AWS Student Builder Group · UNISA · 13 May 2026

Agentic AI in Action.

From chatbots that talk to agents that do. The next chapter of AI in 90 minutes.
Akhil Boddu · Co-Founder, Zaio Institute · Creator, AfricanTechBro
01 / 20
africantechbro.ai·zaio.io
Quick on me, then we go

I build, teach, and run agents every day.

Co-Founder
Zaio Institute
QCTO-accredited tech bootcamps · hundreds of South Africans placed into tech roles.
Creator
AfricanTechBro
100K+ followers on TikTok & Instagram · single video at 32M+ views.
Daily driver
Claude · Hermes · AWS
Composing agents to ship at the velocity of a 10× larger team.
Tonight you leave with
A clear map + a live demo + 5 Claude habits
By the end you'll know what an agent IS — and have a way to use one tomorrow morning.
02 / 20
africantechbro.ai
Foundation 1 · the words everyone mixes up

AI · ML · Generative AI — three nested circles.

AI Artificial Intelligence ML Machine Learning GEN AI Generative AI
AIthe whole field
Any computer system doing something that looks intelligent. Analogy: "transportation" — covers bikes, cars, planes, and rockets.
MLa kind of AI
Computers that learn from examples, not from hard-coded rules. Analogy: a toddler learning "dog" after seeing 100 dogs — not from a book of dog-shape rules.
Gen AIa kind of ML
ML that creates new stuff — text, images, code, music — instead of just sorting or predicting. Analogy: a librarian who became a novelist.
03 / 20
africantechbro.ai
Foundation 2 · where intelligence comes from

How AI learns from data.

Nobody programs the answers. The model figures out the patterns from millions of examples — like training a puppy with treats, just at internet scale.
STEP 01
📚
Show examples
Feed the model massive amounts of data — books, sites, code, images.
STEP 02
🎯
Ask it to predict
Hide the next word / next pixel. Make it guess.
STEP 03
Score the guess
Reward correct guesses, penalise wrong ones — billions of times.
STEP 04
🧠
Patterns emerge
After enough rounds, the model "knows" grammar, facts, code — without ever being taught a rule.
The puppy analogyYou don't write a rulebook for a service dog. You show 10,000 examples, reward the right behaviour, correct the wrong ones — and over time the dog figures out what "sit," "fetch," and "leave it" mean. AI training is exactly that, at the scale of the entire internet.
04 / 20
africantechbro.ai
Foundation 3 · the magic, demystified

A chatbot is autocomplete on steroids.

It doesn't "know" things. It predicts the most likely next word, one at a time, based on patterns it absorbed during training.
You typed
"The capital of South Africa is …"
Model's predictions for the next word
Pretoria82%
Cape Town9%
Johannesburg6%
3%
The whole gameThe chatbot picks a word, adds it to the sentence, then asks itself "what comes next?" — and does it again. And again. Hundreds of times per response. It's predictive text on your phone — but the phone has read the whole internet.
05 / 20
africantechbro.ai
Foundation 4 · you're already using it

A day in your life, silently powered by AI.

06:30
📱
Face ID unlock
Computer vision identifies your face in 0.1s.
07:00
🗺️
Google Maps ETA
Predicts traffic across millions of historical trips.
09:00
📧
Gmail autocomplete
Suggests your next sentence based on how you write.
10:30
🛡️
Spam filter
Quietly blocks 99% of phishing you never see.
12:30
🍔
Uber Eats picks
Personalised by what people like you ordered.
16:00
🎵
Spotify Made for You
A playlist generated for your taste alone.
20:00
🎬
Netflix thumbnails
Different cover image for every viewer, A/B tested.
23:00
🏦
Bank fraud check
Flags weird transactions in milliseconds while you sleep.
You don't start using AI today. You're already in it. The question is — can you direct it?
06 / 20
africantechbro.ai
Foundation 5 · what changed in 2023

AI used to sort. Now it creates.

