From agent harnesses to always-on assistants

AI Engineer15m 36s
0 comments · 0 votesOpen discussionClose discussion
Sign in to join the discussion

    Video summary

    Sam Bhagwat, Mastra's co-founder and chief executive, outlines a progression from model calls to tool-using agents, durable harnesses and always-on assistants. Harnesses add planning, background execution, interruption, steering and persistent sessions so work can continue beyond a single exchange.

    Bhagwat describes the architectural shift from local execution to cloud sandboxes and team messaging. He uses the term claw for systems that notice external events, wake on schedules and learn from prior runs. Automatic skills and harness changes are emerging approaches, not settled answers about safe continual improvement.

    His prediction is partly about user demand: people pull assistants into more tasks as capabilities expand. Bhagwat also expects consolidation, comparing it with mobile applications that survive because they are frequently used or provide substantial value. He advises builders to meet capability expectations while preparing for selective adoption rather than assuming every powerful assistant will retain users.

    Original YouTube thumbnailWatch on YouTube

    Share this page

    Sam Bhagwat wearing glasses and a plain blue T-shirt on black beside the blue and white headline “Always-On Agents”. Framed in blue with WWW.ARTIFICIAL-INTELLIGENCE.VIDEO, 21 July 2026 and duration 15m 36s.

    Sam Bhagwat predicts that agent harnesses will gain persistence, initiative and learning, while lasting adoption will depend on frequent or substantial practical value.