← All classes
Class VK9AK · Workshop

AI agents at work.

When a task needs an agent — and how to make it reliable.

You already know roughly what an AI agent is. This class is about when you actually need one, and how to make it reliable: working through messy information, repeating a weekly job, and remembering what matters.

◷ 2.5 hours◈ Cantonese · EN terms▤ Bring a laptop◉ 10 seats◇ No coding background assumed
Who teaches Horris Tsebio ↗ Finance · Technology · AI · Product · Communication 20+ years in programming & product development
Reserve standard pricing sold out — waitlist open
Standard HKD 780
Two together − 50 each
Three together − 70 each
HtPi members further 5–15%
Best single discount applies max
HKD 710 from · per person
Thu 19 Nov · 19:00–21:30
Kwun Tong · WeMine Office, MG Tower
6 left
Reserve a seat — pay within 24h Join the waitlist

This session is full. If a seat opens (unpaid holds sometimes lapse), the waitlist is offered it in order — leave your email to join.

FPS · PayMe, no fee. Card (Stripe) or PayPal +3%. Registering creates a free HtPi account; paid members save on every class. Transfer & refund rules → · Gift a seat →

About this class

When do you actually need an agent?

“Agent” now gets used for almost anything with AI in it, which makes it hard to tell when you actually need one.

A chatbot with your files and web access already answers a lot. A fixed workflow can repeat the same steps every week. An agent earns its place when the next step depends on what it has just found.

This class is about telling those apart — and then building an agent you can rely on.

01

One case, start to finish

Most of the evening works through one realistic case, done three ways: as a chat, as a fixed workflow, and with an agent. You see where each one holds up, where the agent goes wrong, and how to fix it.

02

Jobs that come back every week or month

How to turn a regular job into a skill that an agent runs on schedule, on its own.

03

Remembering what matters

How to make sure an agent remembers the important things — and how to keep the context window from filling up on a long task.

04

Where it runs, and who is in charge

Cloud or local, steering a running task from your phone, and when splitting work between several agents helps — or only adds cost. Along the way we compare Grok Bot, Hermes, Claude Cowork and Meta’s Muse.

05

Jev: quick, narrow decisions

Jev, from TypeSafe, answers typed questions — pick a category, give a score, yes or no — instead of writing text. When it fits, where it can save tokens and time, and its trade-offs.

06

Who it’s for

People who know roughly what an agent is and want to use one properly at work. If you came to the vibe coding class, this picks up from there; if you’re new to AI, a short refresher at the start covers what you need.

No coding needed. Live demos, exercises in pairs, and the last 20 minutes kept for your own work and your questions.

Think of one job you do every week or month — you’ll use it at the end.

甚麼時候真的需要 AI Agent?

現在差不多任何用上 AI 的東西都叫「Agent」,反而更難判斷:自己到底需不需要一個。

一個可以讀你的檔案、上網搜尋的聊天機械人,已經能回答很多問題。一個固定的 workflow,可以每星期照同樣的步驟做一次。Agent 適合步驟不能預先定好的工作:它會按剛查到的結果,自己決定下一步做甚麼。

這堂課要做的,是先分清這幾種做法,再做出一個靠得住的 Agent。

01

一個案例,由頭做到尾

課堂大部分時間都圍繞一個貼近真實工作的案例,用三種方法去做:聊天、固定 workflow、Agent。你會看到每種方法在哪裏站得住、Agent 在哪裏出錯,以及怎樣修正。

02

每星期、每個月都要做的工作

怎樣把定期工作變成一個 skill,讓 Agent 按時自己完成。

03

記住真正重要的事

怎樣確保 Agent 記住重要的事——以及在長時間的工作中,怎樣避免 context window 被塞滿。

04

在哪裏運行,由誰主導

雲端還是本機、用手機操控正在運行的任務,以及甚麼時候把工作分給幾個 Agent 有幫助——甚麼時候只會增加成本。過程中會比較 Grok Bot、Hermes、Claude Cowork 和 Meta 的 Muse。

05

Jev:快速而範圍窄的判斷

Jev 是 TypeSafe 推出的模型,它不寫文字,而是回答有固定格式的問題——選一個類別、給一個分數、是或否。甚麼時候適合用它、在哪裏可以節省 token 和時間,以及它的代價。

06

適合誰

大概知道 Agent 是甚麼、想在工作上把它用好的人。如果你上過 vibe coding 課,這堂課正好接上;如果你是 AI 新手,開頭會有簡短重溫,讓你跟得上。

不需要寫程式。有現場示範,也有兩人一組的練習;最後 20 分鐘留給你自己的工作和提問。

請先想好一份你每星期或每個月都要做的工作——課堂最後會用到。

Curriculum

How the class runs.

Times are counted from the start of class.

0:00
Chatbot, workflow or agent
A short refresher on models and tools, then where the lines between the three really fall.
0:17
Where it runs
Grok Bot, Hermes, Claude Cowork and Muse compared; cloud or local; steering a task from your phone.
0:30
The case: failure and repair
One realistic case, done three ways — and fixing what goes wrong.
0:58
Break
Seven minutes.
1:05
Recurring work, memory and context
Handing a regular job to an agent; making sure it remembers what matters; keeping the context window in check.
1:35
Several agents, one owner
When to split work between agents, and how to give each one a clear job.
1:47
Jev: quick decisions
Where Jev fits, what it can save, and its trade-offs.
2:10
Your work, and questions
Pairs sketch one recurring job of their own; open discussion.
Good to know

Please bring a laptop. The live demos run on the instructor’s own accounts, so you don’t need any AI subscription for this class.

Full policies →
Horris is highly experienced in the field. He presents complicated and abstract AI concepts in a simple, accessible way, using plenty of live examples. With his guidance, I've already started exploring different business ideas using the agent I built, and I highly recommend this course.— Michelle W. · Multi-Family Officeattended: OpenClaw Hands-On Workshop · 20 Apr 2026
Reserve a seat