Dun and Bradstreet Rebuilt Its 642 Million Business Database for AI Agents in 2026

Dun and Bradstreet just did something almost no legacy data provider has tried at this scale. The 180-year-old credit and risk firm tore down the architecture behind its Commercial Graph, the database that powers nearly 200,000 enterprise customers, and rebuilt it from the ground up so that AI agents could query it the same way human analysts have for decades. Michael Nuñez at VentureBeat reported on May 21 that the rebuild covers 642 million businesses, 11,000 fields per record, and roughly 100 billion data quality checks per month. At Metaintro, we track these enterprise rebuilds because they reshape which roles get hired, which get reskilled, and which get quietly absorbed by agentic AI.

What changed inside the Commercial Graph, and why does it matter for workers?

The Commercial Graph was never a single tidy database. It was a federation of separate systems built for different markets, different use cases, and different decades of business technology, stitched together by custom integrations that human analysts learned to navigate through SQL queries or pre-built interfaces. Credit officers could wait. Procurement teams could wait. Sales reps could click through ambiguous entity matches and pick the right one. None of that translates to a workflow where an AI agent has joined the workforce and is firing off thousands of queries per minute at sub-second latency. Gary Kotovets, Chief Data and Analytics Officer at Dun and Bradstreet, told VentureBeat that the firm now treats agents as a distinct consumer category, on par with the credit analysts and sales professionals D&B has served for nearly two centuries. The database itself almost doubled in five years, from roughly 300 million records to 642 million, and the legacy fragmentation meant agents could not retrieve answers fast enough or consistently enough to be trusted in production. That is the kind of plumbing problem that quietly decides whether a finance team can layer AI skills onto an existing career or watch the whole department get restructured around new tooling.

Why are static relationships the wrong design for AI agents?

Legacy enterprise databases tend to capture point-in-time relationships. A CEO is linked to a company. A subsidiary is linked to a parent. The line is drawn once and frozen. Human analysts understand intuitively that when a CEO leaves for a new firm, the track record might or might not travel with them, and they ask follow-up questions to figure it out. Agents do not ask follow-up questions in the same way, and they certainly cannot wait for a custom analyst report. D&B rebuilt the underlying schema to support dynamic relationships, the kind that propagate when a subsidiary changes ownership or an executive moves to a new role. That single design choice has knock-on effects for jobs. Credit analysts, third-party risk managers, and procurement specialists who used to spend hours stitching together a corporate hierarchy now spend that time validating what the agent already pulled. Workers who can review agent output, catch the misses, and explain the logic to a regulator are exactly the kind of profile that companies are starting to call indispensable. We have written about this profile in our piece on AI agents and the workplace skills that help you thrive, and it is the through-line for almost every enterprise AI rebuild story right now.

How did D&B actually rebuild the Commercial Graph?

The technical sequence matters because it tells job seekers and managers where the new openings will sit. D&B started with consolidation. Fragmented databases were migrated to cloud infrastructure, the schema was redesigned, and a data fabric layer was added to normalize records across regions while preserving local compliance requirements. On top of that unified graph, the firm built a structured access layer designed specifically for agents. Raw SQL access at the latency agents need was not the answer, so D&B packaged its data behind a set of tools and skills exposed through the Model Context Protocol, or MCP, which has become the default channel for agent-to-data integrations across the enterprise software stack. Behind every query sits a match and entity resolution engine that confirms the agent is talking about a verified company rather than a name collision. This is the part of the stack that hiring managers used to staff with dozens of data stewards, master data management leads, and resolution specialists. Many of those roles are not disappearing. They are being redefined, and the workers who survive the transition are the ones who can read the data analyst playbook and translate it into agent-readable schemas. If you want to see how this overlaps with adjacent paths, our guides on essential data analyst skills and the data analyst interview questions walk through the muscle memory employers want.

What is Know Your Agent, and why is identity the next hiring battleground?

