I spent years at Amazon living and breathing the leadership principles. They weren't motivational posters. They were decision-making tools - used daily in hiring loops, product reviews, strategy debates, and the countless judgment calls that shape how a company actually operates.
Recently, I ran a session with AI executives and champions at a customer I'm working with. The question on the table: as AI agents become part of our teams, which leadership principles matter most?
My answer surprised some of them. It's the same principles that always mattered. They just apply differently now - because your "team" is no longer exclusively human.
Here are the seven I believe matter most when your organization is deploying AI agents alongside people.
1. Customer Obsession - The North Star for Every Agent
At Amazon, every significant decision starts with the customer and works backwards. This isn't a value statement. It's an operating method. You write the press release before you build the product. You ask "what does the customer need?" before you ask "what can we build?"
This principle becomes even more critical with AI agents because agents don't have intuition. A seasoned human employee can sense when a customer interaction needs a different approach - when to slow down, when to go the extra mile, when the script isn't working. That intuition comes from years of absorbing what customer obsession actually means in practice.
An agent has none of that built in. If you tell it to resolve tickets quickly, it will resolve tickets quickly - even when the customer needs something else entirely. I wrote about this in my piece on intent engineering: the gap between what you tell an agent to optimize for and what your organization actually needs is where value gets destroyed.
Customer Obsession is the principle that closes that gap. Before deploying any agent, work backwards from the customer experience. Write the PRFAQ. Describe what the ideal interaction looks like from the customer's perspective. Then build toward that - not toward an efficiency metric.
The question for leaders: If your AI agent interacted with your most important customer tomorrow, would the experience reflect your company's values? Or just its KPIs?
2. Ownership - Agents Don't Own Anything. You Do.
Owners think long-term and never say "that's not my job." This is foundational when humans work alongside agents, because agents have zero sense of ownership. They execute. They don't care about the long-term consequences of their decisions. They don't think about what happens downstream. They won't flag something that seems wrong but technically falls within their instructions.
This means human ownership becomes more important, not less. Someone needs to own the agent's outcomes, not just its inputs. Someone needs to ask: is this agent making decisions that align with our long-term interests? Is it creating technical debt? Is it optimizing for today at the expense of tomorrow?
I've seen teams deploy agents and then treat them like set-and-forget automation. That's the opposite of ownership. Agents drift. Their context gets stale. The world changes around them. Without active human ownership, an agent that was perfectly aligned last quarter can be quietly causing damage this quarter.
The question for leaders: Who owns the outcomes of your AI agents - not just the deployment, but the ongoing alignment with your business goals?
3. Invent and Simplify - The Compound Power of AI Innovation
Leaders expect and require innovation and invention from their teams and always find ways to simplify. This is the principle that separates organizations that bolt AI onto existing processes from organizations that reimagine what's possible.
Most companies are still in the "bolt-on" phase. They take an existing workflow, add an AI layer, and get 20-30% efficiency gains. That's useful, but it's not invention. Invention is asking: now that we have agents that can reason, decide, and act autonomously - what entirely new things can we do that were impossible before?
The simplify part is equally important. I see organizations building incredibly complex agent architectures - multi-agent systems with dozens of handoffs, elaborate orchestration layers, custom tooling everywhere. Sometimes that complexity is necessary. But often, a single well-designed agent with clear intent and good context can outperform a complex multi-agent system. Before you add another agent to the chain, ask whether you can simplify instead.
The question for leaders: Are you using AI to do old things faster, or to do entirely new things? And is your architecture as simple as it could be?
4. Think Big - From Personal Productivity to Organizational Transformation
Thinking small is a self-fulfilling prophecy. I've named my company and my newsletter after this principle because I believe it's the one that separates incremental improvement from genuine transformation.
Most AI deployments today are small. One person uses ChatGPT to draft emails. A team uses Copilot for code completion. A department runs a chatbot for customer FAQ. Each of these is useful. None of them is thinking big.
Thinking Big with AI means asking: what would it look like if every knowledge worker in our organization had a team of AI agents working alongside them - each with access to the right organizational context, each aligned with our goals, each continuously improving? What would our company be capable of that it isn't today?
