Lindy 3.0 launches with agent autopilot feature that operates its own computer and manages logins

Summary

Lindy has shipped version 3.0, the most significant update since 2.0 launched in November 2024, and the headline feature is autopilot, a capability that gives each AI agent its own dedicated computer with persistent login credentials. Rather than re-authenticating every session, users authenticate once and the agent retains access, enabling long-running scheduled workflows against paywalled or credentialed services. Flo, Lindy's founder, demonstrated a live example where an agent wakes at 9 a.m. daily, pulls posts from his authenticated Twitter account, rewrites them in a LinkedIn-native format, and publishes directly, all without manual intervention.

The product is live today at lindy.ai, not announced as a roadmap commitment. Pricing starts at $50 per month on a consumption model, which Flo frames as roughly one-tenth the cost of a human employee. On runaway billing risk, the company says it handles overages generously with refunds, and points to declining inference costs as a structural backstop.

Lindy is already using 3.0 to automate its own QA function. An agent wakes hourly, registers a fresh account, attempts a payment, creates a new agent, tests it end-to-end, and pages the on-call engineer on failure. That displaces a dedicated QA engineer role.

Onboarding friction, historically a problem for open-ended agent platforms, is addressed in 3.0 through a plain-text setup interface and a library of pre-built templates across sales, customer support, engineering, recruiting, and personal productivity. Flo estimates agent creation time at roughly 90 seconds.

On the model layer, Flo identifies memory as the single most important unsolved problem for agent platforms. He argues that application-layer memory hacks are a dead end and that foundation lab researchers have advised companies like Lindy to stand down, with native model-level memory expected to arrive in six to twelve months. He expects GPT-5 to represent a meaningful step-change in reasoning quality. His near-term AGI benchmark moment came last week when a Lindy agent, given computer access, spontaneously built and deployed a working website without being explicitly instructed to do so.