The Holak Scale has a section on the two dimensions: individual and organisation. It was one of the most-cited parts after publication - because everyone sees the gap between themselves and their employer.

This article expands on it. For managers - how to actually measure your organisation. For engineers at 9 in a company at 2 - how not to burn out. For AI ambassadors - how to run conversations with compliance and the board.

Why the gap appears

Two systems learn at different speeds:

  • The individual learns in 2-6 weeks. Custom instructions, CLAUDE.md, MCP - tools are public, tutorials are free, feedback is instant.
  • The organisation learns in 12-24 months. Every new stack needs security review, RBAC, data policy, training, change management, budget.

The gap is natural. The problem starts when both sides pretend it isn’t there.

Three typical patterns

Pattern A: individual 9, organisation 2

The engineer runs autonomous agents on pet projects. At work they fire up ChatGPT in a vanilla window because internal tooling doesn’t allow anything else. “Compliance hasn’t approved.”

Cost: wasted talent. The engineer loses motivation, leaves within a year for a company at 5.

Signal: conversations like “that won’t fly here”, “maybe in two years”. Silence in retros around “how we work.”

Pattern B: individual 3, organisation 7

The company rolled out AGENTS.md, stood up MCP servers, hired an AI ambassador. But the average engineer still copies prompts from Confluence - because they don’t know skills built by the platform team exist.

Cost: wasted infrastructure. ROI drops to 10% because only 5% of the team uses it.

Signal: internal dashboards show high token consumption from a small handful of users. “We have great tools, just nobody uses them.”

Pattern C: individual 5, organisation 5 (aligned)

A rare, healthy state. The company invests in education, engineers have a voice in tooling, the weekly AI sync isn’t a hype demo - it’s a reflection on “what’s working.”

Signal: retros include “does our CLAUDE.md / AGENTS.md match practice?”

How to compute the organisational median

Most common mistake: measuring only leaders. “Our AI ambassador is at 9, so the company is at 9.” No. The company is at the median of the sample.

Method:

  1. Sample. Randomly pick 15-30 people across all org levels. Not volunteers - random.
  2. Diagnosis. Use the 30-minute protocol per person. Yes, that’s 7-15 hours total.
  3. Distribution. Record all levels in a sheet. Histogram. Median, not average.
  4. Spread. Median + min + max + p25 and p75. Spread shows whether the company is homogenous.

Red flags in the results:

  • Spread > 5 (e.g., min 1, max 9) - divided company, polarisation risk
  • Median < 3 - most of the team is at chat level
  • p25 = 0 or 1 - every fourth person is in resistance, check whether it’s a choice or fear

What a level-9 engineer in a level-2 company does

A common real-world situation. Five options, from least to most radical.

Option 1: safe channels

Use what the company allows in maximum scope. Custom instructions in the tool you have officially. Context files in a private branch (not the prod repo). Skills locally.

Limit: you’ll get to 4-5 at work. Full 9 stays in pet projects.

Option 2: be a greyhound

Propose a pilot to leadership. 1 team, 1 project, 3 months, defined success. Show the numbers. Win the pilot - and expand.

Limit: requires political energy. Works if you have a director-level sponsor.

Option 3: an education channel

Brown bag weekly, internal newsletter, demo days. Don’t push tools - show value. After a few months adoption grows organically.

Limit: slow, months until impact. Requires patience.

Option 4: AI ambassador role

Negotiate a formal role with your manager. 20% time on adoption, measurable goals, tooling budget. Requires political capital.

Limit: the role only exists in companies that commit to AI. Others won’t even let you define it.

Option 5: change jobs

Sounds radical, but in 2026 it’s a real option. Level 5-6 companies actively recruit from level-2 ones. Market check is simple: if your job regularly blocks you from tools juniors at other companies are learning - it’s time to leave.

Limit: personal decision, not tactical.

What a manager in a level-7 company with median 3 does

The other side: great infrastructure, nobody uses it.

What works:

  1. Internal champions. Pick 1-2 people per team, give them 10% time to be “guides.” Measure usage growth in their team.
  2. Buddy programs. Pair a new user with a champion for 2 weeks. Critical metric: prompts/day before and after.
  3. Use case library. Specific short recordings “how Iwona saves 45 minutes a week on X.” Not generic “AI boosts productivity.”
  4. Defaults. Team-level custom instructions pre-defined. Skills visible in the UI. CLAUDE.md in every repo with a template to fill in.

What does NOT work:

  • A leadership mandate - “everyone must use it.” Increases resistance.
  • A single training - training doesn’t change levels, behaviour does.
  • A dashboard with numbers and no context - “we burned 5M tokens” means nothing.

Governance - who decides

Three decisions that need an owner:

  1. What can be deployed (tools, MCP, agents). Owner: CISO + Head of Engineering.
  2. What is forbidden (data class, production actions). Owner: Compliance + DPO.
  3. What is measured (levels, usage, ROI). Owner: AI ambassador / Head of Productivity.

Without an owner per category, every new MCP triggers a 3-week from-scratch discussion. With an owner - decisions in hours.

What’s next

Measure the organisation quarterly. Three diagnoses a year show the trajectory. Trajectory matters more than a point estimate.

If you’re a level-9 engineer in a level-2 company - pick one of the five options. Don’t try all at once. Measure the result after 3 months. Adapt.

If you’re a manager of a level-7 company with median 3 - pick two champions. Give them 10% time. Measure delta usage in their teams after a quarter.

Everything else is theory nobody remembers in production.