What Self-Improving AI Means For Public Affairs
Nearly every day, a new AI model comes out that changes what's possible. Now the labs say the AI itself may set the next bar. It’s called recursive self-improvement, and the people closest to it think it’s arriving sooner than anyone planned. In today’s short read, we break down what it is, why Silicon Valley thinks it’s nearly here, and what it means for the technology running your public affairs shop.
What Is Recursive Self-Improvement?
Recursive self-improvement, or RSI, is AI building the next, smarter version of AI. Then that smarter version builds an even smarter one. The cycle keeps repeating. The worry is spiraling capability, as better systems build better systems.
Think of it like compound interest. Your deposit earns interest. Then the interest earns interest. The balance doesn’t climb in a straight line. It bends upward. RSI is compound interest on intelligence. Today, humans still pick the research direction and grade the results while AI does a growing share of the work. Full RSI is the moment AI takes the steering wheel, too.
Silicon Valley Thinks It’s Close
This isn’t sci-fi chatter anymore. The labs are publishing receipts. Anthropic co-founder Jack Clark now expects full RSI in 2027, a year earlier than his previous late-2028 call. Claude agents reportedly led 26% of core AI research tasks at Anthropic in August, with roughly 30,000 running at once. OpenAI is singing the same tune. Chief scientist Jakub Pachocki says he strongly expects today’s pace to carry into RSI.
In June, Clark and Anthropic Institute lead Marina Favaro urged the major labs to consider a coordinated, verifiable pause, warning that AI could soon improve itself faster than society can manage the risks. Their proposal was akin to nuclear arms control: inspectors who verify compliance instead of taking anyone’s word for it. In September, CEO Dario Amodei went further, offering outside evaluators desks, badges, and access to Anthropic’s training and safety procedures, plus the right to publish what they find.
What This Means For Public Affairs
Let's set the regulatory fight aside for now and focus on the practical implications for public affairs practitioners. When AI is helping build AI, the gap between today's tools and tomorrow's constantly widens. Your tech stack either rides that wave or gets buried by it.
What's stunning in public affairs is how many tools the industry still uses that are barely technology at all: black-box data sets that claim to contain a decision-maker's influencers; white-labeled chatbots that repackage a subscription you're already paying for; so-called ad "targeting" with no reporting transparency or verification. If you're still using these tools, you're getting left behind by today's AI, and that gap is widening by the day.
Two Ways To Operationalize This Email
🔱 Audit your entire stack in Q4. Budgets are being set and renewal decisions are landing soon. List every tool in your tech stack and what you pay for it. You can’t fix what you haven’t counted.
🔱 Do not sign long-term contracts. In the AI age, technology is moving too fast to bet on any one solution. Public affairs teams need maximum flexibility to evaluate the latest AI tools and decide what is best for their workflows. Insist on per-project pricing for any public affairs technology.