Let's Talk

Human Systems

Capability 03 · Change management · Org design · AI adoption

AI agents now carry out parts of people's work. As they do, people move from doing the work to being accountable for it. We design who does what, who decides, who reviews and who can stop it, so the AI you deploy gets used, trusted and is worth what it costs.

Our Point of View

AI value is ahuman systems question.

The usual story is that AI frees up capacity. In most organizations today it moves capacity instead: into reviewing output, redoing work, reconciling tools and absorbing one change after another.

Whether deployed AI gets used, trusted and scaled depends on people, roles and ownership as much as on the technology. A system that is technically sound can still fail if nobody designed for the people around it. Humans are part of the system, so we design them in from the start, not after launch.

The Evidence

The human bill shows up in the data.

of employees in organizations implementing AI strongly agree it has transformed how work gets doneGallup, May 2026
as likely to say AI transformed their work when their manager actively supports itGallup, May 2026
of organizations have mature governance for agentic AIDeloitte, State of AI in the Enterprise, Apr 2026 (n=3,235)
of the effort behind AI value belongs to people and processes, against 20% for technology and data and 10% for algorithmsBCG, the 10-20-70 principle
Why Agents Change the Question

A copilot assists a person.An agent takes on part of a role.

When AI only helped on request, the question was whether people would use it. Once agents carry out tasks and whole processes, the question becomes who does what, who decides, and who can stop it.

The person's roleAI assistants:Does the work; AI helps on requestAI agents:Supervises, approves or reviews what agents do
The main riskAI assistants:Low or performative adoption, and reworkAI agents:Over-automation, rubber-stamp review, skill erosion and unclear accountability
What to designAI assistants:Fluency, trust, fit with the workflow, manager supportAI agents:Decision rights, review capability, handoffs and role redesign
What to measureAI assistants:Usage, confidence, time savedAI agents:Overrides, escalations, and what each agent may do alone

How much an agent may do alone is a design choice, set workflow by workflow. We raise it only as human oversight catches up.

What We Design

Eight parts ofa human system.

A Human Systems Architecture is the blueprint for one workflow or domain: a design a client can hold, test and change.

01

Work and role design

Which tasks stay with people, which go to agents, and how each role changes as a result.

02

Decision rights

Who decides, who reviews and who can stop it, with one named owner for every workflow.

03

Human and agent handoffs

Handoffs, escalation and override paths, and what a reviewer needs to see to judge the output.

04

Capability and fluency

What people need to know to use AI and judge its output, taught at the point of need.

05

Trust calibration

Relying on AI the right amount: not so little that it goes unused, not so much that errors go unchecked.

06

Change capacity

How much change a team can take on alongside everything else, and what work should stop to make room.

07

Incentives, culture and safety

Whether people can tell the truth about how they use AI, and raise a problem when they see one.

08

Measurement

Human signals beside usage: confidence, rework, overrides and the experience of the people doing the work.

How We Work

In every phase,not after launch.

Human Systems work runs alongside the build, from the first scan to the next workflow. Scope varies: a readiness check inside one AI build, or the full people design around a team working with agents.

  1. 01

    Scan and discovery

    Role interviews, a readiness and change-capacity read, and a map of who touches each workflow.

    → A readiness baseline and a workflow people map

  2. 02

    Design

    Set how much each agent may do alone against the evidence; define decision rights, review points, escalation and stop rights.

    → The Human Systems Architecture

  3. 03

    Build

    Co-design with the people who will use it; give reviewers the information and authority to overrule the AI.

    → Role cards, reviewer guides and handoff specs

  4. 04

    Launch and enablement

    Manager briefings, learning at the point of need, and champions inside the team.

    → An enablement plan and a manager kit

  5. 05

    Measure

    Human signals beside usage: adoption, rework, confidence and overrides.

    → A 30/60/90-day read on adoption

  6. 06

    Expand

    Repeat for the next workflow, and give agents more room only where oversight has caught up.

    → An expansion roadmap

Deliverables

What youcan hold.

  • A readiness baseline, and a map of who touches each workflow
  • A people design for the workflow: roles, decision rights, handoffs and review points, set before the build starts
  • Role cards, reviewer guides, and a decision chart that names people and agents side by side
  • An enablement plan and a manager kit for launch
  • A 30/60/90-day read on adoption, with what the evidence can and cannot support

What it is

  • Design and delivery work inside AI engagements
  • Roles, decisions and handoffs designed around AI workflows
  • Readiness and enablement tied to a business result
  • Capability built inside your team, by working alongside it
  • Applied to our own work at AWSM LABS first

What it is not

  • HR outsourcing or people operations
  • A training vendor selling courses without a result
  • Legal or compliance advice; we work alongside your counsel
  • Advice from a distance
Foundations

Built on provendisciplines.

Human Systems joins four established disciplines and applies them, workflow by workflow, to enterprise AI.

Human systems integration

What it is: From defense and aerospace: plan for people across a system's whole life, from its first requirements. Human costs are system costs, and cheapest to fix early.

Sociotechnical systems

What it is: Design the social and the technical together, around whole tasks and teams, rather than fitting people to a machine afterwards.

Human factors

What it is: Fit tasks, interfaces and information to how people perceive, decide and make mistakes, including automation bias and over-reliance.

Organization design and change

What it is: Structures, incentives, culture and psychological safety decide whether a change sticks.

Let's Talk

Who decides,and who can stop it?

If you are about to put agents into people's work, or your adoption numbers look better than the work feels, let's talk.