Beyond Autonomous Teams: Essence and Accidents in Organizations Delivering Value – Simon Rohrer (Talk Outline)

Simon Rohrer (co-author of Sooner Safer Happier) argues that “autonomy” and “product” are container concepts that mean too little, and reframes teams as value centers needing agency + coherence rather than autonomy. The shape of your value — including its essential coupling — designs your organization at every fractal level, and the same lenses (Viable Systems Model, both-and thinking, agency within boundaries) increasingly apply to AI agents.

1. “Who Are We?” — The Expanding “We”

• An LLM listened to his talks and told him his verbal tic is ”so.”
• Title: Beyond Autonomous Teams — Essence and Accidents in Organizations Delivering Value. Subtitle nods to Toy Story: “To autonomy and beyond.” The original subtitle said product teams — he learned why not.
• Amsterdam anecdote: colleague Marcel asked “do we need to worry about token costs?” Simon pushed back — which we? You-and-me, the office, the tech org, Saxo Bank (the trading platform), or the whole software industry? The word “we” should carry different meanings as we expand who we think we are.

2. What “Autonomy” Really Means

• Tavistock Institute post-WWII socio-technical work (The Loss, Rediscovery, and Transformation of a Work Tradition): as coal mining mechanized, external supervision grew and longwall mining failed until miners rediscovered pre-mechanization autonomy — which made them more efficient.
• Etymology: autonomy = autos + nomos = “self-laws.” Simon doubts teams actually want to be subject only to their own laws.
• Container concepts (via friends Evelyn & Kenny, a Dutch/German idea): words like sustainability, quality, “full-service total solution” — you fill them with your own meaning, so they contain little. Autonomy is one.
• Contrast: Bernard Williams’ “thick concepts” (courageous, generous) — evaluative and substantively descriptive at once.
• João Rosa (interview with Rein): teams need agency and must understand the web of dependencies — “a 20-person start-up differs from a 2,000-person bank differs from a huge global company.” Teams don’t need unlimited authority but “clear boundaries within which they can act freely.”
• Personal analogy: Brexit (10 years ago) sought autonomy — a bad decision; he moved to Denmark, which acts powerfully within the EU (a bigger “we”).

3. From “Product” to “Value” to “Value Centers”

• “Product” is also a container concept.
• Co-wrote Sooner Safer Happier with John Smart, Zsolt (from Budapest), and Miles. Its balanced-scorecard outcome: Better Value, Sooner, Safer, Happier — better = quality, sooner = flow, safer = agile-not-fragile/security/privacy/minimum-viable-compliance, happier = customers/colleagues/citizens/climate.
• Evolution of the phrase: better product faster → better value faster → better value sooner → better value, sooner, safer, happier.

3.1 Where “product teams” came from

• Scrum was inspired by the “New New Product Development Game” (HBR) about brand-new stand-alone products, via the book Wicked Problems, Righteous Solutions — the rugby scrum metaphor and the product owner.
• In a complicated org, what one team delivers is not a product. At Barclays they realized “products” (current accounts, mortgages; or stocks, bonds) aren’t delivered by one team — nor exactly a “service” — so they jokingly called them “a banana.”

3.2 Value streams → value centers

• Sooner Safer Happier called everything a nested value stream (team → team-of-teams → … → whole org).
• Chris Matts: a software value stream isn’t sustainable/repeatable — it’s a one-off that changes every quarter/project/feature (different team combinations). Dave Snowden likewise cautions against manufacturing metaphors.
• New term (with John, for the next book): value center — inspired by cost centers, but reframed positively: don’t call people a sunk cost (not psychologically safe); every team exists because it is valuable. So squads/tribes/pods/canoes → one word: value centers.
• Bank example of nesting: cash equities → listed derivatives → equities → global markets → investment bank → universal bank (with an HR value center inside).

4. The Shape of Your Value

• The web of dependencies is the shape of your value, and it’s unique to each org. Some coupling is accidental (removable); some is essential (the subtitle’s “essence”).
• Vlad Khononov (last year): there’s always some coupling left — you can’t fully domain-driven-design or Team-Topologies it away. Essential coupling is the shape of your value.

4.1 Molecular value shapes

• Hyperscaler (public cloud): very simple atoms — storage, elastic compute, DNS, Kubernetes are near-standalone “micro-products,” so two-pizza teams work (light dependencies on IAM/portal).
• Trading platform (his company): one product/service, but ~80 teams with a large shared kernel — almost every team (beyond onboarding/support) must know about orders, prices, risk, portfolios.
• Universal bank: far more complex still — many orgs have this level.

