The Approval Fits in Your Pocket

A developer we follow shipped real features from his phone recently. No desk, no keyboard, no big screen. An agent did the building; he did the deciding. His conclusion was that some of his best sessions now happen away from the computer.

Strip away the novelty and there's an operations lesson underneath. When the execution moves to agents, what's left of the human's job is decisions. And decisions are small. They fit in a notification.

The desk was for typing

Most operational work historically required a workstation because the work was manipulation: typing into the ERP, reconciling the spreadsheet, assembling the report. The judgment buried inside it ("approve this," "escalate that," "yes, that vendor") took seconds.

Automation separates the two. The manipulation runs on a server all day. The judgment arrives wherever you are: invoice 0117 matched PO-4482, flagged a 2% variance, approve or review? That's a phone-sized question. Answering it from a hallway is the correctly sized interface for the decision.

What has to be true first

Phone-sized approvals only work when the system underneath earns them. Three requirements, from the builds we run:

The summary must be trustworthy. You're approving a description of the work, so the description has to be grounded in the underlying data, with the source one tap away.

The blast radius must be bounded. A yes should commit exactly what was described. Hard guardrails decide what an agent is allowed to do with your approval.

The default must be safe. If nobody answers, the system waits or routes onward. Silence is never consent.

We design our approval queues to that standard, and it changes how teams experience automation. The system does the forty hours. The owner spends four minutes a day steering it, from wherever they happen to be standing.

In production, an agent looks like a quiet queue of well-formed questions, each one small enough to answer from your pocket.