Introduction: The key question for 2026
If you're wonderingwhat prompts to use for business automationin the era of multi-agent models, the short answer is: prompts that leverage the autonomy of systems like Agensh, Carbon-X’s genomic analysis, and Nanbeige4.3-33B’s code generation. In this article, you’ll discover a practical method for designing effective prompts, real-world use cases, and ready-to-use code snippets.
Why multi-agent prompts are revolutionizing business automation
In 2026, multi-agent models like the recently open-sourceAgensh(a harness that coordinates over 1,000 agents without a central orchestrator) are game-changers. Each agent can perform a specific task, share a common Git repository, and communicate via chat, creating a more dynamic workflow than older single-LLM systems.
Concrete benefits for businesses
- Scalability:Add new specialized agents without redistributing the entire model.
- Modularity:Replace or update a single agent (e.g., the data analysis agent) without disrupting the overall process.
- Problem resolution:If one agent fails, the others continue working, increasing reliability.
- Speed:Tasks are distributed in parallel, reducing response times from minutes to seconds.
How to build effective prompts for multi-agent models like Agensh
Designing prompts for a multi-agent ecosystem requires clear task definition and structured communication between agents. Follow this step-by-step approach:
Step-by-step: from goal definition to production
- Define the business goal.
- Identify specific tasks.Extract subtasks such as data extraction, cleaning, analysis, and visualization.
- Assign each task to a specialized agent.Use Agensh’s predefined agents (DataMiner, TextProcessor, etc.) or create new agents with specific skills.
- Write a main prompt that coordinates the agents.The prompt should include:
- Role of each agent.
- Required input.
- Output format.
- Error flags and fallback actions.
- Test the end-to-end flow.Run the prompt multiple times, monitoring outputs and timing.
- Optimize and version.Save effective prompts in a dedicated Git repository for version control.
Here’s an example of a main prompt for sales report automation:
# Main prompt for Agensh - Sales report
# Roles:
# - DataMiner: extracts raw data from CRM and databases
# - Cleaner: cleans and normalizes data
# - Analyst: calculates KPIs and trends
# - Visualizer: generates PNG charts
Goal: Generate a weekly sales report in PDF.
Input: None (uses predefined CRM datasets)
Steps:
1. DataMiner extracts last week's sales data.
2. Cleaner removes duplicates and fills missing values.
3. Analyst calculates GMV, % growth, sales by region.
4. Visualizer creates bar charts for each region.
5. Assembler combines analysis and images into a PDF.
Output: Send the PDF to the specified email and log completion.
Fallback: If an agent fails, notify the manager and use the backup dataset.Leveraging specialized models: Carbon-X for data analysis and Nanbeige4.3-33B for development
While multi-agent models handle the workflow, specialized models bring depth. Two 2026 innovations areCarbon-X(a 14-billion-parameter language model for DNA) andNanbeige4.3-33B(a 33-billion-parameter agentic programming model).
Prompts for genomic data processing with Carbon-X
Carbon-X can read raw DNA sequences and respond in natural language. To integrate Carbon-X into a business automation (e.g., analyzing genetic data for product personalization), use this prompt:
# Prompt for Carbon-X - GWAS variant analysis
Role: You are Carbon-X, a language model for DNA.
Input: VCF file with SNP variants for cohort X.
Goal: Identify loci associated with [disease/target].
Instructions:
1. Read the input VCF file.
2. Convert each genotype to descriptive text.
3. Analyze associations using your integrated knowledge graph.
4. Return a JSON with: {locus, p-value, gene, function}.
Format the output as a JSON array.Prompt for code generation with Nanbeige4.3-33B
Nanbeige4.3-33B excels at writing agentic code and self-correction. To generate a RESTful microservice that exposes an /analyze endpoint, do this:
# Prompt for Nanbeige4.3-33B - Microservice generation
Goal: Write a Node.js microservice that exposes POST /analyze.
Input: JSON with a "data" field (array of numbers).
Output: Calculate mean, standard deviation, and return JSON.
Instructions:
1. Create a Node.js project with Express.
2. Implement /analyze as an endpoint.
3. Use built-in data analysis to calculate mean and std.
4. Add error handling (400 for invalid input).
5. Write unit tests for the two endpoints.
Return the full code with README.Practical example: automating the customer onboarding process
Let’s put the pieces together for a real-world use case: automatically onboarding a new customer in a SaaS company.
- An agent (DataMiner) extracts registration form fields from HubSpot.
- The verification agent calls a third-party service (e.g., DocumentAI) via a prompt that includes the document number.
- The data analysis agent sends the data to Carbon-X to check for anomalies in customer data (e.g., unusual address patterns).
- The text generation agent uses Nanbeige4.3-33B to create a personalized welcome email with integration code.
- The Agensh orchestration agent deploys the new user to the user database and logs the onboarding completion.
The orchestrator prompt that coordinates these agents could be:
Common mistakes and how to avoid them
Over-engineering prompts
Lack of context in multi-agent models
Forgetting to pass context (e.g., customer ID) between agents causes correlation errors. Use explicit variables like {{customer_id}} and ensure each agent receives them.
Conclusion: Transform your processes with the right prompts
The multi-agent models of 2026 offer unprecedented automation, but the key is knowingwhat prompts to use.Design modular prompts that assign specialized tasks to dedicated agents, integrate specialized models like Carbon-X and Nanbeige4.3-33B when analytical depth is needed, and follow a rigorous testing and versioning cycle.
Start with the example main prompt for sales reports, adapt it to your domain, and watch speed and reliability improve within a week.
Next step:Choose a manual process in your company, map out the subtasks, and write your first orchestrator prompt with Agensh. Your future automation team will thank you.
IMPORTANT: We are in 2026. All temporal references must be current. Do not mention past years like 2024, 2023, etc. The content must be fresh and relevant to the present. CURRENT CONTEXT (October 2026): Here are some recent trends and news to inspire you: - Anthropic Launches OSS Scanner, a Free AI Vulnerability Scanner: Anthropic has launched OSS Scanner, a free, opt-in vulnerability scanner for critical open-source projects. The Anthropic OSS Scanner runs the company... [2026-10-25] - Nanbeige Releases Nanbeige4.3-33B, a 33B Open Agentic Coding Model: Nanbeige LLM Lab released Nanbeige4.3-33B, a 33B dense Apache 2.0 model with 512K context. It reports 80.9 on Terminal-Bench 2.1 and will teach the up... [2026-10-25] - Researchers from Stanford and UC Berkeley Introduce TokenGate: Token-Gated Attention Makes Long-Context Prefill Up to 11.3x Faster: Stanford and UC Berkeley researchers released TokenGate, a drop-in learned token pruning module for LLMs. It cuts long-context prefill time-to-first-t... [2026-10-25] Use this current information as inspiration to create an original and relevant prompt for 2026.