Summary
Swami Silva Subramanian argued that AI agents stall in production because teams chase demos while skipping measurement, governance, and clear business outcomes. Drawing on AWS customer work and Amazon’s own “thousand flowers” agent experiments, he showed how mechanisms that spot ROI and adoption signals turned Mesh Claw into the open-source Kiro Crew.
Key Takeaways
- Define the business outcome before starting an AI agent prototype.
- Measure ROI and adoption signals before scaling agent investments.
- Build governance and evaluation into agents from the beginning.
- Let teams experiment, but create mechanisms to spot what works.
- Avoid reactive AI decisions driven by fear of missing out.
“We skip verification because it slows us down, and we actually skip governance because it's not as exciting as the next feature.”
S
Swami Sivasubramanian
Speaker
Share
Up Next
SJD
WatchNext in agenda
Deploying at Scale: The True Cost of AI
Susan Jones, James Waters, Dasha Afanasieva
ai adoptionenterprise ai
ML
WatchAlso on ai agents
The Future of Materials Discovery
Max Welling, Larissa Holzki
materials discoveryai agents
CN
WatchAlso on ai agents
Is AI Killing Software as a Service?
Clay Bavor, Niall Firth
ai agentssaas