Agency Excellence Scale

Five-level scale measuring problem ownership from reporting issues to autonomously solving and shipping solutions.

11 min read

Executive summary

Agency measures how much ownership someone takes over problems — from reporting issues to shipping solutions autonomously
This is a 1-5 scale that is contextual and observable, not a personality trait
High agency (4-5) dramatically reduces management overhead and increases delivery predictability
Most hiring mistakes happen when someone interviews at level 4 but operates at level 2
Use behavioral examples, not resumes, to assess agency accurately

Agency Excellence: The degree of ownership someone takes over problems, ranging from identifying and escalating issues (level 1) to autonomously designing, implementing, and shipping solutions (level 5).

What it includes: Problem ownership, solution design autonomy, execution capability, communication of outcomes.

What it does NOT include: Personality traits (introversion/extroversion), communication style preferences, technical skill depth, or management responsibilities.

Key distinction: Agency is contextual — someone can be high-agency in a familiar domain and low-agency in an unfamiliar one. This is not a fixed personality trait.

Why this matters

Business impact

High-agency talent:

Reduces management overhead — managers spend time on strategy, not hand-holding
Increases delivery predictability — problems get solved before they escalate
Improves client satisfaction — clients see proactive problem-solving, not excuse-making
Enables scale — you can grow revenue without proportionally growing management layers
Improves — clients value autonomous problem-solvers, making high-agency talent easier to place on engagements

Low-agency talent:

Consumes leadership attention — constant need for direction and validation
Creates delivery risk — problems escalate because they wait for permission
Frustrates clients — perception of "just doing tasks, not solving problems"
Blocks scale — requires tight supervision that doesn't scale economically

Cost reality

A level 5 (Solves and ships) engineer at 110k€/year creates less management cost than three level 2 (Identifies causes) engineers at 65k€/year each.

Why: The level 5 requires 2-3 hours of management time per week. The three level 2s require 15-20 hours of management time per week — nearly a full-time manager.

Management Overhead Estimator

Estimate weekly management time and annual cost based on team size and average agency level.

30–100 people

Team size(people)?

Average agency level(1–5)?

Manager hourly cost(€/hr)?

Results

Weekly manager hours

Annual overhead cost

69.120 €

FTE overhead equiv.

0.3

The Scale (1-5)

yaml

canonical_block: competency_scale
version: 1.0.0
concept_id: scale.agency_1_5
updated_date: 2026-02-16

scale_name: "Agency Excellence"
description: "Degree of ownership over problems: from reporting to autonomous solution delivery"
range:
  min: 1
  max: 5

levels:
  - level: 1
    label: "Reports problems"
    description: "Identifies and escalates issues but does not propose solutions"
    observable_signals:
      - "Consistently asks 'what should I do?' when blocked"
      - "Provides status updates without analysis"
      - "Waits for direction before acting, even on routine decisions"
      - "Escalates problems immediately without attempting diagnosis"
    assessment_method: "Behavioral interview questions about blockers and decision-making"
    anti_signals:
      - "Claims to be proactive but always seeks approval"
      - "Describes past work as 'following instructions'"
      - "Cannot give examples of solving problems independently"

  - level: 2
    label: "Identifies causes"
    description: "Analyzes problems to understand root causes but waits for solution direction"
    observable_signals:
      - "Explains why something failed with clear reasoning"
      - "Traces issues to upstream dependencies or config"
      - "Asks clarifying questions about context before proposing action"
      - "Documents problem investigation but waits for next steps"
    assessment_method: "Review past incident write-ups or troubleshooting documentation"
    anti_signals:
      - "Analysis is thorough but never leads to action"
      - "Always ends diagnosis with 'what do you want me to do?'"
      - "Treats problems as interesting puzzles, not things to fix"

  - level: 3
    label: "Proposes options"
    description: "Generates multiple solution approaches with trade-offs and asks for decision"
    observable_signals:
      - "Presents 2-3 approaches with pros/cons of each"
      - "Considers implementation complexity and risk"
      - "Asks 'which approach do you prefer?' after laying out options"
      - "Includes 'do nothing' or 'defer' as explicit options when valid"
    assessment_method: "Work samples showing decision memos or technical proposals"
    anti_signals:
      - "Proposes options but they're not meaningfully different"
      - "Trade-offs are superficial or don't reflect real constraints"
      - "Presents options but clearly wants external validation"

