Questions we answer

Can we trust AI to act on its own?Governance and guardrailsWho allowed the AI to do that?Identity travels with every requestCan someone talk the AI into something risky?Risk-graded autonomyCan we see who did what, and why?One thread from request to resultWhere are the keys kept?One vault, no shared secretsWhy did the AI bill jump, and who spent it?One counted doorwayIs every task getting the right model?Right model, right taskWhat if our AI vendor goes down or raises its price?No single point of supplyAre we paying for tools that do the same job?One tool per jobDoes the AI forget everything between conversations?Governed shared memoryIs this producing real work, or just answers?The digital factoryWhich work should run itself?Managed, not scriptedCan we get more from the Copilot licences we already pay for?Copilot proposes, the platform verifiesWill people actually use it?Meet people where they workHow much time do people lose sorting inbound information?Intake to insight, untouchedCan the same platform run a real business?One governed foundation for every productWhat happens when it breaks at night?Detect, fix, escalateWhere is the list of what is open, and who owns it?One list, closed with evidenceCould we recover if we lost everything?Proven by restoringCan you prove it to an auditor?Challenge before acceptanceDoes it get better over time, or just older?Every lesson becomes a ruleWill this still work after the next upgrade?Clean Core for AIDo our documents match what is actually running?Documentation as a by-productCan we trust AI to act on its own?Governance and guardrailsWho allowed the AI to do that?Identity travels with every requestCan someone talk the AI into something risky?Risk-graded autonomyCan we see who did what, and why?One thread from request to resultWhere are the keys kept?One vault, no shared secretsWhy did the AI bill jump, and who spent it?One counted doorwayIs every task getting the right model?Right model, right taskWhat if our AI vendor goes down or raises its price?No single point of supplyAre we paying for tools that do the same job?One tool per jobDoes the AI forget everything between conversations?Governed shared memoryIs this producing real work, or just answers?The digital factoryWhich work should run itself?Managed, not scriptedCan we get more from the Copilot licences we already pay for?Copilot proposes, the platform verifiesWill people actually use it?Meet people where they workHow much time do people lose sorting inbound information?Intake to insight, untouchedCan the same platform run a real business?One governed foundation for every productWhat happens when it breaks at night?Detect, fix, escalateWhere is the list of what is open, and who owns it?One list, closed with evidenceCould we recover if we lost everything?Proven by restoringCan you prove it to an auditor?Challenge before acceptanceDoes it get better over time, or just older?Every lesson becomes a ruleWill this still work after the next upgrade?Clean Core for AIDo our documents match what is actually running?Documentation as a by-product

The business question · Approvals and guardrails

“Can someone talk the AI into something risky?”

The concept

Risk-graded autonomy

Routine work runs, risky work waits for a person, destructive work is blocked for automation.

How it works: Each action is graded by risk: routine work runs, risky work pauses for an approval in Teams, and destructive work is blocked for automation.
Why it matters: Compliance gets human control exactly where the risk is, while everyday work keeps moving at machine speed.
The result: Approvals arrive where the owner already works, and every yes or no leaves an auditable trail.

See it workingRisk-graded autonomyReal screen

A real approval in Microsoft Teams, approved by Jeffrey Walters, then the agent reporting its own switch to a backup model — A real approval in Teams: the card records who approved it, and the agent reports its own switch to a backup model. Other chats are blurred.

The business problem

Agents can act faster than anyone can review them.

Autonomous agents act faster than any change board can review them. Agents and automated fixers on this platform can restart services and change data. One wrong action could take the platform down or undo the trust it had earned.

▶ Watch the answer: Who is your AI acting for?

Why this concept

Grade every action by risk: routine runs, risky pauses for approval, destructive is blocked for automation

Approving everything trains people to click yes; approving nothing is a gamble. Grading by risk puts human attention exactly where the risk is.

Also considered

Ask a person before every action
Let automation act freely and review afterwards

How it works

Risk-graded autonomy

Action

Proposed by an agent or an automated fixer

Risk grade

Routine runs, risky waits, destructive is blocked

Approval

Risky work waits for a yes in Teams

Action runs

Only once it is allowed

Record

Who said yes, to what, and why

What we put in place

Risky actions post an approval card in Teams, and only the owner can approve.
Every decision is kept with who, what and why.
Destructive actions are blocked outright for automated fixers, and critical services sit on a protected list.
A logged emergency path stays with the owner if normal access fails.

The result

Before

Agents can act faster than anyone can review them.

After

Approvals arrive where the owner already works, and every yes or no leaves an auditable trail.

I put people in the loop where the risk is, not everywhere, so AI stays fast and governed.

In your organization

An approval model compliance can sign off without slowing every team down.

Related case studies

Agent and user authentication

“Who allowed the AI to do that?”

The conceptIdentity travels with every request

Read the case study

Audit trail

“Can we see who did what, and why?”

The conceptOne thread from request to result

Read the case study

Next case study

One thread from request to result

Lessons learned

Contact

Looking for someone to lead AI from pilot to production?

I would be glad to walk you through any part of this live and talk about the role you have in mind.

Email me LinkedIn One-page brief

j.walters@erpaiintegration.com

Go deeper: Lessons · The journey · How it’s built · Live health