Managing 150,000 AI agents would not be only a technology challenge. It would also force database teams to decide how much information each agent can access, what actions it can take, and how people can review its work.
That challenge sits at the center of a larger enterprise question: how can organizations delegate consequential work to AI while retaining control? Mala Ramakrishnan and Vinit Sahni approach that question through privacy, integrity, trust, and governance.
Three ideas that sound alike but work differently
Mala Ramakrishnan, Founder and Managing Partner at Progressive Ventures, has two decades of experience working at the intersection of product, data, and privacy. In an article published on Sep 22, 2026, she draws a line between three ideas that often appear together: “Privacy. Integrity. Trust.”
Those words are used interchangeably, Ramakrishnan states, but they do not mean the same thing. That difference matters when organizations manage large numbers of AI agents, especially agents that work with databases and other forms of enterprise information.
“Integrity runs on disclosure. You can only trust a number if you can see where it came from, who touched it and whether anyone changed it. Audit trails, provenance, receipts,” Ramakrishnan writes. Her point puts visibility at the heart of reliable data work: a result needs a record showing where it came from and what happened to it.
Privacy follows a different path. “Privacy runs on the opposite instinct: lock it down, show less, forget what you don’t need.” An organization that protects information must limit exposure and avoid keeping data that it does not need.
These goals can pull in different directions. Integrity calls for disclosure, records, and a clear history of changes, while privacy calls for restricted access and less information. Every organization settles the tension between privacy and integrity somewhere.
Why scale changes the governance problem
The idea of managing 150,000 AI agents makes that tension easier to see. A database team would face a vast number of agents whose work would need governance, with privacy and integrity shaping how that work could be controlled.
The question is not only whether agents can perform tasks. It is also how an organization keeps control when those agents handle consequential work. Vinit Sahni, Founder of governr, focuses on that exact issue: how enterprises can delegate consequential work to AI while retaining control.
Control requires a way to connect an agent’s work with the information behind it. Ramakrishnan’s description of audit trails, provenance, and receipts provides three clear ideas for that connection. At the same time, her description of privacy points toward limits on what an agent can see, retain, and share.
That creates a practical governance problem. More records can make an agent’s work easier to inspect, but more information can also create a larger privacy burden. Less access can protect information, but it can also leave less evidence for checking how a result was produced.
There is no single answer in the facts presented here because every organization settles the privacy and integrity tension somewhere. The important point is that the decision cannot be separated from the way AI agents operate.
Enterprise adoption remains a work in progress
McKinsey’s 2025 survey shows why this discussion matters. The survey found that 23% of organizations were scaling an agentic AI system somewhere in the enterprise, while 39% were experimenting with AI agents.
Those numbers show a gap between trying the technology and expanding it across an organization. Fewer than 10% of organizations were scaling AI agents in any individual business function, which places the idea of managing 150,000 agents far beyond a routine deployment.
The survey also gives the governance debate a clear setting. Organizations are experimenting with AI agents, and some are scaling agentic AI systems, but fewer than 10% are scaling agents within any one business function. That leaves enterprises facing control questions before large-scale adoption becomes common inside individual teams.
Ramakrishnan’s article was published on Sep 22, 2026, and Sahni’s article was published on Sep 23, 2026. Together, their focus points to the same challenge from different angles: AI adoption needs both trustworthy information and control over delegated work.
For database teams, the central issue is not only the number of agents. It is the system used to govern them. Privacy asks the organization to lock information down, show less, and forget what it does not need. Integrity asks the organization to preserve disclosure, provenance, and records.
At 150,000 agents, those choices would shape how an organization understands its own data and its AI systems. The debate begins with three words, but the work depends on keeping their meanings separate: privacy, integrity, and trust.
Based on
- What managing 150,000 AI agents could look like for database teams — thenewstack.io
- 4 Questions To Ask Before You Let AI Agents Near Your Company’s Data — forbes.com
- As AI Agents Gain Authority, CIOs Must Redesign The Enterprise — forbes.com
- AI Agents Can Do The Work: Companies Hesitate To Let Them — forbes.com