Rebuilding the graph solved data retrieval. It did not solve identity. The authentication models that worked for human users do not extend cleanly to machines, so D&B built what Kotovets calls Know Your Agent, a concept he borrowed deliberately from Know Your Customer compliance frameworks. Every agent that queries the Commercial Graph must map to a verified IP address and register an individual access key, and that key is then treated as an authenticated identity inside the same pipeline that vets human users. Kotovets told VentureBeat that this verification layer is what allows the firm to give agents the same level of trust it gives a longtime credit analyst. The hiring implication is direct. Identity, governance, and compliance roles are quietly absorbing AI-specific responsibilities, and job descriptions are starting to require both regulatory fluency and an understanding of how machine identity actually works. This is the kind of skill stack we explored in the CIO and CHRO AI talent retention playbook, and it overlaps with what enterprises are trying to standardize after trust gaps in enterprise AI agents pushed several vendors to slow their rollouts.

How do multi-agent workflows stay honest about which company they are analyzing?

Most credit, risk, and supply chain workflows do not run a single agent. They chain several. A credit check agent passes its output to a KYC agent, which hands off to a third-party risk agent, which loops back to a procurement workflow. Each step queries D&B independently. Without a mechanism to confirm that every step is referencing the same legal entity, an entire workflow can finish on divergent records and still look clean on the surface. D&B built a business verification agent that sits inside customer workflows as a persistent reference point, available through Google's Agent2Agent or A2A protocol so that it works regardless of which orchestration tool a customer prefers. Kotovets compared the pattern to a digital verification step between agents, and it is the kind of design that quietly raises the bar for what enterprises consider acceptable agent governance. For workers, the lesson is simple. Roles that include agent QA, lineage validation, and workflow design are emerging fast, and they sit comfortably next to the skills that beat job titles in 2026 AI workplaces. Workers who can spot when two agents are operating on divergent records before a regulator does are scarce, and they are starting to command real salary premiums similar to what we documented in the AI skills salary premium piece.

What does the D&B rebuild tell other enterprises about agent readiness?

Kotovets said he has spoken with hundreds of Chief Data Officers and CIOs over the past six months, and the same constraint comes up every time. They cannot build what they want in AI because their data foundations are not standardized, not normalized, and not queryable by agents. D&B had that foundation in part because it has spent decades curating it, and it still had to rebuild. Most enterprises do not, and that gap is now showing up in earnings calls, pulse surveys, and the kinds of stories we covered in the 2026 AI rollouts failing on broken foundations write-up. D&B's playbook condenses into four requirements. Data foundations come before agent infrastructure. Dynamic relationships, not static ones, must be the default. Entity consistency checks have to be engineered into multi-agent workflows from the start. Lineage must be embedded at the schema level, not added once the regulators show up. Each of those requirements creates new openings, from data governance leads to lineage engineers, and they are the kinds of roles many job seekers can pivot into using the same logic we laid out in the older worker AI upskilling edge and the AI reskilling gap piece.

Who benefits when 642 million records become agent-ready, and who has to adapt?

Three groups gain something concrete from this rebuild. The first is data engineers and data stewards inside enterprises who now have a working reference architecture they can point at when their leadership asks why the in-house equivalent is taking so long. The second is risk and compliance professionals who can use D&B's verification model as a baseline when designing agent guardrails for their own organizations, particularly inside regulated industries that watch every audit trail. The third is the broader population of job seekers who recognize that agent-readiness is becoming a top hiring filter and start future-proofing their careers before the next round of restructuring lands. The flip side is harder. Anyone whose daily work consists of clicking through legacy interfaces, running pre-built reports, or stitching together corporate hierarchies by hand needs to start layering an agent-aware skill set onto that work. The rebuild D&B just shipped is not a one-off. It is a template that every B2B data provider, every enterprise SaaS platform, and every internal data team will be expected to copy within the next two years, and the workers who understand the shift first will pick the best openings.

What new job categories should you watch as agent-ready data becomes the default?

