The Engineer Who Decided Nobody Should Debug Half-Asleep at 4am
Topaz Turkenitz spent years keeping distributed systems alive at 99.9% uptime. Then she left to build Vespper, an AI that takes the pager so engineers can sleep. This is how the on-call grind became a company.
There is a specific kind of tired that only on-call engineers know. It arrives at 3am, in the dark, with a phone buzzing on the nightstand and an alert that could mean everything or nothing. You open a laptop you can barely see, squint at a graph, and start asking the same questions you asked last time. Topaz Turkenitz knows that tired. She built a company so that the rest of us might feel it a little less.
Turkenitz is the co-founder and CEO of Vespper, a company in Y Combinator's Fall 2024 batch. The pitch is short enough to fit in a Slack message, which is roughly where it lives: a 24/7 AI on-call engineer that triages your alerts and surfaces the relevant context in seconds. But the reason it exists is older than the company. It comes from years of being the person on the other end of the pager.
01 / The GrindYears of Keeping the Lights On
Before founding anything, Turkenitz was an engineer at Snyk, the developer-security company that grew into a unicorn while she was there. Her work was the unglamorous kind that keeps software companies alive: building and maintaining large distributed systems, running the experiments that grow a product, and holding the line on reliability. Her team's services ran at 99.9% uptime. That number looks clean on a slide. Behind it are a lot of nights.
She has a computer science degree and, less commonly for an infrastructure engineer, a real feel for design. At Snyk she used both. She led one of the company's top growth experiments and spearheaded a project that helped land a Fortune 500 bank as a customer. Earlier, at the global design marketplace 99designs, she drove product work end to end. The through-line is someone who likes to ship and likes it to be good.
Somewhere in all of that, a pattern set in. The alerts kept coming. Most of them were noise. A few were real. Sorting one from the other, again and again, ate hours that could have gone to customers. And the knowledge needed to actually fix things tended to live in one person's head, which was a problem the moment that person went on vacation.
The mission is simple: making on-call stress-free.The founding idea behind Vespper
02 / The DecisionLeaving to Fix the Thing She Lived
Plenty of engineers complain about alert fatigue. A smaller number quit over it. Rarer still is the person who leaves specifically to solve it. That is the move Turkenitz made in 2024, teaming up with Dudu Lasry, a machine-learning and computer-vision engineer whose past includes stints at Viz.ai, Google, and SafeBreach. Between the two of them, they had shipped at both a security unicorn and a medical-AI unicorn. What they shared was the same frustration with on-call.
The result is Vespper: a multi-agent AI system that behaves like an on-call engineer who never sleeps and never forgets where the runbook is. When an alert fires, it triages it, pulls together the relevant signals, and posts a diagnostic hypothesis in Slack while the humans are still finding their glasses. It plugs into the tools teams already run - Datadog, Grafana, Prometheus for observability, PagerDuty and Opsgenie for incidents, Notion and Confluence for the knowledge that would otherwise be trapped in someone's memory.
The design choice underneath it is worth noticing. Vespper is not trying to be another dashboard. There are already plenty of dashboards. It is trying to be the thing that reads them for you, which is a much harder and much more useful problem. That instinct - remove the toil, keep the human for the judgment - runs through everything Turkenitz builds.
03 / The Character:shipit:
If you want a quick read on someone, their GitHub bio is a decent place to start. Turkenitz's is a single item: :shipit:, the little squirrel-in-a-hat emoji engineers drop into a pull request to say it is ready to go out. No paragraph. No manifesto. Just ship it.
Her handle there is topaztee, and the projects she has pinned tell a small, honest story. One is a Chrome extension that adds, in her words, "some flare" to Trello's plain progress bar. It is not a startup. It is not a resume line. It is someone who looked at a slightly boring piece of software and could not resist making it a little better. That is the same reflex that turns a personal annoyance with on-call into a company.
She looked at a boring progress bar and made it better. Then she looked at the on-call rotation and did the same thing.On the builder's instinct
There is also a badge on her profile that reads Arctic Code Vault Contributor. It means some code she wrote is archived on film, in a decommissioned coal mine in the Norwegian Arctic, meant to last a thousand years. She probably was not thinking about geological time when she committed it. That is rather the point - the work outlasts the moment you were tired.
04 / The BetWhy On-Call Is Now a Company
The wager Turkenitz is making is that AI pattern recognition, combined with the operational discipline of AIOps and real domain expertise, can change what it feels like to run software. Not by replacing engineers, but by taking the part of the job nobody signed up for: the reflexive, sleep-deprived triage that happens before the actual thinking can begin.
It is a two-person company operating across Tel Aviv and San Francisco, which is its own kind of always-on. Vespper started life partly in the open, with a public repository under the vespperhq organization - fitting for founders who came up shipping developer tools and who understand that the engineers they are selling to want to see how the thing works.
Ask what success looks like and the answer loops back to that plain mission. Stress-free on-call. An engineer who gets paged, glances at Slack, finds the hypothesis already waiting, and either fixes it in minutes or goes back to sleep because the AI already ruled it out. No heroics. No 4am squinting. Just a quieter, more humane version of the job Turkenitz spent years doing by hand.
The best founders don't invent a problem. They pick one they have already lived - then they refuse to accept it.The shape of Turkenitz's bet
She is early. The company is small, the funding is seed-stage, and the market is crowded with tools that promise to fix incident response. But there is something durable about a founder whose product is a direct answer to her own worst nights. It is hard to fake that. You cannot outsource having been the person on the pager. Turkenitz was, and now she is building the thing she wished she had - which is, in the end, how most of the good ones get made.