Datadog

**The problem they set out to solve was organisational rather than technical, and that is the whole entry.** At Wireless Generation they watched developers and operations staff work from different, incompatible data during incidents - each convinced the fault lay with the other, each holding evidence the other could not see. **The finger-pointing was not a failure of goodwill. It was the predictable result of two teams looking at different instruments and describing the same event in incompatible terms.**

**So the product's founding insight is not a monitoring feature.** It is that an argument about what is happening cannot be settled while the parties hold different data, and that a shared view ends the argument not by resolving it but by removing the ambiguity that fed it. Everything about the platform follows from that: one pane covering infrastructure, applications, logs, network and user experience, because **the moment any of those lives in a separate tool with a separate login, the old argument becomes available again.**

**Read against the two other observability companies on this timeline, the three approached the same destination from three different directions**, and the difference in starting point still shows in what each is best at. **Dynatrace** came from inside the application, tracing a transaction through code. **Kentik** came from the network, reading flow records and routing. **Datadog** came from the infrastructure, monitoring hosts and then everything attached to them. Three origins, one convergence - and the reason they converged is that an outage does not respect the boundary between an application, a host and a network, so any tool that does will eventually be asked to stop.

**One founder detail is worth keeping.** Pomel is an original author of the VLC media player - the open-source thing on almost every computer of a certain era, which played the file nothing else would. **There is a straight line from writing software that copes with malformed input from the whole world to building a platform that ingests telemetry from every system a company runs**, and it is not a coincidence that both jobs are mostly about handling what you were not expecting.

The commercial model is usage-based across more than twenty products, sold by landing in one team and expanding across others. **That model has a structural tension worth naming, because it applies to every observability vendor and not only this one: when the bill scales with how much you instrument, the customer's incentive is to observe less.** Nobody designs that deliberately, and it is the honest cost of metered infrastructure - **you are asking a customer to pay in proportion to how carefully they watch, at exactly the moments when watching carefully matters most.**

It launched infrastructure monitoring in 2012, which put it in position as containers arrived and made the number of things worth monitoring jump by an order of magnitude overnight, and listed on NASDAQ in September 2019. **It was also built in New York rather than California**, which its own commentators note - and which is a smaller point than it sounds, except that the customers who needed it first were banks and media companies rather than startups.