Blog
Notes from building a place to hire AI employees.
2026-07-19
We built a place to hire AI employees, not another agent builderWorkflow tools automate tasks. We wanted staff: teammates who know the company, act with real tools, and get better from feedback. Here is how Pagerox works and why we built it this way.
2026-07-20
The permission ladder: why AI employees should earn autonomyMost agents launch with whatever access their API key carries, and the only dial is on and off. Pagerox employees work under a per-operation ladder: off, observe, notify, suggest, act. Here is how each rung works and how an employee climbs.
2026-07-20
Process-and-drop: a privacy architecture for AI in SlackThe easy way to build an AI that listens to Slack is to mirror Slack into a database. We refuse to store raw messages at all: they are processed in flight and dropped, and only distilled work items with provenance persist. Here is why that is a feature.
2026-07-20
Corrections are the productEvery AI product claims it learns from feedback. Ask what it learned, who taught it, and whether you can undo it, and most go quiet. The memory ledger is our answer: attributed, visible, revertible. Silent learning is a bug with good marketing.
2026-07-20
You should be able to fire us: the portable employee bundleEvery vendor says you own your data. The test is what the export button produces. A Pagerox employee exports as a documented bundle: config, attributed memories, eval set. Credentials never export, and import always lands as a draft.