This is the jump from AI activity to AI fluency that I keep coming back to. Activity gets you percentage-point improvements. Fluency - rethinking how work itself is structured around human-agent collaboration - gets you order-of-magnitude transformation.
The question for leaders: What becomes possible for your organization when AI isn't a tool your people use, but a collaborator your people lead?
5. Bias for Action - Speed Matters in the Age of Agents
Many decisions and actions are reversible and do not need extensive study. This has always been one of my favorites because it cuts through the analysis paralysis that kills innovation.
In the AI agent era, Bias for Action applies in two directions.
First, to organizations deploying agents: don't wait for the perfect architecture, the perfect governance framework, the perfect model. Start. Deploy an agent in a contained workflow. Learn from it. Iterate. The organizations that are learning fastest right now are the ones that deployed agents early and imperfectly, not the ones that are still writing strategy decks about agents they haven't built.
Second, to the agents themselves: the best agents are designed with a bias for action. They don't stall when they encounter ambiguity. They make a reasonable decision, communicate what they did and why, and move forward - with appropriate guardrails and escalation paths. An agent that freezes every time it hits an edge case is not useful. An agent that makes reasonable judgment calls within defined boundaries is powerful.
The key word is "reversible." Give agents the authority to act on decisions that are easily undone. Reserve human judgment for the irreversible ones.
The question for leaders: Are you learning from deployed agents, or still planning to deploy your first one?
6. Dive Deep - You Can't Lead What You Don't Understand
Leaders operate at all levels, stay connected to the details, and audit frequently. No task is beneath them. This principle is the antidote to the most common failure mode I see in enterprise AI: leaders who delegate AI strategy entirely to the technology team without understanding what agents actually do.
You don't need to write code. But you do need to understand how your agents make decisions. You need to read the conversation logs. You need to understand what context the agent has access to and what it doesn't. You need to know what happens when the agent encounters a situation it wasn't designed for.
The leaders who Dive Deep into their AI systems find things that dashboards never show - edge cases where the agent behaves badly, patterns where customers get frustrated, moments where the agent's judgment diverges from what the organization would actually want. These are the insights that drive real improvement.
Without Dive Deep, you're managing a dashboard. With it, you're managing an outcome.
The question for leaders: When was the last time you personally reviewed a conversation between your AI agent and a customer?
7. Earn Trust - The Make-or-Break Principle for AI
Leaders listen attentively, speak candidly, and treat others respectfully. They are vocally self-critical, even when doing so is awkward or embarrassing.
Earn Trust is the principle I believe will define which organizations succeed with AI and which don't. Because trust is fragile with autonomous systems, and rebuilding it is expensive.
When an AI agent makes a mistake with a customer, the damage is different from when a human makes a mistake. Customers extend grace to humans - we all make errors, we understand. But when an AI system gets it wrong - gives incorrect information, mishandles a sensitive situation, makes a customer feel like they're talking to a wall - the trust damage is disproportionate. People don't just lose trust in the agent. They lose trust in the company that deployed it.
This means organizations need to be vocally self-critical about their AI systems. When agents fail, acknowledge it publicly. When you discover your agent was optimizing for the wrong thing, say so and explain what you're changing. Transparency about AI limitations earns more trust than pretending the system is perfect.
It also means building trust between humans and agents inside the organization. Your employees need to trust the AI tools they work with. That trust is earned through transparency (they can see what the agent did and why), reliability (it works consistently), and humility (the agent knows when to escalate instead of guessing).
The question for leaders: If your AI agent made a significant mistake tomorrow, does your organization have a plan for how to communicate that transparently - to the customer and to your team?
The Principle Behind the Principles
Looking at all seven together, there's a meta-principle: the leadership principles that work for human organizations work for human-agent organizations too. They just need to be operationalized differently.
With humans, principles spread through culture - through stories, through mentoring, through watching how leaders handle tough situations. With agents, principles need to be encoded explicitly - in system prompts, in decision boundaries, in goal structures, in feedback loops.
The organizations that will thrive aren't the ones with the best AI models. They're the ones that take the principles that made them great and systematically embed them into how their agents think, decide, and act.
That's the real leadership challenge of our era. Not whether to deploy AI agents - that ship has sailed. But whether you can make those agents reflect the values, judgment, and long-term thinking that define your organization at its best.
The principles haven't changed. How we live them has.