How to write effective prompts for business automation with new LLMs
As we move into 2026, businesses are seeking rapid ways to integrate AI into their daily operations. The key to success lies in crafting prompts specifically designed for modern LLMs. These models can process contexts of up to 1 million tokens and interact with physical systems, such as JD.com’s logistics robots.
Why prompts are the cornerstone of modern AI automation
Prompts have evolved beyond simple inputs; they now serve as the interface that translates human intent into actionable steps for AI agents. With advanced models like DeepSeek-V4.1-Flash, which offer a 1M token context and enhanced cross-layer attention reuse, well-structured prompts can fully leverage extended memory capabilities and reduce the burden on KV caches. This leads to faster response times and more reliable automation for complex tasks.
How to structure prompts for LLMs with extended context
When designing prompts for LLMs with extended context, follow this three-part structure:
- Goal:Clearly define the desired action (e.g., optimize delivery routes).
- Context:Provide relevant data within the token budget (e.g., supply chain status, logistics constraints).
- Constraints:Specify the output format, limits, and allowed actions (e.g., JSON, maximum 5 variants).
Here’s an example prompt for an AI agent managing the supply chain:
<prompt>
Goal: Generate 3 optimized delivery routes that reduce total travel time by 15% compared to the default route.
Available data (up to 500k tokens):
- Current location of each vehicle
- Real-time traffic conditions
- Capacity and temperature constraints for each shipment
Output format: JSON with fields {"route", "total_time", "savings_percentage"}
Constraints: Maximum 3 routes, each route cannot exceed 8 hours.
</prompt>This approach harnesses the model’s extended memory to consider the entire logistics network without repeating context at each step.
What prompt to use for optimizing the supply chain with AI agents
Modern AI agents can operate autonomously once given a clear prompt. For a supply chain optimization agent, include a feedback loop that enables the agent to propose, test, and iterate on routes.
<prompt>
You are an AI agent specialized in supply chain optimization. Each cycle, you must:
1. Analyze input data (positions, traffic, constraints).
2. Propose a set of candidate routes.
3. Calculate time savings for each candidate.
4. Select the best route.
5. Provide a brief justification.
If you cannot achieve a 10% savings, generate new candidates.
Output format: JSON {"selected_route": "...", "savings": "X%", "justification": "..."}
</prompt>This purpose-driven prompt enables the agent to make repeatable and traceable decisions, reducing the need for manual updates.
When to use logistics automation prompts with physical robotics
Physical robotics, such as JD.com’s 3 million robots, require prompts that coordinate both digital and physical actions. Use prompts that include commands for IoT devices and real-time execution status.
<prompt>
Coordinate sorting robots in the logistics center to complete order #12345 within 5 minutes.
Robot information:
- Robot A: capacity 20kg, battery 85%
- Robot B: capacity 15kg, battery 40%
- Robot C: capacity 30kg, battery 10%
Constraints: Use only robots with battery >=30% to avoid downtime.
Output: Sequence of actions with robot ID and action (pick, move, release).
</prompt>By integrating prompts with sensor data, operations become end-to-end and adapt quickly to changing conditions.
Best practices and checklist for business automation prompts
- Goal clarity:Start with a single action verb (e.g., generate, optimize, coordinate).
- Context control:Keep data within the token limit; use summaries for very long sections.
- Format definition:Always specify the exact output format (JSON, CSV, Markdown).
- Constraints and security:Include limits (e.g., budget, time) and security rules (e.g., do not disclose sensitive data).
- Feedback loops:Design prompts that allow the agent to iterate and correct itself.
- Testing and monitoring:Run A/B tests on prompts to measure accuracy and speed; monitor KV cache performance.
Conclusion
Prompts serve as the bridge between generative AI and operational efficiency. By crafting prompts that leverage new 1M token LLMs, coordinate AI agents for supply chain management, and integrate physical robotics, companies can achieve tangible automation as early as 2026. Start with a well-structured prompt, test it quickly, and scale with data-driven feedback loops.
Key Takeaways
- Always define a clear goal, context, and constraints in your prompt.
- Leverage models with extended context to reduce repetitive prefill and optimize KV cache usage.
- Use agent-driven prompts for autonomous supply chain optimization.
- Combine prompts with IoT data to directly control physical robots.
- Implement a testing checklist to ensure prompts remain effective as workloads grow.