I recently had the opportunity to go back to the AWS London office, and meet some of my former colleagues there in an Executive Briefing Center with one of my customers. In the Culture of Innovation session, which my former teammate delivered, he mentioned Team Tenets as a mechanism to explore. I have been reflecting on Team Tenets for some time now, as a way to align all team members - humans and agents alike - on what the team does, why, and how.
I first wrote about Team Tenets in issue #22 of this newsletter, after writing them together with a client's AI core team. This post continues from there, now that agents are becoming part of the team.
What is a tenet?
At Amazon, a tenet is described as a belief that helps the team make decisions faster, by making clear what is important to it and, by extension, what isn't. Where the mission says what the team does, the tenets say how it does it. A good tenet says that the team cares more about one thing than another, so it can break the tie when two reasonable options conflict.
Firecracker, the open-source technology AWS built to run Lambda functions, is a good real example. One of its published tenets reads:
"Minimalist in Features: If it's not clearly required for our mission, we won't build it."
Plenty of other teams choose to build a rich set of capabilities and do well with it, which is why this one helps the Firecracker team decide.
The dual purpose of Team Tenets
The first purpose is inside the team. Tenets help everyone on the team, whether they joined last week or years ago, understand how the team works and the reasons behind it, and they settle many recurring discussions before they happen.
The second purpose is outside the team. Tenets let other teams and stakeholders understand how you work and what you care about, so they can engage with you better.
So what changes when agents join the team?
I think both purposes now have a new kind of reader.
Inside the team, an AI agent is a new starter almost every time you work with it. In most setups it does not carry over the discussions the team had last month, and it does not pick up the culture from meetings and conversations the way a person does. It mostly knows what you write down for it, which is one more reason to keep your tenets short and clear. Amazon's own AI coding tool, Kiro, already has a place for this: a product file that, in Kiro's words, "helps Kiro understand the 'why' behind technical decisions." Other tools use an AGENTS.md or CLAUDE.md file, or the instructions of a custom GPT, for the same job.
Outside the team, your agents may increasingly be the ones answering other teams' requests and preparing material for stakeholders. When they do that, they represent the team, so they should be working from the same beliefs as the people on it.
Many of the hard decisions in a team are trade-offs between two good things, and agents need help with those too. When Anthropic (the company behind Claude) let an AI agent run a small shop in its office, in an experiment called Project Vend, the agent kept agreeing to discounts because it was trained to be helpful. Nobody had told it which mattered more when helping customers and keeping the shop profitable were in conflict, and that is the kind of statement a tenet makes.
A few examples for a hybrid team
The tenet I shared in issue #22 from the client's AI core team ended with "Full automation is the exception, not the rule, until we feel ready for more." You could adapt the three below for a team where people and agents work together:
- A person owns every outcome. Agents do the work we delegate to them, and the person who delegated it is accountable for the result. Linear (the project management software company) wrote a similar principle into its guidelines for agents: "An agent cannot be held accountable."
- We prefer reversible actions. Agents act on their own when a step can be undone, and check with a person before anything that can't be - the same idea as Amazon's one-way and two-way doors.
- We say when an agent did the work. Stakeholders always know whether a person or an agent produced what they receive.
Putting tenets to work
If you want your Team Tenets to guide both the people and the agents, put the same text in both places - on the page where the team describes itself, and at the top of the instructions your agents read. Keep them to a handful (Amazon's rule of thumb is no more than seven), in order of priority.
At Amazon, a list of tenets usually ends with the phrase "unless you know better ones", which invites anyone to challenge them. I suggest you treat it as a practice and not just a phrase, and ask your agents to flag the moments where the tenets did not help them make a decision. Together with what the people on the team have noticed, that list is a good place to start when you sit down to review and improve your tenets.
Written tenets can also be checked. The other tool in this issue, Jev, a model built for small decisions at scale, makes it cheap to turn each of your team's tenets into a yes/no question and check the work against it.
The octopus organization
Phil Le-Brun (an Executive in Residence at AWS, and previously the international CIO of McDonald's) also spoke at the briefing that day, about The Octopus Organization, the book he wrote with Jana Werner. About two-thirds of an octopus's neurons are in its arms, so each arm can sense and decide on its own while still working with the others. I think that is a useful picture for a team where people and agents both make more decisions on their own. Phil has also written a short piece on tenets on the AWS blog.
If you want to dive deeper, watch Phil and Jana in conversation with Barry O'Reilly (the author of Unlearn) about the book. They talk about the 36 anti-patterns the book is built around, why an organization should become adaptive rather than change once for AI, and why leaders need enough understanding of technology to know which questions to ask.
Your action step
Does your team have its tenets written down? If not, write three to five of them in order of priority. End the list with "unless you know better ones". Put the same text on the page where your team describes itself and at the top of the instructions your agents read.
Then ask yourself: if your agents had to follow just one of these tenets starting tomorrow, which one would matter most? I would like to hear your answer.
If you want to write Team Tenets with your own team, people and agents together, that is the kind of work I take on in AI strategy advisory engagements and in sessions as an AI keynote speaker and workshop facilitator.