For leadership teams

Scale AI across the organisation without losing control.

Turn scattered pilots into a governed rollout you can track and stand behind. HAR Works helps leadership teams of mid-size and large Nordic companies decide, decision by decision, what people do and what AI agents do.

  • A ranked list of decision paths, with what autonomy costs on each
  • A shared language for business, IT and HR
  • A named person accountable for every path
Example Decision portfolio · commercial function
  1. Discount on a proposal

    How often
    Daily
    Same way every time
    Partly
    What an error costs
    Material, noticed late

    Owner: Sales Director

    Supervision mode: Approve

  2. Account priorities for the week

    How often
    Weekly
    Same way every time
    Mostly rules
    What an error costs
    Costs time, shows within the week

    Owner: Head of Sales

    Supervision mode: Approve

  3. Routing inbound leads

    How often
    Continuous
    Same way every time
    Nearly always
    What an error costs
    Cheap, needs watching

    Owner: Sales Operations Manager

    Supervision mode: Monitor

  4. Meeting notes into the CRM

    How often
    Daily
    Same way every time
    Mostly rules
    What an error costs
    Cheap, visible at once

    Owner: Sales Team Lead

    Supervision mode: Audit

  5. Non-standard contract terms

    How often
    Monthly
    Same way every time
    Tacit judgement
    What an error costs
    Severe, noticed late

    Owner: Commercial Director

    Supervision mode: Collaborate

Ranked by value. The supervision mode follows from what an error costs.

Six vantage points

The same problem, seen from six seats.

Agentic AI reaches many parts of the organisation at once. Each leader sees a different part of the same gap: the work between people and agents has not been designed.

  • From the business

    Pilots stall before they reach the work

    The demo worked, and the work looks the same. Personal productivity rose while the organisation’s results stayed where they were. This is the wall between experiments and measured business impact.

  • From the leadership table

    The AI strategy stays above the work

    Goals and principles are approved, and the question is where to start. The strategy has no owner, no decision paths and no order of work, and it loses credibility with every month without a visible first result.

  • From the people who supervise

    “Human in the loop” turns into a click

    A person is placed after the agent, whatever is at stake. That says a human is involved and leaves their role open. Approving one agent’s output is a task; approving ten is a job, and supervision drifts into a rubber stamp.

  • From the vendor meeting

    Vendors arrive with agents

    The demo is convincing and the contract is close. The tool then decides how the work is divided, and the cost of errors and supervision stays with you. Changes are cheapest before you sign.

  • From risk and audit

    Who answers for the agent is unclear

    Agents from several vendors run in different units. How many there are, what they can reach and who can stop them is hard to say, and audit, customers and the EU AI Act ask for the record.

  • From HR

    Roles change faster than they are planned

    Work moves to agents one tool at a time. Roles, skills and workload change without design, and supervision becomes work without time set aside for it.

Where you stand

Five levels of AI adoption, with a wall between 2 and 3.

Pilots and shared tools sit at levels 1 and 2. Crossing the wall takes one business problem solved end to end, with a named owner and a measured result.

The HAR adoption model: five levels, with a wall between level 2 and level 3.
  1. Level 1: Individuals experiment
  2. Level 2: The organisation learns from shared tools
  3. The wall: Don’t Scale Chaos
  4. Level 3: Measured business impact
  5. Level 4: Business impact scales across functions
  6. Level 5: Impact becomes self-reinforcing
  • Personal productivity Levels 1–2
  • Organisational productivity Levels 3–5

Find your level in a 30-minute maturity check

How it works

Two loops, with control.

An operating model for humans and agents runs as two loops that feed each other, held together by governance.

The outer loop decides the operating model, and the inner loop builds and runs the agents.

Outer loop

The operating model (HAR)

  1. Discovery
  2. Design
  3. Scale
  4. Reassess

Decides who does which work under which control: person, agent or plain automation, a supervision mode and a named owner for every decision path that matters. Reassess keeps the operating model current as the market and the environment change.

Inner loop

The agents

  1. Build
  2. Run

Builds and runs the agents the design calls for and measures them in use. We build the first agents with you to prove a path; production builds and Run go to your team or an engineering partner you choose.

Governance

Control follows consequence.

The supervision mode of each decision path is chosen from what an error costs, from Command to Audit.

  1. Command

    A person directs each instance, and the agent carries out the intent.

    High attention per instance

  2. Collaborate

    A person works each instance together with the agent.

    High attention per instance

  3. Approve

    A person signs off each output before it takes effect.

    Moderate attention per instance

  4. Monitor

    A person watches the flow and the exceptions it raises.

    Light attention

  5. Audit

    A person samples completed work after the fact.

    Light attention

If no accountability anchor can be named, the decision is not ready for an agent.

  • A named person for every path

    Each decision path has an accountability anchor, a named person who answers for its outcomes and can stop the agent.

  • Drift caught in review

    Reassess reviews look for supervision that has loosened without anyone deciding it, and set it right.

  • A record audit can rely on

    Supervision modes, owners and gate decisions are kept on record with their reasons, for audit, customers and the EU AI Act.

Governance in the method

HAR thinking

A shared language for work between people and agents.

Human Agent Relationship is a way to think about how people and AI agents divide work, decisions and accountability. It grows in the open: in the book, on LinkedIn and on YouTube.

  • The book

    Ihmisten ja agenttien organisaatio by Tapio Nissilä and Niklas Nordling (Professional Publishing Finland, 2026, in Finnish)

    How to design the operating model of an organisation of people and agents.

    About the book

  • LinkedIn

    HAR – Human Agent Relationship: short posts on dividing work between people and agents.

    Follow on LinkedIn

HAR thinking and the nine beliefs

Partners

Domain expertise makes agents work.

What to hand to an agent, what an error costs, which rules apply and what data exists are questions of the domain. Domain partners bring that knowledge and the client’s trust; HAR Works brings the operating model and governance.

Six domains

  • Marketing
  • Sales
  • Customer service
  • Finance
  • HR
  • Procurement and supply

Partner with HAR Works

Where to start

Start where you are.

Each situation has its own first step, with its duration stated up front.

Where to start, by situation
Your situation Start with Duration
We have pilots and we’re unsure where we stand Maturity check 30 minutes
The direction is still open Executive briefing ½–1 day
The leadership team sees AI from different angles Executive briefing ½–1 day
A pilot has stalled Commercial Function Sprint 4–6 weeks
A function has been chosen Commercial Function Sprint 4–6 weeks
Copilot is rolled out, and the work looks the same Commercial Function Sprint 4–6 weeks
A decision path has been chosen Decision path design Agreed for each path before work starts
Agents are already live Living operating model Ongoing
Supervision has loosened and nobody decided it Living operating model Ongoing

All engagements and their pricing models

Evidence

What we can show today.

  • The HAR method at work

    A European B2B software company maps its expansion sales decisions in one day

    10 decision points identified and defined, trigger to outcome

    of the ten, by who does the work
    • 4 to the agent
    • 3 kept with people
    • 3 deferred as their own piece of work

    5 of the ten, pinned down concretely enough to become agent specifications

    Read the case

  • Sample Discovery · Invented company

    What a Discovery produces, shown on a consumer telecom business

    The analysis separates what is known from what is argued. Every finding carries one of three labels, and the document ends with the questions the leadership team should answer next.

    See what it contains

  • The book behind the method

    The method in book form

    The HAR method is set out in a book on how to design the operating model of an organisation of people and agents. The book is written in Finnish.

    About the book

Start with a 30‑minute maturity check.

We place your company on the five-level adoption scale, match your situation to a typical buying trigger and recommend one next step. No charge, and no obligation to continue.