I run multi-agent systems in production and publish what breaks, with the numbers.
Agentic AI in Production is about AI you can trust in production. You get three things here: a weekly newsletter on the failure, the number and the config that fixed it; a podcast; and the latest news in tech.
I started in trading systems, on Fidelity’s online trading platform from 2014 to 2018, then on intraday risk at Merrill Lynch from 2019 to 2021. Every number there had to be right and explainable after the fact. I took that habit to production AI, where the hard part is knowing when the system is wrong before a customer does.
Three rules run through everything I publish. The worker never grades its own work, so checks run outside the agent they judge. Every decision leaves a record you can replay later. Autonomy is earned with evidence, never by assertion.
Measured, not asserted.
Open source: Attestor (attestor.dev), context-clock and prompt-ledger (github.com/bolnet).
Find me on LinkedIn (linkedin.com/in/singhsurendra) and X (x.com/surendra_ai).
