Ben MannTech Lead, Product Engineering

It’s a sign of the times. The technology is extremely valuable, and the industry is growing fast. At Anthropic, we’ve been less affected because people are mission driven they choose between just making money or shaping the future of humanity.

They sound huge, but compared to the value a single person can add, they’re actually cheap. If someone improves our systems by even a few percent, the revenue impact can be enormous.

No. Progress is accelerating. Model release cycles are shorter, and scaling laws still hold. People mistake rapid iteration for slower gains, but the advances are very real.

I prefer the term 'transformative AI.' It’s not about matching humans at everything it’s about causing large scale societal and economic transformation. The Economic Turing Test is one way to measure that.

It’s already here. In customer service, AI resolves over 80% of tickets without humans. In coding, our tools write most of the code, allowing smaller teams to achieve more.

Learn to use the tools ambitiously. Don’t treat them like old tools push for big changes, experiment, and adapt quickly. Those who adapt fastest will thrive.

Yes. Our legal team uses Claude to redline documents. Finance uses it to run analyses on customers and revenue. The productivity gains are huge.

We felt safety wasn’t prioritized enough there. Our founding team came from the safety group and wanted a company where safety was the top priority while staying at the frontier.

They’re not opposites. Working on safety often improves capabilities. Claude’s trustworthiness and personality are direct results of our alignment research.

It’s training AI with guiding principles like human rights baked in from the start. The model critiques and improves its own outputs against these values.

Because transparency builds trust. Policymakers and the public deserve the straight talk, not sugarcoating.

We avoid hyping unsafe tech even when it could grab headlines. Our focus is ensuring benefits outweigh risks by addressing problems early.

Forecasts like the AI 2027 report point to 2028 as a 50% probability. The exponential growth in compute, algorithms, and infrastructure makes it realistic.

Compute chips and data centers plus breakthroughs in algorithms and efficiency. All three drive progress.

I treat 'busy' as the normal state. It's a marathon, not a sprint. Being surrounded by mission driven people helps.

Starting the Labs now Frontiers team. It bridges research and products, creating tools like Claude Code and Model Context Protocol.

You don’t need to be an AI scientist. Roles in product, finance, operations all help drive the mission forward.