9. Securing Execution with a Harness

An execution harness defines the secure sandbox environment where an agent operates, restricting access to approved commands and prompting human approval before executing sensitive scripts.

8. State Tracking with Checkpointers

Checkpointers log the state of an agent at every execution step. If an error occurs, the agent can recover from the last saved state rather than restarting the entire task from scratch.

7. Memory Systems in Agentic AI

Standard chatbots lose context once the session ends. Agentic AI features short-term memory (session history) and long-term memory (storing past experiences in vector databases) to learn and evolve over time.

6. Governing AI: Rules and Prompts

System prompts define the operational boundaries of an agent. They specify the persona, strict rules, and target tasks to ensure safe and predictable behavior. **Prompt Design** A high-quality prompt contains clear formatting rules, exceptions, and instructions for human-in-the-loop triggers if the agent encounters high-risk operations.

5. The Importance of Resources for Agents

Just as chefs need knives (tools) and ingredients (resources) to cook, Agentic AI needs both tools and resources to perform tasks. **What is a Resource?** Resources are the data inputs read by the agent, such as: * Text files, PDFs, and documents. * Database records. * JSON payloads from APIs. Using protocols like MCP, agents dynamically pull and process these raw materials to complete tasks.

Tech.Document your engineering breakthroughs.

A high-performance Knowledge Management System. Track daily bug fixes, preserve code snippets, and build a demonstrable Full-Stack portfolio.

★ Featured Posts

Ye Aung

•Jul 8, 2026•

9. Securing Execution with a Harness

An execution harness defines the secure sandbox environment where an agent operates, restricting access to approved commands and p...

Ye Aung

•Jul 8, 2026•

8. State Tracking with Checkpointers

Checkpointers log the state of an agent at every execution step. If an error occurs, the agent can recover from the last saved sta...

Latest Articles

Ye Aung

•Jul 8, 2026•

7. Memory Systems in Agentic AI

Standard chatbots lose context once the session ends. Agentic AI features short-term memory (session history) and long-term memory...

Ye Aung

•Jul 8, 2026•

6. Governing AI: Rules and Prompts

System prompts define the operational boundaries of an agent. They specify the persona, strict rules, and target tasks to ensure s...

Ye Aung

•Jul 8, 2026•

5. The Importance of Resources for Agents

Just as chefs need knives (tools) and ingredients (resources) to cook, Agentic AI needs both tools and resources to perform tasks....

Ye Aung

•Jul 8, 2026•

4. AI Instruments: Understanding Tools

Standard LLMs only output text. They cannot browse the web, write files, or send emails on their own. Agentic AI uses 'Tools' to p...

Ye Aung

•Jul 8, 2026•

3. Multi-Agent Systems and the Role of Sub-agents

Managing a large project alone is difficult and error-prone. Similarly, handing all tasks to a single AI agent can lead to context...

←