Meta Offers ~95% Discount on Muse Spark in Exchange for Agent Training Data

Skynet Chance (+0.02%): Paying to harvest real-world agent trajectories accelerates the creation of more autonomous, capable agents while transferring proprietary workflow data into frontier training pipelines, modestly widening the capability-versus-oversight gap. The risk is incremental rather than structural, since it concerns data acquisition economics rather than any change in control mechanisms.

Skynet Date (+0 days): Unblocking the data bottleneck for non-software professional workflows would speed the deployment of autonomous agents into more domains, pulling capability-driven risks slightly forward. The effect is a marginal acceleration of an already fast trend rather than a step change.

AGI Progress (+0.02%): The article highlights that agent session data was the driver behind the large 2025 jump in coding-agent capability, and Meta is now paying to acquire exactly that scarce feedback signal for reinforcement learning. Overcoming the shortage of digital traces for complex professional workflows addresses a real limiting factor on general agentic competence.

AGI Date (+0 days): Cheap access in exchange for training rights could substantially increase the volume and diversity of real-world agentic data, compressing iteration cycles toward broadly capable agents. Counterweighted by enterprises' documented reluctance to share proprietary data, which limits how much acceleration this actually buys.