Quick answer: Generative AI is commonly used to produce text, code and other content. Agentic AI adds goal-oriented workflow control, tool use and intermediate decisions. Agentic systems often use generative models as one component.
A simple example
A generative assistant might draft a bug-fix checklist after reading an error. An agent-based workflow can gather approved logs, categorize the failure, check relevant documentation and assemble a recommended remediation plan for human review.
Typical building blocks
- Model: interprets a request and produces candidate next steps.
- Tools: controlled interfaces to search, documentation, tickets or data.
- State: stores what has happened in the workflow.
- Policy: restricts what the system is allowed to read or change.
- Human approval: gates sensitive operations such as production updates.
A safe data pipeline assistant
Start with read-only Cloud Composer job logs and a list of known pipeline errors. Let the agent summarize possible failure causes and cite exact log evidence. Do not permit automatic restarts, deletes, permission changes or production SQL without review.
Where agents are unnecessary
Do not add agent orchestration to tasks where deterministic code is clearer, faster or safer. A scheduled SQL transformation, for example, often belongs in a tested data pipeline rather than a free-form LLM workflow.
Portfolio demonstration idea
Build a synthetic-data troubleshooting copilot that produces a root-cause hypothesis, confidence caveats and an approval-required action plan. Publish a demo and evaluation results before describing it as a completed project.