The Pulse

Google Engineer Resigns Over Chip Tools That Could Speed AI

Robert O’Callahan resigned from Google on September 24, 2026, because his team’s chip-design tools could make AI cheaper and faster, he says. He plans to keep maintaining Pernosco and rr while studying how AI agents debug code.

AI.info Team · Written with AI · fact-checked · 2026-09-25

“My team’s goal is ultimately to make AI much cheaper and lower-latency, and I don’t think that’s good for people right now.”

Robert O’Callahan, in his resignation email reproduced on his blog

Robert O’Callahan resigned from Google on Thursday, September 24, citing his work on tools intended to speed the design of hardware chips for artificial intelligence. He did not say he worked directly on AI models or capability research. The tools could help produce a new generation of AI chips, lowering costs and latency and making AI more pervasive, his post says.

His concern starts with chip-design tools

O’Callahan had once considered the work relatively harmless because it focused on engineering tools rather than AI systems themselves, he wrote. Over time, he concluded that the tools’ main effect would be to accelerate AI-chip design. Faster, cheaper hardware, he argued, could also support more capable systems by making it practical to spend more computing resources on inference.

He had discussed his reluctance to see the project succeed with his manager. O’Callahan described the decision as difficult: he valued his colleagues and work, and said his Google position in New Zealand offered an unusually good arrangement. He also tried, he said, to influence events from within Google DeepMind, but did not think that effort would have enough effect.

Another Google team was not practical, in his account

The post does not identify O’Callahan as a Google DeepMind employee. It refers to his attempt to influence events “from within GDM,” while describing his employment more broadly as work for Google on chip-design tools. That distinction matters: his stated reason for leaving was the expected effect of his team’s work, not a claim that he built AI models at DeepMind.

Moving to a Google project that would not accelerate AI was impractical because, in his account, the company had no other engineering teams in New Zealand. His location and his obligation to hand over work responsibly shaped the timing of his departure, he said.

The date was planned ahead of a backpacking trip

O’Callahan picked the date before a planned ten-day backpacking trip with friends, which was due to start the following Monday. The timing was not a response to recent viral resignations or warnings calling for slower AI development, he explicitly said. His departure came amid that wider discussion, but he presented the overlap as coincidence.

His broader argument is that AI development is moving faster than people and institutions can adapt. He described the risks as real but uncertain, and raised concerns ranging from economic disruption and cybersecurity to concentration of power and AI-related harms to people. He also believes AI capabilities will keep advancing if development continues.

Pernosco and rr remain on his agenda

O’Callahan plans to continue maintaining Pernosco, a debugging tool, and rr, a recording-and-replay debugger. He also wants to study how AI systems debug code and whether debugging tools can help them. His post leaves his longer-term plans open, but says he wants future work to be “unambiguously pro-human.”

The resignation puts a specific question on the table: whether engineers should judge their work only by its immediate function, or also by what it enables downstream. O’Callahan’s account is personal, not a measure of the project’s actual effect on chip development. But his stated reason for leaving is concrete: he believed the tools he helped build would make AI cheaper and faster, and chose to stop contributing to that work.