If failure is cheap (draft, image) → generative, human in loop. If failure has side effects (refund, scheduling) → agentic with guards/logs/rollbacks. If environment changes (physical world, inbox) → agent with memory/evals/kill switch.
Choose: Generative vs Agentic vs AI Agents
When to choose which
- Generative: You need content at scale, output is the end goal, human will review, no tools required.
- Agentic: Multi-step workflow with branches, needs to call APIs/search, success = task completion.
- AI Agents: High-level goal in open world, needs sense/plan/act/learn, requires autonomy + memory.
Example Scenarios
GENERATIVE AI
1. Morning email & Slack
You say: Turn these rough notes into a polished reply to my manager and summarize the long thread below.
It does:
- Polishes tone while keeping your key points
- Extracts action items from the long thread
- Creates a concise draft ready for you to review and send
Why this type fits: Output is content, single-pass, and you review before sending.
2. Content creation for your socials
You say: Write 3 Instagram captions and an Etsy description for these handmade ceramic mugs from my photos.
It does:
- Generates caption options in different tones
- Turns photo details into a product description
- Suggests hashtags and a LinkedIn cross-post
Why this type fits: You need fresh content quickly without calling tools or taking actions.
3. Everyday problem solving
You say: Here is what is in my fridge – give me a meal plan for 2 days, plus tailor my cover letter to this job post and explain this code error.
It does:
- Builds a meal plan from only what you have
- Rewrites your cover letter to match the posting
- Explains the code error in plain language with a fix
Why this type fits: One prompt in, one helpful answer out – no multi-step workflow needed.
AGENTIC AI
1. Inbox & calendar manager
You say: Sort my Gmail, save receipts to Drive, and turn emails with dates into calendar events.
It does:
- Reads Gmail, labels receipts, and files them to your finance folder
- Creates calendar events from emails with dates and times
- Drafts replies within your rules and waits for your approval
Why this type fits: It needs to call Gmail, Drive, and Calendar APIs and follow branching rules.
2. Shopping & errands helper
You say: Here is my grocery and household list and a budget – find the best options.
It does:
- Compares options on Amazon and Instacart within your budget
- Checks reviews and delivery windows for each item
- Prepares a cart and checks out only after you approve
Why this type fits: Multi-step workflow that needs search, compare, and an action with a guard.
3. Travel planner
You say: Plan a weekend in Napa under 300 dollars and avoid conflicts with my calendar.
It does:
- Searches flights and hotels within your budget
- Checks your calendar for conflicts and travel time
- Holds options and books only if you approve
Why this type fits: Goal-driven workflow that orchestrates search, calendar, and booking tools.
AI AGENTS
1. Personal finance agent
You say: Keep my bills paid and help me save without thinking about it.
It does:
- Senses bank balance and due dates to pay bills on time
- Moves spare cash to savings when it is safe to do so
- Learns your spending patterns and nudges you before you overspend
Why this type fits: Continuous sense, plan, act, and learn loop with memory of your habits.
2. Smart home agent
You say: Make home comfortable and efficient while I am out, and keep the fridge stocked.
It does:
- Observes occupancy, weather, and energy prices to adjust thermostat and lights
- Watches fridge camera feed and adds low items to grocery list
- Orders groceries when you confirm and learns your preferences over time
Why this type fits: Lives in your environment, senses changes, and acts autonomously with learning.
3. Health & fitness agent
You say: Help me stay consistent with workouts and meals around my work schedule.
It does:
- Connects watch, food log, and calendar to plan workouts around meetings
- Adjusts meal suggestions based on activity and preferences
- Observes progress and changes the plan when energy or sleep is low
Why this type fits: Needs memory, sensing, and long-term adaptation to your changing life.
The stack: how they build
Generative is the engine. Agentic is the workflow around the engine. Agents are the autonomous system that owns the outcome.
- Layer 01: Foundation Engine (Generative) – Prompt → LLM → Output.
- Layer 02: Workflow & Orchestration (Agentic) – Goal → Model → Tools → Logic → Action.
- Layer 03: Autonomous System (Agents) – Goal, Memory, Sense, Plan, Act, Learn.
Final Takeaway: The future is hybrid, not either/or
Best products start with a job:
- Generate a contract → generative.
- Generate and send it after checking Stripe + DocuSign → agentic.
- Own renewals for top 100 accounts and learn what works → agent.
When to choose which Example Scenarios GENERATIVE AI 1. Morning email & Slack You say: Turn these rough notes into a polished reply to my manager and summarize the long thread below. It does: Why this type fits: Output is content, single-pass, and you review before sending. 2. Content creation for your socials You say:…