The leap from "find me this book" to "write me a new one in the same voice."
Before · Classical ML
📚
The Librarian
Sorts, classifies, recommends. Can tell you which book matches your query — but cannot write a new one.
spam filter netflix recs credit scoring face ID
Now · Generative AI
✍️
The Novelist
Creates new text, images, code, audio, video — in any style, on demand. Trained on the work of every novelist who came before.
ChatGPT · Claude Midjourney · DALL-E GitHub Copilot Suno · ElevenLabs
07 / 20
africantechbro.ai
Foundation 6 · the honest part

The three risks everyone should know.

No fearmongering. Just the real ones — with the real fixes.
RISK 01
🤥 Hallucination
AI confidently makes things up — fake citations, wrong facts, invented quotes — and sounds correct doing it.
MitigationVerify any specific claim. Treat AI like a confident friend, not a textbook.
RISK 02
⚖️ Bias
AI absorbs the prejudice in its training data. If the internet is biased — and it is — the model can amplify it.
MitigationQuestion outputs about people, places, races, jobs. Diversify your sources.
RISK 03
💼 Displacement
AI doesn't replace people — it replaces tasks. The tasks first to go are the ones easiest to describe.
MitigationBecome the person who directs AI. That's the job that compounds in value.
08 / 20
africantechbro.ai

Agentic AI in Action.

From AI that talks to AI that does. This is what the next 30 minutes is about.
09 / 20
africantechbro.ai
The single biggest mental shift

A chatbot is an intern. An agent is an employee.

Same brain. Different permissions, different responsibility, totally different output.
Chatbot
👨‍🎓
The intern in the room
Answers your questions thoughtfully. Will not leave the room. Cannot touch your files, send emails, or take actions on your behalf.
Can do
  • Explain things
  • Brainstorm
  • Draft text you copy-paste
Agent
👩‍💼
The junior employee
You give them a goal. They open the laptop, send the emails, run the reports, file the receipts, follow up — and report back when done.
Can do
  • Browse the web, read files
  • Send emails, save to S3, post to Slack
  • Run code, call APIs, loop until the goal is met
10 / 20
africantechbro.ai
The moving parts

Every agent has the same four pieces.

🧠
Model
The brain
The LLM that thinks. Examples: Claude, GPT, Hermes, Llama. Swappable.
🦾
Harness
The body
The program that runs the loop and gives the brain hands. Examples: Claude Code, Hermes, AWS AgentCore.
🛠️
Tools
The hands
What the agent can actually do. Read a file. Run AWS CLI. Send an email. Search the web.
📒
Memory
The notebook
What it remembers across turns and across days. Today's run knows what yesterday did.
Wrap these four with a loop — think · act · observe · think again — and that's an agent. Every agent. Same recipe.
11 / 20
africantechbro.ai
Connecting to AWS · the toolbox

AWS gives your agent production hands.

Think of these as different tools your agent can pick up. Most of them are one CLI call away.
💬
Lex
Build a chatbot in minutes — think Alexa for your business or website.
→ the chatbot template
📖
Comprehend
Reads text. Pulls out entities, sentiment, topics — without writing any ML.
→ the reading-comprehension robot
👁️
Rekognition
Looks at images / video. Detects objects, faces, text, unsafe content.
→ the eyes
🔊
Polly · Transcribe
Polly = text → speech. Transcribe = speech → text. The mouth and the ears.
→ the voice + the hearing
🌐
Translate
100+ languages, neural translation, real-time. Powers our SA-multilingual demos.
→ the translator
🎓
SageMaker
Train your own custom model when off-the-shelf isn't enough.
→ the workshop
Lambda + EventBridge
Run your agent on a schedule (6am every day) without managing a server.
→ the schedule + deploy
12 / 20
All services available in the AWS Free Tier with limits — perfect for student projects
Live now · ~15 min

Let's run the #1 open-source agent on OpenRouter.

NousResearch / hermes-agent
#1 OpenRouter 224B tokens/day MIT
Open-source agent harness — TUI, persistent memory, scheduling, 40+ built-in tools, MCP support. Free to run. Swap any model.
open-source hermes openrouter aws-mcp web-search
We did not write this. We install Hermes, plug in the official AWS MCP server, then ask it in plain English to find jobs and save the results to S3. That's the "connect to AWS" moment.
13 / 20
github.com/NousResearch/hermes-agent

Let's run an agent.

