Remote bets AI agents replace the dashboard: ships MCP server so agents run payroll

Amsterdam-based global payroll and employment platform that hires and pays workers across more than 100 countries.

When agents become the interface, HR software's winners are owned entities, compliance depth and calculation engines — the backend agents call, not the best dashboard.

For seven years Remote owned 100% of its legal entities and built proprietary payroll calculation engines instead of routing through chains of local aggregators. Van der Voort treated agentic AI as the moment to expose that engine: his own open-source personal agent, named Jim, became the first explorer, allowed to work in Remote 'but he cannot do destructive things.'

In 2026 the company shipped Remote MCP; BambooHR and Workday adopted Remote as the engine under their own products. Van der Voort says more than 85% of the last month's code was AI-written and revenue per employee is up 50% with no layoffs. Remote Build embeds forward-deployed engineers in customer organizations, fed by Remote Labs, an internal AI marketplace; AI-powered payroll runs in 100+ countries.

Seven years of owning legal entities and calculation engines created a clean, agent-callable engine; aggregator chains of local vendors cannot expose one.

Letting BambooHR and Workday build on Remote turns rival HR suites into distribution for its payroll engine.

Scoped, non-destructive agent access answers the scariest objection to agentic AI: payroll errors mean people don't get paid.

Saying on the record that the platform should 'disappear in a good way' stakes out the agent-era thesis first in a category where everyone claims the same AI story.

When interfaces shift from screens to agents, value moves to what agents can't fake: owned entities, compliance depth, calculation engines. Better to announce your own disruption than await Deel's.

As of 2026-10-10, AI-powered payroll runs in more than 100 countries and hiring plans are being re-evaluated toward upskilling over headcount, which the company calls deferred rather than avoided hiring. The MCP security model remains unproven at enterprise scale, and the numbers — 85% AI-written code, 50% revenue-per-employee gain — are self-reported in the SaaStr profile rather than independently verified.