What If You Could Have a Team That Never Sleeps?
Think about the last time a problem in your business felt genuinely impossible. Maybe your supplier keeps missing deadlines, or your best-selling product suddenly stopped moving, or your team spends hours every week on tasks you know should be automated. You’d love to command a team that could test every possible fix, check each other’s work, and come back with a solution — but you have customers to serve and orders to ship. You don’t have 30 hours to think through every angle.
This week, Anthropic announced something that should change how you think about AI. An unreleased AI model made real progress on the Riemann hypothesis, a math problem that has stumped humans for over 150 years. The model didn’t just crunch numbers faster. It acted like a whole research department: testing hundreds of approaches, delegating tasks to specialised sub-agents, and writing up its own findings. The full story is on TechCrunch (source), but the lesson for you isn’t about prime numbers. It’s about what AI can now do on your behalf.
TL;DR: An unreleased Anthropic model made a significant advance on the Riemann hypothesis by working autonomously for 1.5 days, coordinating 60 sub-agents and testing 650 ideas. This proves that AI is no longer limited to simple tasks — it can tackle open-ended, complex problems and find novel solutions. For Malaysian SME owners, that means the next wave of business automation isn’t just about scheduling posts or answering FAQs. It’s about giving you a tireless problem-solving partner.
What This Means
The Riemann hypothesis is about the distribution of prime numbers — those numbers like 2, 3, 5, 7 — and whether there’s a pattern in how they appear. Nobody has been able to prove it conclusively. According to the TechCrunch report, Anthropic’s model made progress by pushing the lower bound for which the hypothesis is known to hold significantly further than ever before. It didn’t solve the problem completely, but it took a real bite out of a 150-year-old mystery.
How it worked is the part that matters to you. An Anthropic staff member without deep math training simply prompted the model to “take a real stab” at proving it, then left it alone. Over the next day and a half, the model spun up 60 sub-agents (source) — think of them as 60 different consultants. Thirty of those consultants tried and failed to produce new ideas. Thirteen contributed ideas that helped. Thirteen were validators, checking the correctness of the work. Two developed the key breakthroughs, and two wrote the initial paper. All together, they consumed 31 million output tokens (source).
AI is no longer just answering questions on command. It is now capable of exploring possibilities, coordinating its own team, and coming back to you with a result — like having a remote team that never sleeps, never gets tired, and never asks for annual leave.
This is what the industry calls “agentic AI” — an AI that doesn’t just respond to prompts, but breaks down a large goal into subtasks, decides how to approach them, and works through them systematically. The way it worked here mirrors how a good project manager handles a big assignment: assign roles, iterate, validate, deliver. Only this project manager works around the clock.
How This Applies to Malaysian SMEs
You don’t need a PhD in math to see how this changes things for your business. The same agentic approach can be applied to problems that come up every day in Malaysia — from optimising delivery routes in Klang Valley during peak hours, to figuring out why your livestream sales dipped among Sabah customers, to planning your next quarter’s stock orders across a dozen suppliers. You can prompt an AI to “take a real stab” at your problem, leave it running overnight, and wake up to a detailed report with several possible solutions — all with reasoning and validation included.
Imagine you run a catering company in Penang. You’re trying to reduce food waste across 15 weekly events, each with different guest counts and menu preferences. Instead of hiring a data analyst on a three-month contract, you could task an AI agent with the problem. The AI would set up sub-agents to analyse historical sales data, study the patterns of last-minute RSVPs, generate 20 different ordering strategies, and have another sub-agent double-check the math on each. Then it would present you with the safest option and the most aggressive one, along with the predicted impact. That’s not science fiction. That’s what the model did with the Riemann hypothesis, just applied to catering numbers instead of prime numbers.
The most practical part is what this means for your daily workflow. In the past, AI tools were like a junior assistant — you gave them a clear instruction and they followed it. Now, AI can be like a thinking partner. You can describe a messy, open-ended business problem in plain language — “our rental business in Johor Bahru is profitable, but we don’t know why we keep losing customers in the second month” — and let the AI decide how to investigate. The article notes that the key ideas came from just two out of 60 sub-agents, and they arrived after many other attempts failed. This trial-and-error approach is exactly how small businesses learn, but now a machine can do it without draining your team’s energy.
There’s also a less obvious lesson: the prompt was written by a staff member without significant mathematical training. You don’t need to be an expert in your domain to use this tool. You just need to know which problem is worth solving. For a Malaysian SME owner, that’s liberating. You already know your business’s pain points. AI agents can now help you find the answer — even if you can’t articulate the perfect question. Start with a broad prompt, let the AI do the deep thinking, and check the results.
Practical Takeaways
- Pick one open-ended problem. Choose something you’ve been putting off — like why certain repeat customers stop buying. Write it down in plain language.
- Let the AI work unattended. Agentic AI thrives when left alone for a period. Give it a night or a weekend, not just a 5-minute chat session.
- Break tasks into roles. If your AI tool supports sub-agents or “multi-agent” mode, use it. Ask for one agent to gather data, another to generate ideas, another to verify the facts.
- Check the output. The model used validators to check correctness — you should too. Have a team member review the AI’s suggestions before you implement them.
- Expect failures. In the math breakthrough, 30 of 60 agents tried and failed. That’s normal. Look for the few good ideas that emerge from many bad ones.
The Bigger Picture
This breakthrough in mathematics is a preview of how business automation will evolve over the next few years. We’re moving from AI that executes predefined workflows to AI that can define its own workflows. That’s a big deal for Malaysian SMEs, where you often cannot afford a huge team of specialists. Agentic AI can act as your strategy consultant, your data analyst, and your quality controller — all in one subscription.
There are still unresolved questions. The article mentions that a group of prominent mathematicians signed a declaration calling for AI-generated proofs to be attributable to human authors. That debate matters for science, but for your business, it’s simpler: you remain the owner, the maker of final decisions, and the one who takes responsibility. AI is a tool that brings you options and evidence, not a replacement for your judgment.
The long-term impact is this: the best way to stay competitive isn’t just doing things faster — it’s being able to think through more possibilities. That’s exactly what the Anthropic model demonstrated. It tested 650 ideas (source). When was the last time your business tested 650 variations of an idea? With agentic AI, you can.
Key Numbers That Show the Scale
| Metric | Value | What It Means for You |
|---|---|---|
| Ideas tested | 650 | A broad search space, not just one or two “best guesses” |
| Sub-agents used | 60 | Parallel work across many different approaches |
| Output tokens | 31 million | Equivalent to many long documents of reasoning |
| Key breakthrough agents | 2 | Good ideas are rare, but AI can find them |
| Failed attempts | 30 agents | Failure is part of the process — like testing hypotheses |
If you’ve been treating AI like a clever assistant that writes emails and creates images, it’s time to raise your expectations. The same model architecture that pushed the boundaries of mathematical research can be pointed at your business’s stubborn problems. The tools may not all be available to the public yet — this model is unreleased — but the direction is clear. When it does become available, the companies that start experimenting first get the biggest advantage.
Start small. Pick one problem this month. Write a prompt that asks for a full investigation. Let the AI try many approaches. Check the results. Then do it again. That’s how you turn a mathematical breakthrough into a business advantage.
Ready to Streamline Your Operations?
Your business should run itself. AutoRunBiz deploys AI agents to automate your daily operations — WhatsApp orders, invoicing, customer follow-ups, and accounting. Book a free 15-min ops audit to see where automation fits your business →