Switch to terminal. Press → when the demo is done to come back.
14 / 20
watching: Hermes + AWS MCP
Recap · in plain language

What you just watched, step by step.

① Installed Hermes
One curl command. The #1 OSS agent on OpenRouter. We wrote zero lines of agent code.
② Picked a model
OpenRouter → 200+ models, one API key. Swappable mid-conversation with /model.
③ Plugged in AWS MCP
3 lines of YAML in ~/.hermes/config.yaml/reload-mcp. USB-C for agents, made real.
④ Talked in English
"Find 3 jobs · save list to s3://…" Hermes did web_search → aws_api MCP → S3 write. No code from us.
⑤ Ready to schedule
Hermes has native cron — hermes cron. The agent runs at 6am every day. You sleep.
You don't build agents in 2026. You compose them. And the AWS MCP server is the single most powerful thing you can plug into one.
15 / 20
africantechbro.ai
The take-home · whether you're a builder or not

Five ways to use Claude in daily life.

Tonight, this week, every week. No coding required.
USE 01
🎓
Study Buddy
Explain anything as if I'm 12, then again at university level. Quiz me.
"Explain Bayes' theorem twice — once like I'm 12, once for a stats exam."
USE 02
✍️
Writing Partner
Emails · cover letters · LinkedIn posts · drafts in seconds, in YOUR voice.
"Rewrite this email warmer but still professional. Match my style — here are 3 past emails."
USE 03
🧭
Decision Sounder
Stuck between options? Paste the context, ask for the strongest case for each.
"I'm choosing between a SWE role at X and a data role at Y. Make the case for each in 200 words."
USE 04
💻
Code Companion
Even non-devs: build small tools, fix bugs, explain unfamiliar code, scaffold projects.
"Write a Python script that renames every PDF in this folder using the date inside it."
USE 05
📚
Learning Accelerator
Summarise papers · turn YouTube transcripts into notes · build a study plan.
"Summarise this 40-page paper into 10 bullets + 5 quiz questions I can review tomorrow."
16 / 20
africantechbro.ai · screenshot this slide
Before you sleep tonight

Three actions. Thirty minutes.

By tomorrow morning you have used an agent. Not theory — your own.
Step 01 · 5 min
Open Claude. Pick ONE of the five uses from the last slide. Paste a real thing from your day. Watch what happens.claude.ai · free tier works
Step 02 · 10 min
Skim the Hermes README. github.com/NousResearch/hermes-agent — see what 40+ built-in tools look like before you install.comment RUNAGENT for the full step-by-step guide
Step 03 · 15 min
Follow the RUNAGENT guide. Install Claude Code, plug in the AWS MCP server, upload a tailored CV to your own S3. 45 min start to finish.africantechbro.ai/guides/run-your-first-ai-agent
17 / 20
africantechbro.ai · comment RUNAGENT for the full step-by-step
What this means for you, in South Africa, in 2026

In a tight market, the composer outships the engineer.

01
Tight budgets are your unfair advantage
Open-source tools + AWS free tier = $0 to start. A Bay-Area startup burns that on their AI bill before lunch — you're forced to be efficient, and efficient is the meta.
02
"Knows AWS + uses AI" is the 2026 SA dev premium
Every SA recruiter values AWS. Every agent loves AWS CLI. You become more leverageable as a hire AND as a builder.
03
Build agents for yourself first
Job-hunt agent. Study agent. Side-hustle agent. Every one is portfolio + résumé + interview story in a single artifact.
You're not behind. You're early.
18 / 20
africantechbro.ai
Two ways in

The deck. The step-by-step. Pick one — or both.

Comment BYOS for the full talk + write-up. Comment RUNAGENT for the 45-min spoonfeeding guide that walks you through every install command, every AWS step, every fix. I'll DM you the link.
🌍
Web
africantechbro.ai
🎓
Structured path
zaio.io
📸 Take a photo
19 / 20
You're not behind.
You're early.
Q&A · drop them in the chat · 10 min · let's go
20 / 20
Thank you · africantechbro.ai · zaio.io

Keyboard controls

Next slide→ / Space
Previous slide
First slideHome
Last slideEnd
Toggle speaker notesS
Toggle fullscreenF
Show this help?
Print all slides (PDF)⌘P