Selector
speaks network.

01The problem they saw

Networks broke, quietly, for a long time before the rest of the software industry noticed. Cloud-native got the headlines. Kubernetes got the conference talks. The actual wires - the BGP sessions, the SD-WAN tunnels, the carrier MPLS, the F5 load balancers - kept getting more complicated and less observable. By 2019, a network operator at a large telco was juggling more signal sources than a NASA flight director, and roughly the same amount of free time.

Selector's two founders had a closer view of this than most. Kannan Kothandaraman had run products at Juniper Networks. Nitin Kumar had been a Juniper Fellow, the company's senior-most engineering rank, where he drove software architecture across the platform. They had spent fifteen years building the things that broke. They knew exactly how the breakage looked from the inside.

What was missing, they decided, was not more data. There was already too much data. What was missing was an opinion. Something that could look at the data and say: here is what happened, here is why, and here is what to do.

02The bet

In 2019, "AIOps" was already a fashionable acronym. It was also, mostly, marketing. The first wave of AIOps tools were anomaly detectors with a thesaurus. They flagged. They scored. They rarely explained anything.

Kannan and Nitin made a more specific bet. They thought network operations was not one AI problem - it was three, stacked.

That third part - the language part - turned out to be the unlock. Selector built what it now calls a network-specific LLM: a model trained on the dialect of network engineering, the way operators actually phrase a question at 3 a.m. Coupled with the knowledge graph, the LLM does not hallucinate. It cites.

It was a quietly contrarian thesis: that the right model for AIOps was not a general one but a domain-soaked one, the way a good translator soaks in the language they translate.

03The product, in plain English

Selector sells a single platform with four faces, depending on who is asking.

Network LLM & Copilot

You type: "Why is latency up on the Frankfurt edge?" The copilot reads the live telemetry, walks the topology graph, finds the misconfigured route, and writes a one-paragraph answer that links to its sources. This used to be a forty-five minute exercise. Now it is a sentence.

Correlation & Root Cause Analysis

Alert storms get collapsed into actual incidents. Selector says one of its customers reduced ticket volume by an order of magnitude after correlation kicked in. Whether that is true in your environment depends on your environment, but the direction is clear.

Operational Digital Twin

A live, queryable model of the network used to simulate change, predict impact, and plan capacity. The name sounds like a slide deck. The thing itself is genuinely useful: you can ask "what breaks if we drain this region?" and get an answer that is not a meeting.

Multi-Cloud Observability

Launched in May 2026, this is the connector tissue between on-prem networks and the public clouds that increasingly sit on top of them. Selector calls it closing the "network-to-cloud visibility gap," which is jargon for: your AWS team and your network team can finally read the same screen.

04The proof

Pitch decks are cheap. Carriers are not. A telco's procurement cycle is twelve to eighteen months, and they do not buy software because the demo looked good.

Selector's public reference list includes Bell Canada, whose Senior Manager of Infrastructure & Software Cloud Delivery has gone on the record. The rest of the customer roster is, in the great tradition of enterprise software, "various Tier-1 telecommunications providers and cloud service providers." The contracts are large and the renewal rates, according to the company's funding announcements, are unusually sticky.

The investor list - Atlantic Bridge, SineWave Ventures, Two Bear Capital, Ansa Capital - is heavy on funds with operator backgrounds and a thesis around infrastructure software. None of them are tourists.

05The mission, said plainly

Selector frames its mission as eliminating downtime for the world's most complex networks. The version they will say in a keynote is grander; the version operators care about is smaller and more honest: fewer 3 a.m. pages, fewer war rooms, fewer dashboards no one reads.

The company is engineering-led in a way that is easy to miss in their marketing. Their team has imported people from Juniper, Cisco, Uber, Meta, Nutanix and VMware - which is to say, from companies that ship at scale and have opinions about how networks fail. There is a craftsmanship vibe to the culture that does not always survive a Series B; for the moment, at Selector, it has.

06Why it matters tomorrow

The networks Selector targets are about to get worse, not better. Every enterprise is now multi-cloud whether it meant to be or not. AI workloads are pushing more east-west traffic than most networks were ever designed to carry. The number of devices keeps going up; the number of qualified network engineers, stubbornly, does not.

This is the secular tailwind. Selector is positioned, almost suspiciously well, to ride it. They have a domain model nobody else has bothered to build. They have a customer base that is happy enough to keep doubling its spend. And they have founders who can, plausibly, claim to have built parts of the problem they are now solving.

It is 2:47 a.m. on the Eastern Seaboard again. The NOC is quieter than it used to be. One engineer is on duty. She types a question into a chat box and reads a paragraph back. The packets, somewhere, get faster.

The dashboards are still on. Mostly out of habit.