Last week, Anthropic announced something quietly revolutionary: an unreleased model made progress on the Riemann hypothesis, one of mathematics' 150-year-old unsolved mysteries with a $1 million bounty on it.
But here's what matters to founders: how it actually solved it.
The model didn't solve Riemann alone. It coordinated 60 subagents, each with a specific role. Two agents developed core mathematical ideas. Thirteen contributed supporting ideas. Thirty explored unproven approaches. Thirteen validated correctness. Two synthesized and wrote the paper.
In 1.5 days, testing 650 different ideas, this distributed team made more progress than centuries of human mathematicians working solo.
This isn't just math. It's your blueprint for solving pricing strategy, go-to-market plans, and technical architecture decisions.
The Problem With Solo Founder Thinking
Most founders solve hard problems alone. You think hard. You believe your first good idea. You execute. You hit a wall you didn't anticipate.
Why? Because one brain has one perspective. When you're thinking about pricing, you're seeing it from an economics angle. You miss the psychology angle (what will customers actually pay?). You miss the finance angle (is this sustainable?). You miss the sales angle (can your team explain this?).
Anthropic tested 650 ideas on the Riemann hypothesis. Most founders test 10-20.
The Multi-Agent Approach
Here's the agent breakdown from Anthropic:
Strategic core (2 agents): Develop the key ideas and set direction.
Contribution layer (13 agents): Generate supporting ideas for the core agents.
Exploration layer (30 agents): Test unproven approaches (most fail, some spark new directions).
Validation layer (13 agents): Check if the logic actually works.
Synthesis layer (2 agents): Write it up and communicate clearly.
Each agent knows its job. No conflicting priorities. Clear feedback loops.
Now apply this to your startup's hardest problem.
Example: Pricing strategy.
You need strategic agents asking "What maximizes revenue while staying competitive?" You need explorers asking "What if we tried per-transaction pricing?" You need validators asking "Is this financially viable?" You need synthesizers asking "Can our sales team actually explain this?"
One founder brain answers one of these questions. Five agent roles answer all of them—simultaneously.
How to Actually Do This
Start simple. Pick your hardest problem (pricing, GTM, technical architecture, hiring). Define 3-5 agent perspectives on that problem.
Use Claude API (or Claude.ai Pro for testing). Set up a system prompt that defines each agent role, then prompt them to work through your problem across 2-3 turns.
Cost? $5-50 in Claude tokens for a complex problem. (Compare to $5,000+ for a consultant who gives you one opinion.)
Time? 1-3 hours instead of weeks of solo overthinking.
The key: You're not asking AI to solve the problem. You're asking AI agents to explore the problem from multiple angles. Then you—the human—validate and decide.
Why This Works
The Riemann result wasn't magic. It was decomposition and delegation at scale.
When Anthropic ran this, they weren't hoping for a breakthrough. They structured the work so breakthroughs became likely.
Most founders hope for good decisions. Structured founders engineer good decisions.
Multi-agent thinking gives you:
- Multiple perspectives (not one opinion)
- Massive idea exploration (650 vs 10)
- Built-in validation (agents check each other)
- Clear reasoning (you see how they arrived at conclusions)
- Actionable synthesis (written up and ready to implement)
The Next Step
You don't need to understand advanced AI prompting. You need to understand decomposition.
What's your hardest founder problem right now? Pricing? Go-to-market? Technical direction? Hiring?
What perspectives does it need? (Market perspective? Psychology? Finance? Sales? Execution?)
Who should challenge your assumptions? (Define agent roles.)
How would different perspectives approach this? (Run them through the problem.)
Then validate with humans and execute.
Anthropic proved that hard problems aren't solved by thinking harder. They're solved by thinking from multiple angles simultaneously.
Your business deserves that same rigor.
For more on building with AI, visit Bitroot.