Operational AI · human-in-the-loop

The AI proposes. A person disposes.

Human-in-the-loop is not a disclaimer, it’s an architecture: the AI gets its own role in the permission system with read-and-propose rights and nothing else, every consequential action waits for a human click, and every decision lands in an audit log. This page shows the actual mechanics, from systems we run in production.

01. Staged autonomy

Autonomy is a ladder, not a switch.

The failed AI project has a recognizable shape: full autonomy on day one, a bad week two, and a quiet shutdown by month three. Production systems climb instead. Each workflow starts at read-and-propose, earns its way up with evidence, and some workflows correctly never leave rung two — money movement being the obvious one.

Fig. — The autonomy laddertrust is earned per rung
  1. 1Reads & summarizeszero risk, zero lift
  2. 2Proposes, human approveswhere operational AI earns trustwe deploy here
  3. 3Acts, human reviews afterearned per workflow, with audit trailwe deploy here
  4. 4Acts autonomouslyonly for reversible, low-consequence steps
Most failed AI projects jumped to rung 4 on day one. Production systems climb one rung per workflow, with evidence at each step.
02. The approval queue

One surface where judgment actually happens.

The loop needs a place for the human, and that place is a queue: everything the AI proposes, waiting with full context and a one-click decision. Done right, reviewing is faster than doing the work ever was, and the queue doubles as training data — every correction teaches the system your operation’s rules.

Fig. — The approval queuefrom Tandem, our TMS
Rate con extracted — L-2048, Chicago → Dallas, $2,165proposed by the agent · intakeApprove
BOL date differs from rate con — Jul 12 vs Jul 11proposed by the agent · verificationReview
Invoice INV-1042 ready — verified docs attachedproposed by the agent · billingSend
The AI never sends money or messages on its own. Every consequential action waits here for one human click, and every decision lands in an append-only audit log. Live in production in Tandem.
03. The permission model

A prompt is not a security boundary. Permissions are.

Telling a model “never send payments” is a wish. Making the AI a database role that cannotsend payments is engineering. In our systems the agent is a constrained principal: its credentials can read and propose, the approval step runs under the human’s credentials, and an append-only audit log records both. The same discipline applies to AI-generated code headed for production — which is why this page and our production-readiness practice rhyme.

Fig. — The operational loopevery deployment, same shape
01
Capture
emails, PDFs, portal screens, texts become structured records
02
Structure
records join to the jobs, loads, or cases they belong to
03
Propose
the AI drafts the action: an extraction, a match, an exception flag
04
Approve
a person rules on it — one click, full context
05
Act & log
the system executes and writes the audit trail
The loop with the boundary drawn: steps 1–3 run under the AI's constrained credentials, steps 4–5 under a person's. The seam between them is the security model.
04. Why this wins

Slower autonomy, faster adoption.

Operators trust a system they can overrule, so they actually use it. Regulators and enterprise buyers accept a system with an audit trail, so it survives review. The model improves on your corrections, so the queue gets quieter every month. Rung-two systems compound; rung-four moonshots get unplugged. That’s the whole argument, and it’s why every example in this cluster, from Tandem’s intake to DocuPaint’s inspections, is built this way.