Cisco-sponsored MIT Tech Review piece pegs multi-agent system failure rates at 41–87%
This is labeled sponsored content from MIT Technology Review Insights in partnership with Outshift by Cisco, and should be read as vendor positioning — but the numbers it cites are worth knowing. Vijoy Pandey argues the path to superintelligence is horizontal scaling across coordinating agents rather than bigger models, and states current multi-agent systems fail at rates between 41% and 87% because agents can't hold shared goals. Cisco's proposed fixes are shared intent via cognition state protocols (internal testing claims goal agreement rose from about a third of scenarios to 93%), a persistent 'cognition fabric' for cross-session memory, and CASA continuous authorization — all riding on AGNTCY, an open-source agent connectivity layer now under the Linux Foundation.
↳ Follow the thread