This week marked one of the most exciting and challenging parts of my DevOps Micro Internship (DMI) Cohort 3 with Agentic AI. Before this week, I viewed AI coding assistants as tools that simply answered questions or generated code. After spending time with Claude Code, I now understand that modern AI development assistants can become active collaborators when provided with the right context, permissions, and project knowledge.
One of the biggest topics I explored was Claude Code itself. Rather than only asking questions, I learned how Claude Code interacts directly with a project, reads files, performs validations, executes commands, and follows instructions stored within the project. This changed how I think about AI-assisted development. It is no longer just about prompting—it is about creating an environment where AI can work effectively and safely.
Another major topic was Hooks. I learned how hooks allow specific actions to occur before or after commands are executed. During the assignment, I created custom hook scripts that validated operations and logged Terraform commands. This introduced me to an important DevOps principle: automation should not only make work easier but also improve consistency and reduce mistakes.
I also gained practical experience with permissions. Rather than allowing every command automatically, Claude Code can request approval before executing sensitive operations. Understanding why these safeguards exist helped me appreciate the balance between automation and security in modern development workflows.
Perhaps the most memorable part of the week was working with Claude's memory system.
At first, the assignment appeared straightforward. I created a project memory containing CSS design constraints, mobile breakpoints, and a rule to never suggest JavaScript for the project. However, when I started a new Claude session, the assistant ignored the memory and searched through my CSS files instead.
Instead of assuming something was broken, I began investigating.
I learned where Claude stores project memories, how MEMORY.md serves as an index, and how project-specific memories are organized under the .claude/projects directory. I also learned the importance of comparing expected behavior with actual behavior before making changes. Although it took persistence to understand what was happening, the experience strengthened my debugging skills far more than if everything had worked perfectly on the first attempt.
Throughout the week, I also became more comfortable using Git to inspect staged changes, manage project files, and ensure that only the required files were committed. Rather than simply using git add ., I learned to carefully review what would become part of the repository and why.
This week taught me an important lesson: successful DevOps engineers are not defined by how quickly they find answers but by how methodically they investigate problems. Modern AI tools are incredibly powerful but understanding how they operate behind the scenes makes them even more valuable.
One habit I plan to implement going forward is to investigate before assuming. Whenever something behaves differently than expected, I want to verify the evidence, understand the system, and identify the root cause instead of immediately searching for a workaround. I believe this mindset will make me a better engineer over time.
Week 2 was more than learning new tools. It strengthened my confidence in debugging, automation, AI-assisted development, and systematic problem-solving. I am looking forward to applying these lessons as I continue my DevOps journey.
