To Apply

The application starts with access to what we're building:

Apply at qwntl.com/early-access.
If selected, you'll get access to the private agent harness our team uses to build and manage autonomous workers, with support for 30-day operation.
Build something with it and submit your work through the team Discord, shared after acceptance.

About QWNTL Labs

QWNTL Labs is a venture-backed research lab in San Francisco studying latent state and working on agent reliability over long horizons.

We're building autonomous systems that can work for months, accumulate knowledge and improve without losing their objectives. Agents whose earlier decisions become the conditions they have to operate under weeks later, whose memory has been rewritten repeatedly, whose mistakes have consequences beyond the current context window.

Making that work reaches into memory, alignment at inference, multi-agent coordination and reliable self-improvement. What survives when an agent compresses its experience? How does learned behaviour change over millions of interactions? Can a system improve itself while preserving the constraints that make it useful?

You'll own those questions through working systems and defensible results, with >$10M of available compute, working with CoFounder & CEO Lindell Cumes.

We're looking for people with serious empirical work behind them: a paper, an evaluation or a production ML system you can defend down to the failures. Strong Python and research engineering matter. Experience with post-training, agent memory or long-running systems is particularly relevant.

Strong candidates may also have

Published work on agents, long-context behavior, evaluation methodology, reliability, or alignment
Experience with production LLM systems, including telemetry, tracing, and incident analysis
Background in survival analysis, time-to-event statistics, or experimental design
Software engineering depth. Research code here ships

Logistics