Manifesto for Agentic Teams

Manifesto for<br>Agentic Teams

We are discovering better ways of building software by combining human judgment with AI agents.

Outcomes<br>over<br>output

More code is not more value. We measure what ships to users, not what ships to the merge queue.

Context<br>over<br>tooling

The best AI tools fail without accessible knowledge. We invest in making our organization's context machine-readable.

Orchestration<br>over<br>execution

Our job is to direct, review, and decide. We delegate production to agents and focus human energy where judgment matters.

Continuous flow<br>over<br>fixed ceremonies

We ship when ready, not when the sprint ends. We plan around review capacity, not production capacity.

Fluid boundaries<br>over<br>team silos

We design for collaboration across team lines: shared context, open channels, people and agents move to where the work is.

That is, while there is value in the items on the right, we value the items on the left more.

Principles

01

Our highest priority is to deliver value to users. Agents that produce output nobody uses are waste, not progress.

02

AI is an amplifier, not a miracle. It accelerates what's already there. If the foundation is weak, AI makes it worse faster. We invest in the foundation first.

03

We treat Context Debt like Technical Debt. Undocumented decisions, tribal knowledge, and stale documentation are liabilities. We pay them down systematically because our agents depend on it.

04

We measure outcomes, not velocity. Features shipped, problems solved, users served. Not PRs merged, lines generated, or story points completed.

05

The bottleneck has shifted. Code is cheap, judgment is scarce. We organize our team around review, verification, and decision-making, not around production.

06

Every agent has an identity, an owner, and a decision log. We don't deploy autonomous systems without accountability. We encode governance into pipelines and guardrails so compliance happens at deployment speed, not committee speed.

07

We delegate deliberately. The team has shared criteria for what gets delegated to agents, what stays human, and what's a collaboration. This is a team discipline, not an individual preference.

08

We recognize and reduce AI Waste. Overproduction, review bottlenecks, governance theater, context loss, over-prompting, wrong granularity, unused output, and manual work that should be automated. We audit regularly and fix relentlessly.

09

Small batches, continuous deployment, automated quality gates. Agent output follows the same engineering standards as human code. No exceptions, no shortcuts.

10

We build shared knowledge, not individual expertise. Prompt patterns, agent configurations, and delegation criteria belong to the team. When someone leaves, the knowledge stays.

11

We talk openly about how AI changes roles and identities. Senior engineers design systems instead of writing code. Juniors learn by reviewing agent output and understanding why. We create new narratives, not new anxieties.

12

We plan in short cycles with honest assessments, measurable goals, and named owners. Review, learn, adjust, repeat. Long-term roadmaps are speculation. Disciplined execution in tight loops is where value lives.

13

Most waste hides at team boundaries: handoffs, waiting, and context that doesn't travel. We treat cross-team friction like any other bottleneck: we measure it, own it, and reduce it.

14

Fluid boundaries don't mean no ownership. Every piece of work has a named owner. Owners pull in whoever they need, across teams and roles, and stay accountable for the outcome.

15

We invest in the interfaces between teams: shared context, clear APIs, open documentation. When collaboration across boundaries is cheap, people do more of it.

Stefan Grothkopp

2026

This declaration may be freely copied in any form,<br>but only in its entirety through this notice.