It’s been two weeks since Verge politics reporter Makena Kelly dove into the IT tangle that has taken down unemployment insurance systems across the country. Since then, another 4.4 million people have filed claims — bringing the total to more than 20 million people since mid-March — and the situation has hardly improved. Some people report having to call hundreds of times before getting through, keeping them in limbo for weeks.
A lot of the blame has fallen on COBOL, a 1950s-era programming language that a lot of the systems still run on. But this isn’t like old hardware breaking down, and COBOL isn’t exactly broken. The problems with running a decades-old programming language are more subtle — so subtle that you can run for years without any obvious problem.
The problems only show up when you suddenly need to handle an unprecedented surge in traffic and you find out your state unemployment system can’t scale up the way a service like Netflix or Zoom can. But to understand why that is, you need to understand the way network management has changed over the past 20 years (that is, the shift from pets to cattle) and how technical debt can lock you into the old way of doing things.
And most importantly… you have to look at the Big Picture.