  - level: 4
    label: "Recommends solutions"
    description: "Proposes a specific solution with rationale and asks for approval to proceed"
    observable_signals:
      - "Here's what I recommend and why (not 'what should we do?')"
      - "Anticipates objections and addresses them proactively"
      - "Provides implementation plan with timeline and risks"
      - "Asks 'approved to proceed?' not 'what do you think?'"
    assessment_method: "Reference checks asking about decision quality and autonomy"
    anti_signals:
      - "Recommendations lack conviction or hedge excessively"
      - "Asks for approval on trivial decisions"
      - "Changes recommendation based on first objection"

  - level: 5
    label: "Solves and ships"
    description: "Autonomously solves problems, implements solutions, and communicates outcomes"
    observable_signals:
      - "Here's what I shipped and why (not 'can I ship this?')"
      - "Proactively communicates outcomes and next steps"
      - "Takes ownership of results, including failures"
      - "Operates within understood boundaries without permission loops"
    assessment_method: "Portfolio review of shipped work + peer feedback on ownership"
    anti_signals:
      - "Ships without communication (cowboy coding)"
      - "Autonomous but makes decisions that ignore constraints"
      - "High output but creates integration problems for others"

How it works

The agency progression

Every level responds differently to the same situation. Consider the checkout API slowing down at peak:

Level 1 — Reports: "The checkout API is slow. Can you take a look?"
Level 2 — Identifies causes: "The checkout API is slow — I traced it to slow database queries under load. Not sure what to do next."
Level 3 — Proposes options: "Checkout API is slow. Likely a database query issue. We could add an index, add a caching layer, or rate-limit traffic at peak. Which direction do you want to go?"
Level 4 — Recommends: "Checkout API is slow — it's a missing database index. I recommend we add it tonight in a short deployment window; risk is low and the fix is straightforward. Okay to proceed?"
Level 5 — Ships: "Checkout API was slow — it's fixed. Missing database index. Added it, tested on staging, deployed. Back to normal. I'll monitor through peak tomorrow."

The situation is identical at every level. The management cost is not.

Key mechanism: Information + autonomy

Agency is the product of:

Domain knowledge — understanding the problem space
Constraints awareness — knowing boundaries and trade-offs
Trust earned — track record of good judgment
Permission granted — explicit or implicit authorization to act

Example: A junior engineer might be level 2 on a new codebase but level 5 on a system they built. Agency is contextual.

Example: CaseCo Mid

CaseCo Mid (data & AI consultancy)

A series of client-facing incidents kept requiring senior engineer involvement to close. The underlying issue wasn't the bugs — it was that most engineers handling incidents were operating at L2: strong on diagnosis, but consistently stopping short of a decision. Every incident needed a second person to make the call.

Decision

Map agency levels across the team using the scorecard. Redirect client-facing incident rotation to L4+ engineers only, and use incidents as development data for the L2 engineers — not delivery vehicles.

1Ran a retrospective on six months of client incidents. Found that roughly 70% required a senior to make the final call — even on technically straightforward problems.
2Scored the eight engineers in the incident rotation. Six were consistently at L2: they diagnosed well but always ended with 'what do you want me to do?'
3Reassigned client-facing incidents to the two L4 engineers. The L2 engineers continued to contribute — handling investigation, writing post-mortems — but without carrying client accountability.
4Set quarterly agency check-ins. Within two quarters, two engineers had reached L3 and were handling minor incidents independently.

Outcome

Senior management time on incidents dropped from roughly 3 hours per week per senior to under 1. Two L2→L3 promotions freed up rotation capacity. Client satisfaction improved — not because incidents stopped, but because response became faster and more decisive.

Key insight: Level 5 doesn't mean "never ask permission." It means understanding when to ask vs. when to act.

