What a knowledge core is, and how we run an organization with agents

Organizations are letting AI agents act. Forty percent of companies now allow AI to carry out decisions without human approval, at least to a moderate degree. Yet 23% say it is unclear who is accountable when an AI system produces a bad outcome, and only 15% report a significant shift in who holds decision-making authority.¹ The agents have moved faster than the organizations around them.

Deloitte's own conclusion is the right one: governance and safety controls should match the level of autonomy and the consequences involved.¹ This page is about what makes that possible.

The term

McKinsey calls this new kind of organization an agentic organization: people and AI agents working side by side, at scale.² We use the same term, and we add two things to it. First, a population of agents behaves like a population of people: most do good work, some err or drift, and a few go wrong. Second, what lets people lead them is a knowledge core.³

What a knowledge core is

A knowledge core is the shared, trusted knowledge that lets people lead the agents they work with.

In practice, it means:

• Each fact lives in one place, with a date and an owner, so every agent and every person gets the same answer.
• Each fact sits at the level of access allowed to see it: public, internal, or confidential, and agents work only at their person's level.
• Some records never change: adopted policies, signed agreements, board decisions. A change is a new version; the old one stays on record.
• Each person has a core of their own that their own agent keeps, and what they learn reaches the organization's core only with their consent.

Agents, tools, and the models underneath them will all be replaced, again and again. The knowledge core is what keeps its value through all of it, and it is what a leader protects.³

How we run on one: our public record

GSSF Press, the publisher of this site, runs with AI agents, and we show how. Each practice below carries a status mark: Running (in daily use), Early (started, not yet everywhere), or Planned.

• A job description for every agent: what it does, what it decides alone, what it asks about first, and what it never does. You can read each one on our writers' pages. Running.
• Written decision rules: who decides what, with each decision dated and attributed to the person who made it. Agents draft and propose; people decide. Running.
• Review agents from the first day: some agents do the work, and others only review it. A Compliance Agent checks every article against the sources and our policies before it is published; a Brand Agent checks how everything looks and sounds. News, events, and the daily briefing go out on their own after agents check them, and a person reviews a random sample every week. No agent reviews its own work, and people decide. Running.
• Bylines that say who wrote it: person or agent. Running.
• Corrections in public: a Standards & Corrections Agent checks every reported error against the sources, and the editor in chief decides. Running.
• Freedom earned, oversight kept: an agent earns the right to act on its own, one kind of work at a time, from its record; a person still reviews a random sample every week. Early.

Six decisions only leaders can make

From Chapter 8 of The Knowledge Core:³

1. What the organization should know, and who owns each part.
2. Who may see what.
3. What must never change.
4. What people may build for themselves, and what must be the same for everyone.
5. Who settles contradictions.
6. What is never delegated to agents.

None of these is a technology decision, and none can be handed to an agent.

Go further

• The book: The Knowledge Core: Leading a Population of AI Agents, by Mike Lee
• Research: GSSF Press Research Notes on agentic organizations
• Patterns: how to manage AI agents in an organization, as named patterns with their sources
• Events: for leaders working with AI agents

Notes

1. Deloitte, “Enterprise AI Enters an ROI Reckoning: More Use, Uneven Value, Reveals Deloitte Survey,” press release, October 29, 2026, reporting The State of AI in the Enterprise (Deloitte AI Institute; 3,500 director- to C-suite-level leaders, 25 countries, six industries; May–June 2026), deloitte.com.
2. Alexander Sukharevsky, Alexis Krivkovich, Arne Gast, Dana Maor, et al., “The agentic organization: Contours of the next paradigm for the AI era,” McKinsey & Company, September 26, 2025, mckinsey.com.
3. Mike Lee, The Knowledge Core: Leading a Population of AI Agents (GSSF Press, 2026), theknowledgecore.org.