May 26, 2026

4 Minute Read or Listen ->

🔱 Sign up for a free demo today!

See how the latest AI and digital advocacy tools can transform your work.

THIS WEEK'S EDITION

The Three Worlds Of AI Are Coming Into Focus

The defining feature of the current agentic AI era is compute scarcity—meaning that even the most advanced AI labs in the world don't have enough computational power to serve current demand. The market response is already underway, quietly sorting organizations into three worlds of AI. In today's short read, we show you what's emerging, and which world most public affairs organizations should be targeting.

Compute Scarcity Defines The Worlds

The agentic AI era has one defining feature that is creating three AI worlds: there is not enough compute to go around. The most powerful AI labs in the world are literally turning down business because they can't serve current demand.

Last month, OpenAI's CFO Sarah Friar said it plainly: "We're making some very tough trades at the moment and things we're not pursuing because we don't have enough compute." Anthropic is in the same position. For most of 2026, Claude users have hit tighter quotas, rate limits, and peak-hour throttling. More such changes are certainly on the way.

The bet behind the AI buildout—largest infrastructure investment in history—is that the industry will eventually build its way out of the compute crunch. Maybe. But "eventually" is a 24-to-36-month solution at best—and the agentic era, which consumes 10x to 100x more compute per user than the chat era ever did, is here now.

The Efficiency Escape Hatch

Quietly, however, the market is tackling the compute crunch differently—by using less compute, more intelligently. Three approaches are doing the heavy lifting:

🔱 Context: As we covered last week, feed a model your documents, your CRM, your history, your domain knowledge—and a smaller, cheaper model can outperform a larger, generic model.

🔱 Local and open-source models: Google's Gemma 4 and NVIDIA's Nemotron 3 family can run on some laptops, a workstation, or an office server—no data center round-trip required. They're not as powerful as the frontier models alone, but when given context, they can be even stronger. These are just two of the open source models rapidly catching up to the flagship models that get almost all of the attention.

🔱 Mixture of Experts architectures: This is a model that activates only the specialists relevant to each question, not the entire network. Think of a hospital: instead of every specialist examining every patient, a routing system sends each question only to the right experts and leaves the rest of the building idle. The result is vast capacity at a fraction of the operating cost.

The combined effect of these and other efficiency-first approaches can be dramatic. Fin Apex (a customer service LLM built on an open-source model) runs at roughly 1/5 the compute of frontier models, with better results. Smaller models built on similar principles run at 1/10—and in some cases as little as 1/100—of frontier compute, while still delivering frontier-quality results in their domain.

The Three Worlds Of AI

Step back from the technical details and a strategic picture starts to emerge. Three worlds of AI are taking shape, and organizations are sorting themselves into one of them whether they realize it or not.

World One: Subscription Land

This is where almost every public affairs organization currently lives. Pay a per-seat license to OpenAI, Anthropic, or Google. Get a powerful but generic frontier model. The limits of this world are significant:

🔱 Token and context limits. Every conversation has a ceiling. Heavy users hit it constantly—the vibe coding era of unlimited experimentation is over.

🔱 Rate limit throttling. As the labs ration their own scarce compute, you get cut off—especially during peak hours, including in the middle of critical workflows.

🔱 A ceiling on the value you can create. The model knows only what it is trained on, not the things that make your work distinctive—your stakeholders, your past advocacy, your human intelligence. So it ends up doing generic versions of work that demands specialization.

If you live exclusively in World One, you are paying premium prices for throttled access to a generic model. 

World Two: In-House Engineering

The most sophisticated companies are building their own way out of the compute crunch. They engineer proprietary context layers, deploy open-source and local models on their own hardware, post-train models on their proprietary data, and use Mixture of Experts architectures to slash compute costs. World Two requires real engineering muscle, real budget, and real organizational commitment. For almost every public affairs organization, this world is out of reach. If you're too small, you can't afford it. If you're too big, you can't get the organizational buy-in. 

World Three: Purpose-Built Tools

This is the world most organizations will eventually live in—and the one most public affairs teams should be targeting now. You don't build the AI yourself. You buy purpose-built AI tools from vendors who have already done the engineering work for your specialized domain.

These tools combine the best of Worlds One and Two. They use frontier subscription models under the hood when needed, layer in proprietary context and domain-specific post-training, and run on efficient architectures. They're built for your job, not generic chat.

Fin is the leading example in customer service, but the same pattern is now emerging across many domains—including public affairs. These are the vendors building specialized public affairs tools, not the ones white-labeling a chatbot.

What This Means For Public Affairs

If you're still in World One, it means you missed the vibe coding era—meaning you didn't build and operationalize your own custom tools when compute was abundant. If you stay in World One, you're only going to fall further behind. You will pay rising prices for tighter throttling on tools that don't know your work. World Two is technically possible but practically unreachable for most teams. World Three—buying purpose-built tools from vendors who have already done the engineering—is where your leverage is.

Two Ways To Operationalize This Email

 🔱 Audit which AI world you currently live in. If your AI strategy is just a stack of ChatGPT, Claude, and Gemini subscriptions, you are in World One—and falling further behind by the day.  

🔱 Move beyond the headline tools in evaluating AI. Instead of just using the latest frontier model, seek purpose-built tools for your specific workflows. Ask your vendors: What context do they pull from? What public affairs domain training have they done? How do they keep that institutional knowledge current?

🔱 If you're enjoying this content, please consider forwarding this email to a colleague or friend. 

🔱 If you're not already a subscriber, please sign up here to stay up to date on the latest developments in political technology.

🔱 If you have suggestions for future newsletters or any other feedback, we'd love to hear from you.