RAY, the AI operator inside Spectrum

An operator that works beside your engineers, and proves itself as it goes.

RAY does the same work a technician does, through the same screens, with the same approvals. It does more of that work on its own only where the evidence shows it is safe, and it shows that evidence to you.

See RAY at work → How autonomy is earned
Human parity

No secret back door for the AI.

Everything RAY can do, a person can do on a screen in Spectrum, with the same gates. A capability that only the AI can perform is treated as a defect.

That one rule is what makes RAY safe to adopt. Your team can always see, repeat, take over or undo what RAY did, because RAY used the same buttons they would have.

  • Same recordsEvery RAY action leaves the same call note, audit entry and history a person's action leaves, and names the person who handed RAY the work.
  • Same approvalsWhere a person needs sign-off for a change, so does RAY.
  • Same courtesy to usersRAY offers people the choices a good technician would: restart now, in an hour, tonight, or not today. It respects the answer.
  • It asks earlyWhen RAY is unsure or out of options it stops, says what it tried and what it thinks, and asks a person. Asking is a success.
The operating loop

From the first alert to the lesson learned.

RAY follows the full lifecycle of a fix. The highlighted step is where a person decides.

DetectNotices alerts, failed patches, policy gaps and user reports, and groups them into one problem.
Log the callOpens a call on the right client's desk, or joins the one already open.
InvestigateRuns live reads on machines, logs and the network, and writes findings on the call.
ResearchWeighs vendor guidance and known issues, and records the likely cause with its confidence.
PlanWrites the steps, the risk, the blast radius, the undo and how success will be checked.
ApproveA person approves, changes or declines the plan. Nothing runs before this.
Agree with the userOffers the affected person a convenient time and honors the answer.
DeployRuns the plan in stages, pilot first, with live progress anyone can watch.
VerifyConfirms the end state and checks that nothing else broke.
CloseWrites what was wrong and what fixed it, and resolves the call.
WatchLooks again at one day, seven days and thirty days, and reopens if it returns.
LearnTurns the case into a lesson and into evidence for the capability it used.
Modes

You choose how much to hand over.

Every client starts with RAY watching. Each step up is a deliberate decision, made per client, and limited by what each capability has proven.

Mode 1

Watch

RAY reads, investigates, researches and writes notes and plans. Nothing runs.

Mode 2

Propose

RAY sends plans for approval. A person approves each one before it runs.

Mode 3

Drive, with a person watching

RAY runs approved steps while a named person watches live and can stop it at any moment.

Mode 4

Drive within a handover

RAY runs proven capabilities on its own inside the scope and limits a person set. Each run is still a call, a plan, a record and a check.

The capability ladder

Autonomy is earned on evidence, one capability at a time.

Every specific thing RAY can do sits on one of seven rungs. The rung is calculated from RAY's record, never typed in by anyone, and the final rung is switched on by a person for each capability and each client.

Rung 0Cannot yetNo route, recipe or access exists.
Rung 1Needs a personRAY prepares and researches; a person does it.
Rung 2LearningA recipe exists and runs only on rehearsal machines.
Rung 3ProposesRAY plans it; a person approves every run.
Rung 4Drives, watchedRAY runs it while a person watches.
Rung 5ProvenRuns on approval with no watcher, after a strong record and no collateral.
Rung 6May run in a handoverSwitched on by a person, per capability, per client.
Demotion is automatic. A confirmed side effect or a reversal drops the capability a rung on every client, at once. Your own estate first. Rungs are earned on the MSP's own machines before any client's.
The capability register

See exactly what RAY can and cannot do.

The register lists every capability, its rung, the evidence behind it and what it needs to move up, with charts of how fast that edge is growing.

"Cannot yet" and "not assessed" are shown, never hidden. When RAY reaches the edge of what it can do, it asks a person, learns from how they did it with their consent, and the edge moves outward.

Guardrails

Designed so that a mistake is small, visible and undone.

No secrets to the model

RAY is told which account to use, never its password. Credentials stay in the vault and are used by the system on RAY's behalf.

A breaker for harm

If a change causes collateral damage, the run stops, the capability is demoted, and people are told what happened.

Undo before change

A plan names its undo before it runs. Where there is no safe undo, RAY proposes how to make one.

Staged, never all at once

Pilot machines first, then wider groups. Each stage must pass before the next one starts.

Remote access with consent

Before RAY or a technician connects, the user sees the call number and what will be done, and chooses to allow it.

It improves the system

When a screen or flow is slow or clumsy, RAY proposes an improvement. People decide what gets built.

Questions we hear

About RAY

Does RAY run anything on its own from day one?

No. Every client starts in Watch mode, where RAY investigates and writes plans but nothing runs. Moving to Propose, and later handing over proven capabilities, is your decision.

What happens if RAY gets something wrong?

The run stops, the change is undone where an undo exists, and the capability is demoted one rung on every client automatically. The call shows exactly what happened and who approved it.

Will RAY replace my engineers?

RAY takes the repetitive, well-understood work so your engineers can spend their time on the problems that need judgment. It asks them when it reaches its limits and learns from how they solve the problem.

Can my team see everything RAY has done?

Yes. Every action is on a service desk call and in the audit trail, in plain language, with the plan, the approval and the verification result.

Watch RAY work a real call.

We will walk you through a fix from the first alert to the verified close.

Request a briefing →