The dirty secret of modern network operations is that we have too much of a good thing. We’re drowning in telemetry, yet when the backbone has a hiccup at 3 a.m., we’re still stuck staring at a dozen different dashboards trying to play “spot the difference.” Most tools today are vertical silos that excel at telling you their specific world is on fire but fail to explain why the neighbors are screaming. Selector AI is taking a swing at this mess by moving away from the old way of siloed monitoring toward a horizontal, multi-domain reality. We had a chance last month to take a peek at what they’re working on during Networking Field Day 40.
The Death of the Vertical Silo
If you’ve spent any time in a Network Operations Center (NOC), you know the frustration of vertical harmony in a crisis. The network team says the links are green, the server team says the CPUs are fine, and the application team is certain the database is the bottleneck. This happens because our data is trapped in vertical stacks. Selector AI changes this by utilizing a horizontal data lake that ingests data from over 300 sources. It doesn’t matter if it’s an SNMP trap, a streaming telemetry metric, or a random log from a cloud gateway.
The real magic here is the shift from ETL (Extract, Transform, Load) to ELT (Extract, Load, Transform). In a traditional ETL model, we often strip away metadata during the Transform phase to save space or simplify the data. As a result, we lose the very context needed to correlate a BGP flap in Chicago with a spike in application latency in a VPC in Vancouver. By loading everything first and preserving metadata and exact timestamps, Selector ensures that every signal stays pinned to a master timeline. This is the difference between seeing two unrelated sparks and seeing a fuse burning toward a powder keg.
Correlation Is Not Causation
The industry loves to throw the term AI around, but most “AIOps” tools are just fancy regression testers. They see two things happen at once and assume they’re related. In practice, just because your BGP session dropped at the same time a server crashed doesn’t mean the BGP flap caused the crash. It might be the other way around, or a third factor like a power failure might have nuked both.
Selector’s Data Hypervisor layer handles the heavy lifting of deduplication and noise suppression, but the real intelligence lives in its causation modeling. The system understands the underlying physics and logic of networking. It knows that a physical link degradation at Layer 1 is the most likely culprit for a BGP flap at Layer 3. By applying this architectural awareness, the platform can move past similarity grouping and actually tell you the sequence of events. It stops the blame game by identifying the first domino to fall.
Agents and the New Troubleshooter
The industry is moving toward a world where we don’t look at data; we interrogate it. Selector’s use of an agentic Copilot is a nod to how modern engineers actually work. Instead of clicking through five menus to find an interface’s optical RX levels, you use a natural language interface to ask the bot in Slack. The system consults internal expert tools focused on routing or synthetics, stitches the data together, and hands you a summary in plain English.
Taking it a step further, the platform handles both non-intrusive and intrusive automation. It can automatically run a traceroute or a ping mesh across 40,000 devices to verify liveness before an operator even logs in. If it finds a problem with an optical connection, it can trigger an Ansible playbook to flap the port or clear the counters. This isn’t just about speed. It’s about augmenting the L1 or L2 engineer with the institutional knowledge of a senior architect.
The Metadata Tax
None of this works if your house isn’t in order. AI isn’t a magic wand that can fix a network named by a committee of mythology nerds. If your WAN links are labeled “Link_A” in one system and “Chicago_Core_01” in another, correlation will fail. Selector addresses this through intensive onboarding workshops focused on metadata hygiene. You can’t just “plug and play” this level of insight. You have to map the environment and establish business rules for alert routing and logic first.
Ultimately, the goal is to get to a point where the system can recommend specific internal knowledge base articles or tell you exactly which vendor to call. By embedding your own documentation into the analysis, the platform stops being a generic tool and starts acting like a member of the team.
Bringing IT All Together
The era of dashboard staring is coming to a close. We don’t need more graphs. We need more answers. Selector AI’s strength isn’t just that it collects a lot of data, but that it respects the nuances of network architecture enough to distinguish a symptom from a root cause. Use the AI to find the needle in the haystack, but make sure your senior engineers are the ones deciding when to burn the hay.
For more information about Selector and their platform, make sure to check out their website at https://Selector.ai. To see their entire presentation from Networking Field Day, head over to the presentation event page here.