Action: Agency Assessment Scorecard

Use this scorecard during interviews or performance reviews:

Behavior	L1	L2	L3	L4	L5	Evidence
Identifies problems	✓	✓	✓	✓	✓	Ask: "Tell me about a recent bug you found"
Analyzes root causes	—	✓	✓	✓	✓	Ask: "How did you diagnose it?"
Proposes multiple options	—	—	✓	✓	✓	Ask: "What alternatives did you consider?"
Recommends specific solution	—	—	—	✓	✓	Ask: "What did you recommend and why?"
Ships autonomously	—	—	—	—	✓	Ask: "Did you need approval to proceed?"

How to use:

Ask behavioral questions (past examples, not hypotheticals)
Mark highest consistent level observed
Look for anti-signals (claims vs. reality mismatches)
Validate with track record and reference checks for levels 4-5

Effective interview questions:

"Tell me about a recent problem you ran into. What happened, and what did you do with it?" — listen for whether they report, diagnose, propose, or fix
"What did you do once you understood what was wrong?" — tests whether they stop at diagnosis or push through to action
"Did you need anyone's approval to move forward?" — reveals the actual agency ceiling
"Tell me about a time you fixed something without being asked." — tests whether autonomy is claimed or real

For levels 4-5, use in reference checks:

"How much direction did this person typically need on problems within their domain?"
"Can you give me a concrete example of them resolving something without being told how?"

Evidence by level:

Levels 1-3 can be validated by someone who has worked closely with the person day-to-day — a colleague, a project lead, a peer. A credible first-hand account that "yes, this is how they typically behave" is enough. Formal evidence isn't required.

Levels 4-5 require a track record. The evidence builds over time through having owned difficult or high-stakes work repeatedly, without major incidents, and with visible autonomy. Reference checks at this stage serve as confirmation — you're triangulating a picture the track record already suggests, not building one from scratch.

Pitfalls

Confusing agency with extroversion or visibility

High risk

When quiet engineers who ship consistently are assessed as lower-agency than talkative ones who frequently escalate.

Impact

Low-agency extroverts get promoted into roles requiring autonomy they don't have. High-agency introverts are under-utilised and eventually leave.

Hiring for claimed agency in interviews instead of demonstrated agency

High risk

When interview answers about autonomous problem-solving are taken at face value without reference validation.

Impact

Hire interviews at level 4 but operates at level 2. High management overhead, slow delivery, client-facing incidents that require COO involvement.

Expecting level 5 agency in unfamiliar contexts

High risk

When a new hire who operated at level 5 in a previous company needs 2-3 months of ramp time and is penalised for it.

Impact

Conflict between expectation and reality. Manager loses confidence in the hire. Hire becomes cautious and drops to level 3 behaviour permanently.

Punishing autonomous action when it fails

High risk

When an engineer ships a fix that causes a regression and the response is 'you should have asked first' rather than 'let's improve the testing process.'

Impact

Agency drops across the team. Engineers start over-escalating to protect themselves. Management overhead increases as a direct result of the response to one failure.

Technical Scale — Evaluate technical depth separately from agency
Business Scale — Understand business context awareness
Competency Model — See how agency combines with technical + business
Talent Readiness — Use agency as a readiness dimension
Assessment Framework Tool — Full assessment templates and interview guides

What decisions this enables

Which engineers are safe to assign to client-facing incidents without manager involvement
Whether a team's management overhead is a people problem or an agency-level mismatch
How to structure an upskill programme for engineers who are technically strong but low-agency
When to accept a candidate who interviews well but references flag as needing direction
How many senior engineers a team actually needs versus how many it could replace with higher-agency mid-levels

Agency Excellence Scale

Executive summary

Definitions

Why this matters

Business impact

Cost reality

Management Overhead Estimator

The Scale (1-5)

How it works

The agency progression

Key mechanism: Information + autonomy

Example: CaseCo Mid

Action: Agency Assessment Scorecard

Pitfalls

Confusing agency with extroversion or visibility

Hiring for claimed agency in interviews instead of demonstrated agency

Expecting level 5 agency in unfamiliar contexts

Punishing autonomous action when it fails

Next

What decisions this enables

FAQs