06 · Team

People

Roles rewritten so everyone knows what AI does, who checks, and who is responsible.

Stage: Operate and improve

Your problem

Tools got added; job descriptions did not. When an AI-assisted task goes wrong, nobody owns the error.

Adoption stalls because staff do not know what is safe to hand over — and fear what handing over means.

Why it matters

Field research shows the gains and the damage both come from the boundary: people over-trust AI on tasks outside its capability. Written hand-offs — AI does, human checks, human only — are how a small business makes the boundary explicit and keeps responsibility with people.

Where this does not apply: A one-person business has no hand-offs to write — but the money and risk limits still apply to you.

Before, then with AI

One role gets a clear human–AI job description

A business that added AI tools but never rewrote who is responsible for what.

Illustrative example — not a live customer result

Before

  1. Tools were added, but job descriptions stayed the same

    Owner

  2. When an AI-assisted task goes wrong, nobody owns the error

    Staff

  3. Adoption stalls because no one knows what is safe to hand over

    Staff

With AI

  1. Pick one role and mark each task: AI does · Human checks · Human only

    Human only

    Output: One-page role card

  2. Set money and risk limits AI cannot cross alone

    Human only

    Output: Written limits

  3. Run one scenario drill and fix the hand-offs that failed

    Human reviews

    Output: Updated role card

Where this fails: AI must not decide hiring, pay or dismissal. This area is about clarity of responsibility, not automating people decisions.

Flagship use case

Rewrite one role with clear AI / review / human-only lines

For businesses that added AI tools without rewriting responsibility: pick one role, mark every task as AI does · Human checks · Human only, set hard limits, and drill one scenario before relying on it.

Open the full use case →

Other parts of this area

  • Redesign roles
  • AI basics for everyone
  • Onboard faster
  • Measure the new way

How to start

  1. Pick the role AI already touches most

    Manager · Chosen role

  2. Draft the task split in a small working session

    Manager + role holder · Draft role card

  3. Set limits AI cannot cross alone

    Owner · Written limits

  4. A person signs off the final card; AI never decides people matters

    Manager · Approved card

  5. Drill one scenario; every failure updates the card

    Manager · Updated card + review date

Tools to look at first

Can your current tools do this? These are category candidates, not a recommendation to buy.

Notion + Notion AI

Knowledge base

Good for: A structured Company Brain with owners and last-updated dates · Role cards, rules pages and checklists in one place

Not for: Heavy document/email workflows · Financial records

Works with: Works alongside either Microsoft or Google stacks as the knowledge layer.

Official pricing · Data handling

Alternatives: SharePoint, Google Drive + Docs · Disclosure: No commercial relationship. Category candidate, not a verified fit for every business. · Reviewed 2026-10-19

Microsoft 365 + Copilot

Documents & knowledge

Good for: Company knowledge in SharePoint/OneDrive with permissions · Drafting in Word/Outlook where your files already live

Not for: Businesses not already in the Microsoft ecosystem · Accounting-specific workflows

Works with: Best if your business already runs on Outlook, Word and Teams.

Official pricing · Data handling

Alternatives: Google Workspace, Notion · Disclosure: No commercial relationship. Category candidate, not a verified fit for every business. · Reviewed 2026-10-19

Google Workspace + Gemini

Documents & knowledge

Good for: Shared Drive as a lightweight Company Brain · Drafting in Gmail/Docs where your team already works

Not for: Deep Microsoft-ecosystem shops · Line-of-business accounting tasks

Works with: Best if your business already runs on Gmail, Docs and Drive.

Official pricing · Data handling

Alternatives: Microsoft 365, Notion · Disclosure: No commercial relationship. Category candidate, not a verified fit for every business. · Reviewed 2026-10-19

Evidence

External research

AI has a jagged frontier — know which side your task is on

Consultants using AI did significantly better on tasks inside the model’s capability and worse on tasks outside it. Choosing the right task matters more than choosing the fanciest tool.

What this does not show: Professional consultants on controlled tasks; your jobs need their own small trial.

Source: Harvard Business School — Navigating the jagged technological frontier (field study with BCG consultants) · Independent academic study; subjects were professional consultants, not SME owners. · Reviewed 2026-10-19

External research

Klarna’s two-sided automation lesson

Klarna first announced its AI assistant handled a large share of service chats; later reporting covered the company re-expanding human service after quality concerns. Automation volume and service quality must be read together.

What this does not show: A large fintech, both sides partly company-narrated. The takeaway is the need for human escalation paths, not a chat-volume target.

Source: Bloomberg — Klarna turns from AI to real-person customer service · Independent press coverage; shows the limits of the earlier vendor claim. · Reviewed 2026-10-19

Do one thing

Do one thing: write one role card

0/5 done · saved on this device

The prompt includes placeholders — never paste customer personal data or company secrets into a tool that has not passed your data check.

Next

What would make this useful for your business?

Ask a question or report a factual issue. Mason reads every submission; it shapes what we research and publish next.

Put AI to Work With Clear Human Roles · Boteam