage, you are essentially teaching your AI to speak a language that only one company understands. This is a fine way to run a monopoly, but a terrible way to run a civilization.

A tragic three-act play for everyone who spent their career on SEO and UI design.

Why Plumbers are the New Poets

We are currently obsessed with the “Models” those shimmering, mathematical pop-tarts. However, the real heroes of the next decade won’t be the ones building bigger brains. They will be the ones building better foundations. They are the “plumbers” of the digital world, and they have three main jobs:

• Open Standards: Because secrets are just technical debt in a trench coat. Standards like the Model Context Protocol (MCP) and Agent Communication Protocol (ACP) allow agents to form “teams” rather than just isolated islands of silicon.
• Hybridity: Your data should be able to live on a server in your closet or a server in the sky without throwing a tantrum.
• Governance: Because an agent without a sense of “Truth” is just a high-speed rumor mill.

Governance is an Accelerator

Industry research shows that 27% of AI efficiency gains stem directly from strong governance. Organizations that invest heavily in AI ethics report 34% higher operating profit from their AI investments than their less-governed peers.

Proof that being 'ethical' is actually just a really elaborate way to make 34% more money

Memory: The Nervous System of Intelligence

Large Language Models are, by nature, stateless and they have the memory of a goldfish. To make them useful, we have to build a “Unified Memory” layer or a shared cognitive infrastructure.

• Short-term Memory: The context window. Remembering what you said five seconds ago.
• Long-term Memory: The library. Implemented with vector embeddings or knowledge graphs to store facts and business rules across sessions.
• Procedural Memory: Learning skills, so the agent doesn’t have to relearn how to file a report every Monday morning.

Without a shared memory layer, Agent A will remain ignorant of what Agent B discovered. By implementing this shared layer, organizations reduce millions of tokens in unnecessary overhead, moving from O(n²) context reading to O(n) efficiency.

Toward a Future of Clean Plumbing and Real Trees

In 1962, a small device called “Shoebox” was shown at a World’s Fair. It was the world’s first speech-recognition system. Back then, we were promised that technology would eventually give us “a little peace and quiet.” Instead, we got 2,000 pages of reports every minute and a world where we are all “appendages to machines, institutions, and systems.”

The shift toward the “Agentic Enterprise” is projected to generate nearly $6 trillion in economic value by 2028. But that value only appears if we stop worrying about the “Magic” and start worrying about the “Ready.” If the data is ready, the AI will follow. And if the AI follows, maybe we can all go outside and look at a real tree for a change.

Keep your data open. Keep your engines fast. And for heaven’s sake, keep your plumbing clean.

Works cited

1. IBM watsonx.data Accelerates GenAI Data Analysis - Intel, accessed May 11, 2026, https://www.intel.com/content/www/us/en/customer-spotlight/stories/ibm-watsonx-data-customer-story.html
2. What Are Data Silos? | IBM, accessed May 11, 2026, https://www.ibm.com/think/topics/data-silos
3. Agentic AI, explained | MIT Sloan, accessed May 11, 2026, https://mitsloan.mit.edu/ideas-made-to-matter/agentic-ai-explained
4. What Is Apache Iceberg? | IBM, accessed May 11, 2026, https://www.ibm.com/think/topics/apache-iceberg
5. Making Data-Driven Decisions with IBM watsonx.data, an Open Data Lakehouse on AWS, accessed May 11, 2026, https://aws.amazon.com/blogs/apn/making-data-driven-decisions-with-ibm-watsonx-data-an-open-data-lakehouse-on-aws/
6. IBM/watsonx-data - GitHub, accessed May 11, 2026, https://github.com/IBM/watsonx-data
7. Common Data Quality Issues in 2026 (+ How to Fix Them) - Prospeo, accessed May 11, 2026, https://prospeo.io/s/common-data-quality-issues
8. What’s new in watsonx.data - IBM, accessed May 11, 2026, https://www.ibm.com/docs/en/watsonxdata/standard/2.1.x?topic=overview-whats-new-in-watsonxdata
9. Memory architecture is the real bottleneck in multi-agent AI, not prompt engineering - Reddit, accessed May 11, 2026, https://www.reddit.com/r/AI_Agents/comments/1r7e8jo/memory_architecture_is_the_real_bottleneck_in/
10. What Is AI Agent Memory? | IBM, accessed May 11, 2026, https://www.ibm.com/think/topics/ai-agent-memory
11. What is Agent Communication Protocol (ACP)? - IBM, accessed May 11, 2026, https://www.ibm.com/think/topics/agent-communication-protocol
12. Beyond hype: Accelerate innovation in a multi era hybrid cloud maze, accessed May 11, 2026, https://www.ibm.com/new/product-blog/beyond-hype-accelerate-innovation-in-a-multi-era-hybrid-cloud-maze
13. Agentic Commerce: How AI Agents Are Changing Ecommerce - SPXCommerce, accessed May 11, 2026, https://www.spxcommerce.com/blog/agentic-commerce-ai-agents-marketplaces/
14. Agentic commerce: architectural implications for product and engineering teams, accessed May 11, 2026, https://stitch.money/blog/agentic-commerce-architectural-implications-for-product-and-engineering-teams
15. What Is Agentic Commerce? The Complete Guide for Retailers - Paz.ai, accessed May 11, 2026, https://www.paz.ai/agentic-commerce
16. What Is Agentic Commerce? | IBM, accessed May 11, 2026, https://www.ibm.com/think/topics/agentic-commerce
17. From Clicks to Protocols: The Technology Stack Behind Agentic Commerce - IBM iX, accessed May 11, 2026, https://ibmix.de/en/blog/technology-stack-behind-agentic-commerce
18. From checkers to chess: A brief history of IBM AI, accessed May 11, 2026, https://www.ibm.com/new/product-blog/from-checkers-to-chess-a-brief-history-of-ibm-ai
19. Memory Scaling for AI Agents | Databricks Blog, accessed May 11, 2026, https://www.databricks.com/blog/memory-scaling-ai-agents
20. IBM and Red Hat, accessed May 11, 2026, https://www.ibm.com/products/cloud/redhat
21. How governance increases velocity | IBM, accessed May 11, 2026, https://www.ibm.com/thought-leadership/institute-business-value/en-us/report/ai-governance-trends