Key Takeaways

  • 90% of large enterprises have purchased enterprise licenses for frontier models like Claude or ChatGPT, but buying seats does not mean workflows change.
  • Enablement lags raw access: about 60% of companies run structured internal education to show staff how to prompt, build workflows, and apply models.
  • Engineering teams lead agent deployment at 50%, yet half of those firms run setups so primitive they barely qualify as coding agents.
  • The diffusion cliff sits at non-engineering back-office operations, where production agent usage crashes to just 5%.
  • Alex Lieberman captures this drop-off with Lieberman's 4-Milestone Enterprise AI Adoption Framework.

Lieberman's 4-Milestone Enterprise AI Adoption Framework

Alex Lieberman models enterprise adoption as a race with four clear checkpoints. Most companies stall early because buying software licenses requires zero operational change.

  • Milestone 1: General-Purpose LLM Rollout (~90% Adoption): Deploy basic enterprise access to frontier general-purpose LLMs (such as Claude or ChatGPT) across corporate employee seats.
  • Milestone 2: Training and Enablement (~60% Adoption): Provide structured internal education teaching employees specific model capabilities, interface features, prompt skills, and practical workflows rather than expecting them to figure it out independently.
  • Milestone 3: Engineering Coding Agents (~50% Adoption): Equip software engineering and product teams with dedicated AI coding agents to automate code generation and technical implementation.
  • Milestone 4: Production Agents for Non-Engineering Work (~5% Adoption): Deploy autonomous background agents into live production environments to handle repeatable, non-engineering back-office processes across business functions.

Lieberman notes that enterprise leaders are often stunned by these figures: “I think especially at AIE when I tell people about what percentage of companies have reached these milestones, they're kind of mind-blown by it.”

The gap between Milestone 1 and Milestone 2 reveals a common corporate failure: IT buys thousands of seats and assumes employees will teach themselves. Lieberman argues that enablement “never gets the love that it should get but is arguably more important or as important as the technology itself.”

By Milestone 3, half the market drops out. Even within the 50% using coding tools, Lieberman observes that “every other enterprise that we speak to either hasn't rolled out a coding agent or what they are using is closer to not having a coding agent than having a coding agent.”

The real cliff arrives at Milestone 4. Software engineers have clear test suites, pull request reviews, and Git version control to catch agent errors. Back-office teams in finance, compliance, legal, and operations lack automated guardrails. Because their business processes are rarely standardized or written down, autonomous execution stalls out. As Lieberman puts it, “this is where the biggest diffusion gap is from raw intelligence to enterprise usage. Agents in production for non-engineering work drops to like 5%.”

When This Works (and When It Doesn't)

This framework works when diagnosing why an enterprise software rollout failed to produce measurable operational returns. Consultancies and internal champions can pinpoint whether a company lacks raw tooling, employee skills, or process definitions.

It breaks down for early-stage startups and small agencies. A twenty-person team does not need formal enablement tracks to push agents into production. Startups also skip directly from Milestone 1 to Milestone 4, orchestrating autonomous scripts to handle customer intake or billing without buying enterprise coding agents. The four-milestone sequence assumes legacy corporate hierarchies where legal, procurement, and middle managers slow diffusion across non-technical departments.

What to Do With This

Audit your company against the four milestones before buying another SaaS tool. If you run a startup or lead an internal team, list every back-office task your operations staff handles manually three or more times a week, from invoice reconciliation to customer onboarding triage.

Pick one process that already has written rules and an unambiguous outcome. Run an agent against that exact task with human review for five days, measure error rates, and promote it to autonomous production once accuracy matches human staff. Do not spend budget on company-wide prompts or broad subscriptions until your team turns one discrete back-office task into an automated, working background agent.