Lesson 5. Realistic Expectations and Risks

Lesson 5. Realistic Expectations and Risks#

Why This Matters#

AI agents aren't magic. They make mistakes, need tuning, and require oversight. Being honest with clients and understanding limitations helps build long-term relationships.

Key Idea#

An AI agent is a tool, not a replacement for humans

Agents excel at routine tasks, but complex situations need humans. Success = right task selection + configuration + oversight.

What Agents Do Well#

✅ Good at:

• typical questions with clear answers

• collecting and structuring data

• following rules to perform actions

• fast 24/7 response

• handling high volume of requests

Examples:

• answer an FAQ question

• book a client for a service

• qualify a lead by criteria

• send a reminder

• fill a spreadsheet from a form

What Agents Do Poorly#

❌ Poor at:

• non-standard situations

• conflicts and emotional requests

• tasks requiring expertise

• decisions outside the rules

• working with incomplete data

Examples:

• customer complains and demands compensation

• need to evaluate a unique project

• task requires creativity

• data is contradictory

Typical Risks and How to Reduce Them#

1. Agent Makes Mistakes

Risk: the model may "hallucinate" an answer or misunderstand the request

How to reduce:

• give the agent a knowledge base instead of relying on "general knowledge"

• limit responses: "if you don't know — say you don't know"

• test on typical requests

• add a "contact a human" button

2. Data Leakage

Risk: the agent may "remember" sensitive data or show it to another user

How to reduce:

• don't pass personal data to the agent unless necessary

• use anonymization

• configure security rules

• choose platforms with data protection

3. Platform Dependency

Risk: the platform may get more expensive, change, or shut down

How to reduce:

• choose stable platforms (Zapier, n8n)

• maintain documentation and backups

• don't lock into one platform completely

4. Agent "Doesn't Understand" the User

Risk: user writes in an unusual way, agent gets confused

How to reduce:

• add example phrasings to the prompt

• test on real requests

• train users ("write clearly")

• have a "didn't understand → human" option

How to Talk to Clients About Risks#

Bad:

"Everything will be great, the agent will do it all"

Good:

"The agent will handle 70–80% of typical requests. The other 20–30% are non-standard situations — better to hand those to you. We'll configure the agent to know when a human is needed."

Ethics and Rules (2026)#

In 2026 it's important to:

• Transparency: users should know they're talking to an agent

• Control: customers should be able to contact a human

• Data: comply with personal data rules (GDPR, local regulations)

• Honesty: don't promise what the agent can't do

"Does the Task Fit?" Checklist#

A task fits an agent if:

• done regularly (10+ times per week)

• has clear rules or a knowledge base

• doesn't require deep expertise

• impact is measurable (time/money/quality)

• there's a plan B (what to do if the agent errs)

A task does NOT fit if:

• too complex and unique every time

• data is confidential and critical

• no clear rules

• impact is unclear

Check Your Understanding#

1. What do agents do well, and what poorly?

2. What are typical risks when working with agents?

3. How do you reduce the risk of agent errors?

4. Why is it important to be honest with clients about limitations?

5. What ethical rules matter in 2026?

---

Module 1 Practice#

Assignment 1: Opportunity Map#

Choose 3 niches (e.g., beauty salon, online store, agency) and list 2–3 tasks for each where an agent would add value.

Format:

Assignment 2: Mini Impact Estimate#

Take one task from the table above and estimate impact:

• how many times per day/week it's done

• how long it takes

• cost of that time

• automation percentage

• total savings

Assignment 3: Describe Yourself#

Write a short description of yourself as a provider (3–5 sentences):

• who you are

• how you help

• what problems you solve

• why clients should work with you

Example:

"I help small businesses automate routine work with AI agents. I build bots for support, client booking, and lead qualification. No code, fast, with guaranteed results. In 3 months I launched 12 projects; average client savings — ~$200/month."

---

Review Questions#

1. How does an agent differ from ordinary automation?

2. Which tasks are best suited for an agent?

3. What impact metrics do you know?

4. Who makes the decision to implement in a small business?

5. Why there's no "magic" and where are the limits?

6. How do you explain agent value without tech jargon?

7. What is a "process" in simple terms?

8. Why should the first projects be small?

Answers:

1. An agent performs actions (records, sends, analyzes), not just follows a script

2. Repeatable, typical, with clear rules, delivering measurable impact

3. Time, money, quality

4. Business owner or department head

5. An agent is a tool; it needs configuration and makes mistakes in non-standard situations

6. "An agent is a helper that works 24/7 and handles routine, saving your time and money"

7. A sequence of steps to achieve a result

8. Faster results, clearer impact, fewer risks

---

Module Summary#

You've learned:

• what AI agents are and how they differ from bots

• where they're used and what problems they solve

• how to calculate and show impact

• who's involved in projects and how to communicate with them

• what risks exist and how to reduce them

Next step: Module 2 — learn how AI works under the hood (no math!) and how to write effective prompts.

---

Check Your Knowledge#

Quiz: AI Agent Basics#

Flashcards#

---

Materials for the Site#