---
id: "framework-autonomous-scrum"
type: "framework"
source_timestamps: ["§ The \\\\\\\"Autonomous Scrum\\\\\\\""]
tags: ["organizational-design", "agile"]
related: ["claim-autonomous-scrums-outperform", "entity-united-airlines"]
steps: ["Form an interdisciplinary team of strictly 6 to 8 people.", "\\\"Assign the team a narrowly defined", "discrete", "and consequential objective.\\\"", "\\\"Unshackle the team from bureaucratic approval chains", "granting them permission to ACT", "not just recommend.\\\"", "Equip the team with AI agents for deep subject matter expertise and rapid feedback loops.", "Require any leadership veto of the scrum's actions to be time-bound and evidence-backed."]
speakers: ["Jonathan Rosenthal", "Neal Zuckerman"]
sources: ["governance"]
sourceVaultSlug: "hbr-seg-governance"
originDay: 7
articleStem: "hbr-sig-59-consensus-decision-making"
sourceUrl: "https://hbr.org/2026/04/decision-making-by-consensus-doesnt-work-in-the-ai-era"
sourceTitle: "Decision-Making by Consensus Doesn’t Work in the AI Era"
---
# The Autonomous Scrum Architecture

A structural redesign of team dynamics that builds on traditional agile methodology (see prerequisite [[prereq-agile-methodology]]) but departs from it by shifting the team's mandate from *advisory* to *authoritative*. Traditional scrums recommend and escalate; Autonomous Scrums own outcomes and have permission to act. The architecture requires leadership to cede power, tolerate more frequent missteps in favor of speed, and provide a robust data environment. AI agents act as 'savants' for these scrums, providing deep subject-matter expertise and instantaneous feedback on actions.

**Steps:**
1. Form an interdisciplinary team of strictly **6 to 8 people**.
2. Assign the team a narrowly defined, discrete, and consequential objective.
3. Unshackle the team from bureaucratic approval chains, granting them permission to **ACT, not just recommend**.
4. Equip the team with AI agents for deep subject-matter expertise and rapid feedback loops.
5. Require any leadership veto of the scrum's actions to be **time-bound and evidence-backed** (see [[action-require-evidence-backed-vetoes]] and [[concept-pocket-veto]]).

The performance claim is [[claim-autonomous-scrums-outperform]]; the implementation action is [[action-empower-autonomous-scrums]]; the primary case study is [[entity-united-airlines]] (2002–2006, six cross-disciplinary working groups). It pairs naturally with the [[framework-ovis]] decision-rights model, which clarifies who owns and who may veto within and above the scrum.

**Calibration (from enrichment):** Adjacent to well-established patterns — Amazon's 'two-pizza teams' and Single-Threaded Owners, and the Spotify model (squads/tribes/chapters/guilds). Fully empowered scrums, however, are not a panacea: at scale they require clear mission boundaries, platform/enabling teams, and portfolio-level governance to avoid local optimization and fragmentation.


## Related across articles
- [[concept-enc-teams]]
- [[framework-ovis]]
- [[concept-modular-leadership-systems]]
- [[concept-first-line-defense-shift]]