Capability 01 of 04 · Organization
AI Adoption Model
How your organization rolls out AI tools - from individual experiments to org-wide strategy.
5
maturity levels
15
practices
15
guides
L1 · Stage 01Assisted
Practices - what it looks like3 guides
- Adoption via bulk license purchaseThe big-bang license purchase is the most common first move in enterprise AI adoption, and the most reliable predictor of failure.guide→
- Initial enthusiasm fades to low usageEnthusiasm → Silence → Shelfware is the name for the failure arc that almost every unstructured AI tool deployment follows.guide→
- Powerful tools on an unprepared processThe Ferrari Engine in Fiat 126p is a metaphor for a specific and common failure pattern: installing powerful AI capability into an engineering process that cannot take advantage of it.guide→
Criteria - what to measure
- 01AI tools have been adopted (licenses acquired)
- 02Adoption is tracked informally
- 03At least some developers are experimenting with AI tools
- 04Organization has not banned AI tool usage outright
L2 · Stage 02Delegated
Practices - what it looks like3 guides
- Pilot teams (2-3 teams)A structured pilot is the antidote to the big-bang license deployment.guide→
- Adoption spreads through peer networks rather than mandatesAdoption spreads through peer networks rather than mandates: practice travels between people who work together, along the relationships that already exist, and a directive from above does not move it.guide→
- Pilot metrics; track cost-per-merged-PR from day onePilot metrics are the set of measurements you define before a pilot starts that determine whether the pilot succeeded and whether to expand.guide→
Criteria - what to measure
- 012-3 pilot teams are designated with explicit AI adoption goals
- 02Adoption spreads through peer advocacy rather than a mandate, and the org can show where it spread
- 03Pilot metrics are defined and tracked (adoption rate, usage frequency, developer satisfaction)
- 04Pilot results are shared with the broader organization
- 05Champion has direct access to leadership for escalation
L3 · Stage 03Systematic
Practices - what it looks like3 guides
- Platform team owns AI tooling: a central proxy or agent registry supplying identity, cost tracking, sandboxing, observability and evals by default - declare an agent once, get production-readiness in minutes rather than weeksA central proxy or agent registry hands out identity, cost tracking, sandboxing, observability and evals by default, so declaring an agent takes minutes rather than weeks.guide→
- Internal Developer Platform with AI layer, plus enablement run as a named programme with attendance you can count (a standing guild, guided hackathons, hands-on labs) rather than a launch emailThe self-service platform teams already deploy through gains an AI layer, and the enablement beside it runs as a named programme with attendance you can count rather than a launch email.guide→
- Standardized agent setup per team, but no mandated tool - the platform is standard, the choice is free, and outcomes are what get measured; "bad day protocol" for model and harness regressions, plus a vendor-exit planEvery team starts from the same agent baseline - context injection, permissions, monitoring - but no tool is mandated: the platform is standard, the choice is free, and outcomes are what get measured.guide→
Criteria - what to measure
- 01Platform team formally owns AI tooling (selection, provisioning, security, baseline configuration)
- 02Internal Developer Platform includes an AI layer (standardized agent setup, self-service provisioning)
- 03Standardized agent setup exists per team (every team has a working AI environment by default)
- 04New developer onboarding includes AI tool setup that completes in under 30 minutes
- 05Platform team tracks adoption breadth (% of developers with active AI setup)
L4 · Stage 04GovernedMost teams aim here
Practices - what it looks like3 guides
- AI-assisted work is the default path rather than an initiative, and the org advances by removing its next bottleneck rather than by buying tokensAn AI-first development culture is one where agents are the default approach to development tasks, not an option that some developers use sometimes.guide→
- Agent fleet management as disciplineOnce a developer runs several agents at once, start one and check back stops working: agents need scheduling, monitoring, failure handling and quotas.guide→
- Developer supervises agents rather than authoring most changesIn Steve Yegge's model of AI adoption stages, Stages 6 and 7 represent a fundamental shift in what a developer does.guide→
Criteria - what to measure
- 01AI-assisted work is the default path, not an initiative: a large and stable majority of developers use agents daily
- 02Agent fleet management is a recognized discipline with defined practices
- 03Developer role has shifted toward supervising agents rather than authoring most changes
- 04"Span of control" metric is tracked (how many agents a developer can effectively supervise)
- 05Organization benchmarks its adoption against external reference data rather than against its own launch week
L5 · Stage 05Self-improving
Practices - what it looks like3 guides
- Centralized agent orchestration: scheduling, placement and lifecycle handled by a platform, not per teamA central orchestration layer schedules, scales and recovers agent workloads across the organization, the way Kubernetes does for containers.guide→
- Human-at-the-wheel, not human-in-the-loop"Human-in-the-loop" describes an approval model where humans review and approve individual agent actions before they execute.guide→
- Organization optimized for agent throughput, not human throughputOrganizations are designed around assumptions about how work gets done.guide→
Criteria - what to measure
- 01Centralized agent orchestration system exists ("Kubernetes for agents")
- 02Developer role is "human-at-the-wheel" (strategic direction, not task-level involvement)
- 03Organization is optimized for agent throughput, not human throughput (meetings, processes, tooling all agent-aware)
- 04Agent orchestration system handles scheduling, resource allocation, and failure recovery
- 05Organization measures agent utilization as a key infrastructure metric
From the Field
Recent releases, projects and discussions the AI Radar classified into this capability.
article
The Tokenpocalypse Is Here: Companies Are Scrambling To Stop Spending So Much on AI
discovered
An orchestration runtime for multi-agent AI systems. Declare agents, tools, and policies as YAML; Orloj schedules, executes, routes, and governs them for production-grade
discovered
Open-source workspace protocol for AI agent systems. If OSA / Claude Code is the employee, Canopy is the office.
discovered
Your AI agent skills, finally organized. A macOS app to browse, edit, and manage skills across Claude Code, Cursor, Codex, Windsurf, and Amp.
release
crewAIInc/crewAI
article
Kelsey Hightower: What the AI Hype Machine Won't Tell You
article
QCon London 2026: Team Topologies as the ‘Infrastructure for Agency’ with AI
article
Codex now offers more flexible pricing for teams
Which level is your team at in AI Adoption Model?
The criteria above are what we check in a facilitated assessment. Run it yourself to place this capability, see which gates you have passed, and compare it against the other 3 in Organization.
Start the assessment →