Architecture — The Self-Healing Nervous System
The three unshakeable pillars: Steering Wheel, Data Fuel, Ecosystem of Agents.
Discipline in the Self-Driving Enterprise is no longer carried on the shoulders of people alone — it is transferred directly into the architecture of the company. Three pillars hold the foundation: Data Fuel, the Ecosystem of Agents, and the Steering Wheel.
Pillar S — Steering Wheel: strategic guidance and the ethical Glass Box
The Steering Wheel is the leader's interface to the autonomous machinery. It removes the scary black box and replaces it with Glass Box logic: every algorithmic decision is transparent, explainable, auditable.
Three critical functions run through the Steering Wheel:
- Dynamic policy injection — when a sustainability rule changes, the policy is fed into the system and all 50 agents update their behaviour within milliseconds.
- Shield module — the autonomous resilience of the enterprise. A set of agents whose only job is to monitor compliance and security of other agents.
- On-the-loop responsibility — the leader is not inside routines but on top of the whole, addressing only strategic deviations.
Pillar D — Data Fuel: the data refinery
Drucker's adage — what you can't measure, you can't manage — needs an upgrade. In an autonomous world, simply measuring is the past. Data must be fluid, contextual, intelligent. We call this Data Fuel.
Most Nordic companies sit on data dumps. That is crude oil — valuable, but in a form that won't fly a plane. To turn it into jet fuel it needs to be refined through vectorisation, which transforms raw data into a semantic space where AI doesn't just find information but understands its meaning.
The technical enabler is a Cross-Cloud Lakehouse mindset, where data flows freely across cloud boundaries while staying tightly governed. Tools like BigQuery and Managed Lustre feed data to processors up to ten times faster than before — vital for millisecond decisions.
| Step | Description | Technology |
|---|---|---|
| Capture | Multimodal data compilation | Cross-Cloud Lakehouse |
| Vectorisation | Conversion to semantic understanding | Gemini Embedding 2 |
| Processing | Real-time refinement and enrichment | Vertex AI & BigQuery |
| Activation | Data fed to agents for decisions | TPU Direct Storage |
Pillar E — Ecosystem of Agents: a 50-champion swarm
Forget the illusion of one omniscient AI — that is a rigid monolith. The SDE architecture builds an Ecosystem of Agents: a swarm of around 50 specialised agents, each a master of their narrow field. One handles currency risk, another optimises packaging materials, a third nurtures brand tone.
They communicate using Agent-to-Agent (A2A) protocols and Model Context Protocol (MCP) standards.
The communication topology shapes the system:
- Mesh (Peer-to-Peer) — direct agent-to-agent communication, maximum flexibility.
- Hub-Spoke (Star) — a central supervisor controls task distribution, critical for auditability.
- Hierarchical (Tree) — manager agents oversee subordinate agents.
Silicon-to-Software: the technological foundation
The SDE vision requires deep integration between silicon and software. Eighth-generation Ironwood TPU chips are designed for the agent era: TPU 8t for massive pre-training and embedding management, TPU 8i for inference and agent-to-agent communication, with a Collectives Acceleration Engine that cuts latency 5×. Together with the Gemini 3 family, complex reasoning and long contexts no longer limit the depth of autonomy.
Strategic questions: the breakdown of market structures
- When the customer no longer makes the purchasing decision — if agent negotiates with agent in milliseconds, brand must be replaced with algorithmic credibility.
- From supply chain to self-directed network — a dynamic Agentic Supply Web that absorbs global disruption without human guidance.
- The death of product-centricity — Christensen's Jobs to be Done taken to its conclusion: we sell autonomously realised outcomes.
- Trust as code — encode mutual trust into agent contracts. A transparent Purple Cow.
- Competition between nervous systems — winners have the most flexible architecture and the fastest learning cycle, not the most data.
- Autonomous JTBD ecosystem — the company becomes a concierge that reads needs before customers notice them.
Scaling and the new role of humans
The journey toward 50/50 mode — half of operational activity fully autonomous — does not mean replacing people. It frees them from Inbox Dread.
The leader becomes an orchestrator who regulates the system's Creativity Temperature. That requires Brené Brown's vulnerability: the courage to admit you don't control every technical step but trust the architecture and the transparency it provides.
Autonomy requires rigorous architecture. Data must be refined into fuel, intelligence distributed to specialised agents, and leadership delivered through a real-time Steering Wheel and Shield protection.
Sketch your S/D/E on a single page. If any pillar is missing, you don't yet have an SDE — you have a pilot.