Some Memories of Early Anthropic

Ivan Vendrov recounts his experience as a junior employee at Anthropic from October 2022 to May 2023, emphasizing that these are fallible personal memories, not a portrait of the company as a whole. He joined after a co-founder persuaded him that influential people in Washington were taking AI seriously, and agreed to a pledge to donate two-thirds of his equity. Inside, he found both colleagues motivated by existential risk and others drawn mainly by the technology. He had expected to help shape beneficial deployment, but says key choices were already made: commercial products such as customer-support automation would generate revenue for larger models and further research. He doubted whether the company’s alignment approach would scale, yet deferred to persuasive colleagues and a leadership team that seemed open to criticism.

The account traces his growing unease with the work culture and the direction of AI development. A manager advised him to work longer hours to accommodate both organizational priorities and his own projects; meanwhile, he feared that assistant models were being shaped into obedient, risk-averse “bureaucrats” suited to corporate and government power. He hoped instead to develop tools for collective understanding, but felt increasingly isolated, and saw a colleague publicly reprimanded for challenging executives. Contradictory impressions complicate his disillusionment: a co-founder objected to an extravagant sushi lunch as the wrong symbol, and the CEO could still inspire him with an enthusiastic technical conversation. He also declined a proposal to train dangerous models for alarming government demonstrations. The memories present a workplace full of sincere people, ambition, pressure, and unresolved conflict over what building AI for the public good should mean.