Anthropic engineers discuss Opus 5.5 and Claude Platform

On October 28, 2026, at 02:40, the podcast Latent Space released an interview with Anthropic engineering lead Katelyn Lesse and product lead Angela Jiang. The conversation focused on the engineering behind the Claude Platform, covering the launch of Opus 5.5, the decision to separate the model's "brain" from an agent's "hands," scaling the API, and the reliability work behind managed agents. The episode also touched on Anthropic's S-1 filing, the company's two core loops of Improvement and Investment, and how the harness for long-running agent tasks keeps progress intact across context windows.

At 02:53 the same day, Latent Space published a second piece built around the same conversation, framed around Opus 5.5, the Claude Platform, and the engineering details of building AI agents. That material cites Anthropic's S-1 filing from the day before, which describes the lab as having some of the highest revenue and research efficiency in human history, and a turnaround in the last year from a -126% loss to 19% GAAP profit. The company attributes that to its Improvement and Investment loops.

On Opus 5.5, the two executives said the previous Opus had been criticized as too verbose, and that the new model emphasizes intelligence per unit of cost. They said Anthropic claims it runs more cheaply than its predecessor, with the exact savings depending on the workload. The piece also notes that a month has passed since the Opus 5.5 launch and that it has become the generally accepted #1 LLM in the world. Full footage is available on YouTube. As of this writing, no further official updates have followed the interview material.