Boris writes https://www.linkedin.com/posts/bcherny_steps-of-ai-adoption-activity-7483695059843043328-LBg_

I talk to engineers at other companies every day and hear the same thing: one person is 10x’ing their output with Claude but the rest of the org hasn’t caught up. Watching teams adopt AI, I keep seeing the same 4 steps. I mapped them out here: Steps of AI Adoption https://lnkd.in/ggWDMepq There’s no one right path through the steps. Every team and company is different. But at each step, tokens aren’t enough to move you forward: to get to the next step, you need to find and break down the next set of bottlenecks, and build up the next set of guardrails. In practice that means giving Claude ways to verify its own work end to end. It means enabling auto mode for permissions, defaulting on automated code review and security review, and using interfaces that let you manage multiple agents at once (Agent view in CLI, Desktop app, iOS and Android apps, Tag). To get to higher levels it means /loop, /batch, dynamic workflows, and worktree isolation for subagents. It’s not about a single feature, but rather using the right features with the right guardrails that enable Claude to automate entire classes of work in a way that your team can trust the output. Once your teams are bought in, how do you track it? Usage is worth watching (e.g. a dashboard), but it measures activity, not return. A better question: would you have spent engineering effort on this anyway? If yes, how much and what would it have cost in manual eng-hours? That’s your return. The bigger payoff comes when fixing and maintaining happens in the background and your teams can focus on building. That’s when you start doing things that weren’t even in range before. Anthropic is on step 3 and pushing toward 4. Personally, I just hit level 4. Curious where you are — what step is your team on?

Steps of AI Adoption

Boris ChernyJul 16, 2026

Step & your role	Agents	What it looks like	What’s the bottleneck	Products that help with each step	Guardrails
0: Gated	0	Only older or lighter/faster models are approved, latency compounds through AI gateways and custom auth, no MCP governance, internal access to AI tools is gated or process-heavy. No IT infra or approval path for hosting Claude-created code or artifacts; outputs only exist locally.	Legacy security and approval processes, focuses on cost-per-token containment vs. outcomes, lack of true technical voices in decisionmaking.	Claude.ai chat	SSO/SCIM plus role-based access Org-level budget caps Deploy inside existing approvals/IAM Data governance package
How to get from step 0 to 1: Executive/buyer alignment and escalation of blockers; frameworks for launching Claude securely
1: AssistedYou + an agent (a pair)	~1	One engineer, one agent, mostly supervised—a fast pair programmer. You run one session at a time and review almost every change before it merges. Unlock: A change that used to fill an afternoon becomes something you finish between meetings.	Your attention and the need to inspect each response and code edit. Due to low trust for the model’s output and lack of self-verification, you feel you must read everything, so you never look away. Work is synchronous: you sit and watch while Claude works, rather than moving on to the next task.	Claude Code in the Desktop, CLI, or IDE Claude Cowork, Claude Design Usage via Anthropic API, Bedrock, Vertex, or Microsoft Foundry Claude Code analytics dashboard + Analytics API Compliance API for Claude Enterprise Plan mode to review intent before edits	Per-seat spend caps Centrally managed model/effort settings Centrally managed policy OpenTelemetry export into existing SIEM/observability stack
How to get from step 1 to 2: Run more than one agent at a time; a self-verification loop you trust (tests + build + lint + e2e testing with a real dev environment); auto mode, to avoid blocking permission prompts; automate code review
2: ParallelOrchestrator	~10	One engineer orchestrates 5–10 agents at once, each on its own worktree or git checkout, jumping between them. Claude checks its own work—tests, build, lint, security scan—before you see it. Auto mode is always on. Automated code review and security review are on by default. Output multiplies, you review final diffs rather than keystrokes, and your backlog of maintenance work starts shrinking. Claude writes most of the code. Unlock: A backlog that used to take the team weeks becomes one engineer’s afternoon of orchestration.	Reviewing output. You’re hand-writing less code and instead checking six streams of it, and this takes up more of your time. Prompting and steering the model as you juggle sessions.	Auto mode Agent view Claude Code Review Claude Security Review Claude Code on Mobile, cloud execution in Desktop Usage via Claude Teams or Claude Enterprise Claude Tag (do a single task) Worktree isolation in CLI and Desktop Remote control, so you can monitor your agents from your phone	Analytics to monitor team usage Automatic code quality enforcement: lint, automated tests, typecheck Claude powered end-to-end verification (eg. using the Claude Chrome extension or iOS/Android simulator MCP) Manual code review, code merge, and security review. Hold the same quality bar for human and agent-generated code Pre-approve common safe bash and MCP commands in settings.json
How to get from step 2 to 3: Give Claude a way to pull in context (let Claude read code, wikis, discussions); agency and code review speed (agents may touch code owned by other teams); break up your work into loops and routines; let Claude kick off Claude
3: Supervised autonomyManager of managers (an org tree)	~100	Claude writes all or nearly all of the code. “Did you read the code?” becomes “what context was the model missing and how do we solve it for next time?” Unlock: Claude proactively does work that you would have had to kick off manually before. Maintenance and cleanup that used to wait for someone to find the time now runs continuously in the background.	Trust in the loop and your team’s decision throughput. The agent tree is too deep to babysit and your trap is scaling agent count before the loop has earned widespread trust. Ensuring tokens are used efficiently as usage increases. Requires monitoring (via OTel or Analytics) and a culture that encourages experimentation while controlling costs once internal use cases find PMF. Ask yourself: is this something an engineer would have done?	Subagents with worktree isolation (so parallel agents don’t collide) Routines, /loop, /batch, and /goal to fan out repetitive work Dynamic workflows Claude Tag (have it monitor a channel or data source and kick off tasks proactively)	Automatic code review Automatic security review Agent sandboxing CLAUDE.md and Skills to encode standards Tune Auto mode classifier based on your team’s usage Manage token use with model selection, advisors, LSPs, breaking up CLAUDE.md into lazy Skills
How to get from step 3 to 4: Scaled automation of domain-specific use cases (eg. code migration, fuzzing, feature-building, feedback remediation)
4: AI-nativeVP steering by intent	~1,000+	The loop is fully closed and most agents are kicked off by Claude. Hundreds to thousands of agents run; you steer by intent and monitor by exception. Unlock: The quarter-long migration becomes a workflow you kick off and check on.	Identifying and automating work at scale, and enforcing the right guardrails for each type of work.	Claude Agent SDK to programmatically build and schedule agents Claude Tag (active in most Slack channels, auto-responding to posts)	Cost controls for automation Model selection for automation

Source: https://claude.ai/code/artifact/bfdfaef9-bc62-4dfe-ba9e-c58a26c9accf