The D&B rebuild is a useful preview of the job descriptions that will dominate enterprise data hiring through 2026 and into 2027. Agent governance leads will write the rules for which agents can touch which records and how their access is audited. Lineage engineers will own the traceable path that every agent answer has to carry, so that a regulator can click back to the original source without anyone scrambling. Entity resolution specialists, a role that already exists inside master data management teams, will expand into agent-facing work where every query must resolve to a verified entity rather than a fuzzy name match. Schema architects who understand how to model dynamic relationships rather than static ones will become the equivalent of the cloud architects of the early 2010s, in short supply and well-paid. We have written about adjacent emerging roles in our coverage of the agentic AI digital colleague, and the pattern is consistent. Every time enterprise software adds an agent layer, a small wave of net-new roles appears, often before the official job titles catch up. Job seekers who position themselves at the seam between traditional data work and agent infrastructure are likely to find the cleanest path through what could otherwise look like a confusing rebuild.

How should job seekers respond to the D&B rebuild this quarter?

The honest answer is that the rebuild itself does not change your week. The signal it sends does. When a 180-year-old data company tells the market that its core asset had to be rebuilt for AI agents, every Chief Data Officer who reads that headline starts asking whether their own systems can clear the same bar, and the answer is almost always no. That triggers budget conversations, vendor evaluations, and a fresh round of hiring or restructuring. The workers who do best in that window are the ones who can speak both languages. They understand how the legacy systems actually work, because they used them, and they can read a Model Context Protocol spec well enough to explain what an agent will need from those systems next quarter. If you are sitting in a data role today, this is the moment to take a free MCP tutorial, to read one well-explained breakdown of agentic workflows, and to ask your manager whether your team has a written plan for agent readiness. If you are interviewing, expect the question. If you are a hiring manager, expect candidates to ask whether your stack is closer to D&B's rebuilt graph or to the fragmented legacy that came before it. The companies that can answer honestly will hire the strongest profiles, and the companies that cannot will struggle to recruit the same workers they were courting six months ago. Either way, the bar for what enterprise data looks like has just moved, and the careers that adapt earliest will be the ones still standing when the rebuild becomes the default rather than the exception.

Related Articles

• Agentic AI career skills will define the next wave of hiring in 2026
• AI agents have joined the workforce in 2025
• AI agents and the workplace skills that help you thrive
• How to become a data analyst without a degree
• Essential data analyst skills for 2025
• Data analyst interview questions for 2025
• The CIO and CHRO AI talent retention playbook for 2026
• Enterprise AI agents and the trust gap shaping jobs in 2026
• Five skills that beat your job title in 2026 AI workplaces
• The AI skills salary premium for prompt engineering and MLOps
• 2026 AI rollouts are failing on broken foundations
• Five ways to future-proof your career in every industry changing in 2026

People Also Asked

Q: Why did Dun and Bradstreet need to rebuild a database that was already serving 200,000 customers?

A: Because the existing Commercial Graph was a federation of legacy systems designed for human analysts who could wait for query results and click through ambiguous matches. AI agents need sub-second latency, dynamic relationships, and verified entity resolution, none of which the legacy architecture could deliver. Customers pushing agents into credit, procurement, and risk workflows hit those limits fast, which is what triggered the rebuild.

Q: What is Know Your Agent and how is it different from Know Your Customer?

A: Know Your Agent is a verification model D&B built so that every AI agent querying its data must map to a verified IP address and a registered individual access key, then be treated as an authenticated identity in the same pipeline as a human user. It borrows the spirit of Know Your Customer compliance, but applies it to machine identities, which the existing user authentication models were never designed to handle.

Q: Which jobs are most likely to benefit from D&B's rebuild, and which face the most pressure?

A: Data engineers, lineage owners, entity resolution specialists, agent governance leads, and schema architects who can model dynamic relationships are likely to see strong demand. Workers whose daily job is clicking through legacy interfaces or compiling reports by hand face the sharpest reskilling curve, because agent-ready data pipelines will absorb much of that routine work over the next two years.

Future-proof your career, with Metaintro you can navigate the agentic AI shift before your company does, and land in the roles that come out stronger on the other side of the rebuild.