4.2 Value designs organizations

• Conway’s Law — everyone knows it. Ex-colleague Mel (book Follow the Money) extends it: finance/funding, not just org structure, shapes systems.
• Sanchez & Mahoney (1996): ostensibly organizations design products, but products design organizations — Simon takes it one step further: value designs organizations, at all levels (fractal), and constrains the autonomy/coherence trade-off.

5. The Viable Systems Model — Five Questions

• A cybernetic theory (Stafford Beer), modernized. Every value center (team at any level) faces five questions:
  • What value am I delivering? (an I question)
  • How do we coordinate? (e.g., stand-up / timetable)
  • How do we fit together? (recombine tickets into a coherent whole)
  • What is out there for us? (above-and-beyond — next week, extra scope)
  • Who are we? (identity/purpose — “what are we here for?“)
• Left-side = I questions; the rest = we questions. They recur fractally (team → team-of-teams → department) against an environment, with a feedback loop between purpose, value delivered, and environment.

5.1 Purpose = what it does

• Simon Wardley: start with purpose (the “why”), the Strategy Cycle.
• Stafford Beer: “The purpose of a system is what it does” (POSIWID). You don’t get to choose your purpose. If a hospital A&E sends 50% of patients away sicker, its current purpose is making people sick. “We are what we do” — not what we ought to do — at all levels.

6. Both-And Thinking (Duck & Rabbit)

• The duck–rabbit illusion (1800s German magazine; Wittgenstein’s aspect perception) — it’s both. Simon and Sinan emphasize both-ands over either-ors.
• Autonomy vs. coherence (not “alignment”): it’s both. Ivo Velitchkov’s Essential Balances — dove of freedom (left) and glue-or-lubrication (right, which can be both); ”organizations are made of organizations.”
• Order vs. disorder (Ralph Stacey): managers face a paradox of intertwined order and disorder; most prefer order but must handle both — hence Simon prefers “coherence” (more disorder than order lately).
• Governing vs. enabling constraints: governing constraints suit order (align early); for disorder you cohere later.
• Nominalism vs. universalism (Barry, post-structuralism; Derrida vs. Deming): maybe nothing is truly common across organizations, but you must pick a pragmatic floor — “survival is not mandatory.” Mark Eddleston’s book on why change fails cautions against linear transformations.
• Disclaimer: this talk is not org design — it offers lenses to think about the org you already have.

7. From Autonomy to Agency + Coherence

• Agency is a better word than autonomy (per João too), balanced with coherence.
• Supporting thinkers:
  • Carol Gilligan (via Pat Swaital): moral development peaks in the interplay of care for others and the interconnected self.
  • Jay Bloom / David Bakan: a drive toward both agency and communion (the coherent whole, the bigger “we”).
  • Alicia Juarrero (governing/enabling constraints): self-assertion and integration (Koestler’s holons).
  • Fred & Merrelyn Emery: less “autonomy,” more the humble “elbow room.”
  • Daniel Pink’s Drive (autonomy/mastery/purpose) rests on Deci & Ryan’s self-determination theory — whose third element is relatedness (Richard Ryan/Bowden note it’s about the group), balancing autonomy and coherence.
• Worry: even agency + coherence risk being container concepts — he tries to make them thicker.

7.1 Sensemaking, strategy, policy

• Sensemaking happens at the individual/team level (interacting with the environment), but there must be a bigger “who are we” — you give sense and make sense, top-down and bottom-up at once.
• Strategy isn’t handed down like tablets — Roger Martin’s call-center example: a rep listening hard to complaints feeds strategy upward. Top sets direction; frontline info must feed back up.
• Policy/governance: set governing constraints from the top and enabling constraints (platforms — Gregor’s point — that let you do more, not less).
• Novo Nordisk (Copenhagen) security incident: teams had their own dev platforms, GitHub orgs, no standard security policy; they had to shut down Azure DevOps/GitHub access and rebuild with centralized governance — proof some governance must come from the top, not be handed to autonomous teams.

8. Time Frames, “The Most Important Line,” and Language

• Jay Bloom: different actors think in different time frames — CEOs in years, people in days, AI agents in minutes (or hours if you work with them a lot). Alistair Cockburn’s new book (unifying user stories, use cases, story maps) on the power of verbs is highly recommended.
• “The most important line in software” — the chart of the time horizon of tasks LLMs can complete (50% success), rising steadily. A Meta interview (Bloomberg Odd Lots) measures agency simply as “how long an agent runs without human intervention” (in hours) — Simon thinks agency is a much thicker concept than that.
• Cross-language check: German socio-technical writing says partially/semi-autonomous (more humble); Dutch/French/Hungarian favor “self-steering teams” (no direct “autonomy”); Agile is largely self-organization. “Human agency” → actionability / capacity to act; “AI agent” has become a loanword everywhere, so English uniquely conflates agency and AI agent. Coherence ≈ ”hanging together” (Dutch, Hungarian).
• Conclusion on wording: come down from “autonomous teams” but come up from mere “self-organization” — “teams with agency.”

