Agentic AI?
Agentic AI is software that coordinates specialized reasoning, information, and actions to achieve an objective—even when the path changes along the way.
What is agentic AI?
At its core, it is still software. A main application—the orchestrator—coordinates smaller, specialized components called agents.
Handles one specialized part of the problem.
Handles another specialized part.
Returns findings to the orchestrator.
The business problem
The couple wants an outdoor wedding. But rain affects more than location: food, music, photographs, decorations, guests, timing, and cost may all change.
Changing condition
The weather forecast changes as the special day approaches.
Desired experience
The ceremony, meal, music, photographs, and celebration should still happen.
Need for coordination
The application must compare alternatives and recommend the best path.
The optimized state
The agent team
Determines whether the outdoor plan remains viable.
Identifies indoor, tented, delayed, or alternate-location options.
Preserves meal service and dietary requirements.
Protects music, décor, photography, dancing, and guest experience.
Tests cost, contract, and approval constraints.
Prepares and sends updates only after a plan is approved.
Retrievals and tools
Agents
Interpret information, compare options, and return recommendations.
Retrievals
Wedding plan, couple preferences, contracts, guest list, venue rules, and budget records.
Tools
Weather service, venue calendar, calculator, email, SMS, maps, and booking systems.
Who coordinates whom?
The orchestrator owns the overall objective. It assigns focused work to specialist agents, receives their results, and decides what should happen next.
receives results
What does an agent use?
An agent is a specialized application component. To complete its assignment, it may retrieve stored information, call an LLM for reasoning, or use a tool for live data or action.
Specialist Agent
Receives a task from the orchestrator, gathers what it needs, performs its focused work, and returns a result.
How the relationships form a loop
The orchestrator does not simply call every agent once. It evaluates each returned result and decides whether the current wedding plan is good enough.
ACCEPTABLE → continue toward execution
NOT ACCEPTABLE → revise the plan, call another agent, retrieve more information, or loop again
Meet Maya and Jordan
The orchestrator gathers evidence
Thunderstorms are likely from 4:30–6:30 PM, with lightning possible.
The indoor ballroom is included, but it must be selected four hours before the ceremony.
The live band and formal first dance are identified as must-have experiences.
The ballroom fits all guests and dinner tables—but not the eight-piece band with a dance floor.
Candidate Plan 1 fails
Move the entire wedding into the ballroom
This avoids the storm and preserves the meal, schedule, and guest comfort. However, the band cannot fit while retaining a dance floor.
The orchestrator loops: retrieve more information, call additional agents, and test another arrangement.
The system searches for a better path
A separate covered pavilion can hold the band and dance floor after 7 PM.
Storms should pass by 6:45 PM, with a low rain probability afterward.
Dinner can begin indoors while the pavilion is dried and prepared.
Extra labor, flooring, and guest guidance cost $2,850—within the contingency budget.
Candidate Plan 2 meets the objective
Indoor ceremony and dinner; covered-pavilion band and dancing
The ceremony begins on time in the ballroom. Dinner follows indoors. After the storm passes, guests move to the covered pavilion for the live band, first dance, cake, and sparkler send-off.
The approved recommendation becomes action
Wed Not Wet recommendation
- Move the 5 PM ceremony and dinner into the ballroom.
- Prepare the covered pavilion for the band and dancing beginning at 7 PM.
- Use the $2,850 contingency allocation for extra labor and temporary flooring.
- Send revised instructions to guests, vendors, the wedding party, and transportation staff.
- Continue weather monitoring; keep the full indoor plan available as a fallback.
Agentic AI is coordinated software working toward an objective
Objective
The application begins with a desired outcome. In Wed Not Wet, the goal is not simply to move indoors—it is to preserve the wedding experience.
Orchestration
The orchestrator selects agents, evaluates results, determines what is missing, and decides whether to stop or continue.
Optimization
The system can loop through multiple candidate plans until it finds one that satisfies the objective within acceptable constraints.
Specialized Agents
Different agents contribute focused reasoning about weather, venue, catering, experience, budget, and communications.
Retrievals and Tools
Agents retrieve stored context and may use live tools, but tool availability does not mean every tool must be used.
The LLM
The LLM is not the whole application. It is a reasoning and language capability used by the orchestrator or agents where helpful.
What this example intentionally simplifies
Educational Example
Wed Not Wet is designed to introduce agentic AI concepts. Real-world systems are typically more complex and domain-specific.
Not an Implementation
This lesson focuses on the architecture and flow of an agentic application rather than programming languages, frameworks, prompts, or code.
Production Systems
Operational systems often include approvals, security controls, audit logging, permissions, monitoring, deterministic validation, and governance requirements beyond this introductory example.
Wed Not Wet
A simple wedding case reveals the core structure of an agentic AI application: a fixed objective, specialized agents, retrieved context, optional tools, LLM-supported reasoning, contingent paths, and a loop that continues until a satisfactory outcome is found.
The path changes as the system learns.
The experience was preserved.
Check Your Understanding
Answer all five questions, then select Submit Assessment.
Watch Wed Not Wet work the problem
Current optimization run
Checking storm timing and lightning risk.
Reviewing ballroom and pavilion capacity.
Testing must-have features against the candidate plan.
Checking contingency limits and added costs.
Evaluating whether the objective is satisfied.
The system does not follow one fixed path
The orchestrator calls different agents as new information appears. It may reject a candidate, loop, retrieve more context, and test a better plan before stopping.