Abstract:

AI agents are programs that work on their own for long stretches: they remember what they have done, call outside services that sometimes fail, and must recover from crashes without a human stepping in. Now imagine running thousands of them at once. Engineers at Ericsson faced the same challenge in the 1990s with telephone switches that could never go down, and they built Erlang and a toolkit called OTP to solve it, long before the web or multi-core chips became everyday technology. Francesco Cesarini, who worked on the original OTP team, explains these ideas from scratch, with no prior Erlang or Elixir knowledge needed. You will learn how OTP turns thousands of small, isolated programs into a few reusable design patterns, how "supervisors" restart anything that crashes, and why this 30-year-old approach makes Erlang's virtual machine, the BEAM, an ideal place to run AI agents today, including agents that write code.

Bio:

Francesco Cesarini founded Erlang Solutions and has spent three decades building, teaching and championing the Erlang ecosystem. He joined Ericsson's Computer Science Laboratory as an intern in 1995, where he worked on the first release of OTP and later contributed its process manager and debugger at Erlang Systems. He co-authored Erlang Programming and Designing for Scalability with Erlang/OTP, both published by O'Reilly, and lectures at Oxford University. He sits on the board of the Erlang Ecosystem Foundation and co-hosts the BEAM There, Done That podcast. Today he focuses on why the ideas behind Erlang and OTP matter more than ever in the age of agentic AI.