9. Bringing It to AI Agents

• The same questions apply to AI agents: with five agent tabs open, what value is each delivering and how do you make them coherent?
• João’s “clear boundaries” = the harness/context boundaries for agents; agency = acting freely within them.
• Patrick de Brabander: share context/skills at different levels — an individual builds a skill, shares it with the team, then possibly the department/whole org. The expanding “we” matters in agentic development.
• Context engineering is “the most important thing” in the AI-agent world (he could talk 40 minutes on it; more sophisticated than it looks).

10. Closing — Three Moves

• Product → value (you deliver value, even if “product” doesn’t fit).
• Autonomy → agency + coherence.
• Autonomous product teams → a nested and networked organization (not either/or — the bigger-and-bigger “we”; you needn’t traverse the hierarchy to deliver value; the “nested & networked” image is from Integration Training).
• The shape of your value determines what you can do and your agency/coherence trade-offs (dependent on your value shape, org, and history). Purpose is not static, not cascaded top-down, not necessarily aligned — continuously negotiated up and down, hopefully coherent most of the time.
• ”Who are we? We are the purple stage, Craft Conference, the software engineering community, hopefully value-development professionals. This is we.”

11. Q&A

• Line between leader-of-tribe and leader-of-department? “It depends.” Spotify-style leaders at all levels; Turn the Ship Around leader-leader model — push decisions to the information (department → tribe → squad), except governance/regulator decisions that must sit at the right level.
• Committees instead of individual leaders for coherence? Yes — his org uses a technology council with broad representation; no individuals making decisions; can’t put 3,000 people on every committee, so make forums as representative as possible.
• Challenges balancing agency and coherence — changed with AI? Continual and context-dependent (governing vs. ordered/disordered). Example: static code-analysis standards across the org — AI has made this easier (“just tell the AI to read this and do it”; get teams to clean up tech debt, per Gregor).
• How should leaders balance? Everything is contextual — recognize the balance is a continuous negotiation, not a once-a-year pick.
• How should teams think about the shape of their value? Step back; find the accidental vs. essential; decouple/API what you can (contracts) and accept continual negotiation where you can’t.
• Components of good context engineering? “Come see me afterwards” (open invitation).
• Order/disorder ↔ Apollonian/Dionysian? “Almost certainly” — he’ll go ask Claude.
• Layers with agency but unwilling to take responsibility? Common — improve team composition, give psychological safety, encourage capacity to act; a red flag when people don’t find their own purpose; good leadership finds each person’s purpose to compose the team purpose.

People, Books & References Cited

• Simon Rohrer — speaker (Saxo Bank); co-author, Sooner Safer Happier.
• John Smart, Zsolt, Miles — Sooner Safer Happier co-authors.
• Marcel — Amsterdam colleague (token-cost anecdote).
• Evelyn & Kenny — container concepts (Dutch/German).
• João Rosa — agency, web of dependencies, “clear boundaries.”
• Bernard Williams — thick concepts; Wittgenstein — aspect perception (duck/rabbit).
• Chris Matts — value streams as one-offs; Dave Snowden — manufacturing-metaphor caution.
• Vlad Khononov — residual/essential coupling.
• “Mel” — Follow the Money (finance shapes systems); Sanchez & Mahoney (1996) — products design organizations.
• Stafford Beer — Viable Systems Model, POSIWID; Simon Wardley — purpose/Strategy Cycle.
• Ivo Velitchkov — Essential Balances; Ralph Stacey — order/disorder; Alicia Juarrero, Koestler (holons); Fred & Merrelyn Emery — elbow room.
• Carol Gilligan, Pat Swaital, Jay Bloom, David Bakan — agency & communion.
• Daniel Pink (Drive); Deci & Ryan — self-determination theory; Richard Ryan/Bowden.
• Roger Martin — strategy from the frontline; Mark Eddleston — why change fails; Alistair Cockburn — power of verbs.
• Novo Nordisk — centralized-governance case; Gregor (Hohpe) — platforms as enabling constraints.
• Patrick de Brabander — sharing context/skills across levels; Integration Training — nested & networked image.
• Concepts: value centers, essential vs. accidental coupling, Conway’s Law, container/thick concepts, governing vs. enabling constraints, both-and, sensemaking, agency + coherence, context engineering.

Video: https://www.youtube.com/watch?v=95YZ2ntBl9M — Transcript via yt-transcript.sh; exhaustive outline generated from the full transcript.