[ Orchestrated ] agents work the portals, humans hold the gate[ NS · supervised-portal-agents ]
Supervised Portal Agents
For the closed portals you can't integrate with, the ones locked behind multi-factor authentication (MFA), supervised agents do the swivel-chair work, with a human approving every critical action.
[ 01 ] The problem
Some of your most painful work happens in systems you will never get an API for: the insurer portal, the regulator submission system, the legacy core platform, the third-party portal behind multi-factor authentication. So a person logs in, navigates, copies data across, and submits, over and over. This swivel-chair work is exactly what a £50-a-month SaaS cannot touch, because the system is closed by design. It's the highest-value, most-stuck terrain in your operation, and almost no one will go near it.
[ 02 ] What we build
- We deploy supervised Claude AI agents that operate the portal the way a trained member of staff would, logging in, navigating, reading, and entering data, but under strict human control.
- Every critical action pauses for explicit per-action human approval before it happens. Nothing high-stakes is autonomous.
- Payments and bank-detail changes are never automated, full stop, and regulator submissions sit behind a hard pre-submission gate.
- Everything the agent does is logged in a complete audit trail.
- The top rung, 'agents work the portals, humans hold the gate', built only on top of reconciled data and proven lower rungs. We don't lead with it; we earn our way up to it.
[ 03 ] What you get
- Supervised agents that operate your closed, MFA-gated, or un-integratable portals
- Per-action human approval on every critical step, enforced before execution
- Permanent red lines: no automated payments, no automated bank-detail changes, hard gates on regulator submissions
- Role-based access control and a complete, exportable audit trail of every agent action
- A supervision and exception workspace where your team approves, edits, or halts the agent
- A failure and incident playbook so a stuck or unexpected state escalates safely to a human
Why it matters
- Built on every rung below it: we only deploy once data is reconciled and the reviewed pipeline is proven
- For our transport-fleet insurance client, supervised insurer-portal agents are the planned destination after the inbound pipeline and write-back are live
- Governance isn't a constraint we apologise for here; it's the entire reason this can be trusted in a regulated operation