On the panel
Dheeraj Khanna, VP of Platform Engineering
Shares responsibility across software, data and infrastructure, so teams speak one language instead of pointing fingers.
Naveen Ede, Principal Platform Engineer
Defines the rules each layer has to hold, gives each one an owner, and brings every signal into one view for AI to analyze.
Dmitry Blinov, Sr. Director of Engineering
Built data quality checks over 10 years, on data about the data rather than customer data, because a data platform can fail its customers while every service is up.
Why reliability is hard for platform teams
Software and data companies ship changes all the time, across code, settings, databases and data formats. Most failures follow a change, but finding which one means checking many tools by hand.
Industry research finds that 83% of organizations run four or more observability tools, and that engineers spend 40% of their time on incident management.
The hardest failures are silent: every service is up and every dashboard is green, but the product or the data is wrong. Those failures are found late, often by a customer.
The challenge
A silent failure leaves every service up while the product or the data is still wrong. Reltio’s checks cover so much data that Dmitry calls analyzing it a clear use for AI.
Naveen cites DORA research: about 70% of failures come with a change. Changes land in settings, database migrations and data formats, and the clues sit in different tools, so most triage time goes to finding which change caused the failure.
Where AI helps
An agent can look at every source at once and pick out the signal that matters, which leaves engineers to decide whether to roll back or fix forward.
The panel agreed on the ground rules for agents in production: the same access rules as people, clear contracts for what an agent may do, masking for sensitive data, and a full record of what each agent did for auditors.
“The analysis phase has definitely shrunk with the AI coming into play.”
Habits to drop
Dheeraj would retire static runbooks in a wiki, so the runbook arrives with the alert instead. Naveen would make an AI agent the first stop when an alert fires, instead of the command line. Dheeraj would also have engineers ask one connected source a question, instead of checking seven by hand.
“There are much easier ways to ask questions and find answers now than to go to seven different sources and take 20 minutes to get an answer you can get in 7 seconds.”
What NeuBird, the AI SRE, does for software and data platforms
NeuBird gives engineering time back and gives the AI agents teams build the same governed production context from day one.
Resolve
Find the change behind the failure
NeuBird checks 15+ sources in parallel, lines up recent changes against the first errors, and finds the root cause in under 5 minutes at 94% accuracy.
Prevent
Hear about trouble first
NeuBird watches for the early signs of trouble and warns 30 to 60 minutes before a slowdown becomes an outage, instead of waiting for a customer to notice.
Operate
Keep runbooks and alerts honest
NeuBird tunes noisy alerts, flags runbooks that no longer match how systems work, and tracks repeat problems between incidents.
Your agents
The same context for your own agents
The agents and coding tools teams already use, such as Claude Code and Cursor, reach the same production context over MCP, with the same access rules and a full record of what each agent did.
