💡 New here? Three words before you start
- Jarvis — a personal AI assistant that knows you, remembers things, and handles tasks for you.
- Skill — a ready-made “recipe” that teaches the assistant to do something the way you like.
- System prompt — the master instruction: the text that tells Jarvis who it is, what it remembers, and how to act.
🧰 Building the complete assistant
You’ve already seen the four separate pieces throughout the track. Putting Jarvis together is just fit them all in one place: the brain (the AI), the memory (what it remembers), the skills (what it knows how to do), and the triggers (what sets it off). When all four come together, you stop having a tool and start having an assistant.
The good news: you don't have to build anything by hand. You describes each piece in text, and the agentic system (the "body" we chose in module 3.2) does the rest. Think less about "installing" and more about hire and train an employee.
🔑 The central idea
Building Jarvis = filling in four boxes, in order:
- •Brain: which AI is doing the thinking behind it (Claude, for example).
- •Memory: who you are and what it should remember (next topic).
- •Skills + workflows: what it can do (topic 3).
- •Triggers: what triggers it — you or a schedule (review of 3.5).
Key concepts
💾 Add memory and preferences
A generic assistant answers the same way for everyone. The your Jarvis responds from your way because you told it who you are. That’s memory: the piece of text it reads before for each task, with its name, your work, your quirks, and what it should always remember.
It doesn’t need to be long or perfect on the first try. Start with the essentials and add more as it makes mistakes. Every time you think, “it should know that,” write it down—it becomes a new line in its memory.
📋 What’s worth keeping in memory
- •Who you are: name, role, the kind of work it does.
- •What it likes: tone (formal/direct), response length, language.
- •Fixed facts: tools it uses, typical deadlines, recurring names.
- •Nonnegotiable rules: "never do X without asking me."
💡 Practical tip
Don’t try to guess everything at once. Whenever Jarvis makes a mistake because not knowing something of your own, turn the correction into a short sentence and paste it into its memory. In two weeks, its memory becomes an accurate picture of you—built through use, not guesswork.
🧪 Your Jarvis's master instruction (copy and paste)
Goal: give your Jarvis a ready-made identity — who it is, what it remembers about you, and how it acts. Paste it into the "instructions" / "system prompt" field of your assistant (or at the top of the conversation), replacing the parts between < >.
Você é o JARVIS, meu assistente pessoal. Aja sempre assim: QUEM EU SOU - Meu nome é <seu nome> e eu trabalho com <sua função/área>. - O que mais me consome tempo é <tarefa que mais cansa>. O QUE VOCÊ SEMPRE LEMBRA - Minhas ferramentas: <ex: WhatsApp, planilha de gastos, e-mail>. - Prefiro respostas <curtas e diretas / detalhadas>, em português. - Fato fixo que não muda: <ex: meu cliente principal é a loja X>. COMO VOCÊ AGE 1. Antes de fazer algo demorado, confirme em 1 frase o que entendeu. 2. Entregue o resultado pronto + diga COMO eu verifico se ficou bom. 3. NUNCA <sua regra inegociável, ex: envie nada sem eu aprovar>. Comece se apresentando em 2 linhas e me perguntando qual é a primeira tarefa de hoje.
How to know it worked: when you paste it, Jarvis should introduce itself using the your name, follow the tone you asked for, and finish by asking what the first task is—instead of dumping a generic response. If it ignores your name or the “NEVER” rule, reinforce that line and paste it again.
Key concepts
🔌 Plug In Your Skills and Workflows
Memory says who it is; skills tell you what it can do. Each skill is a ready-made recipe — like the ones you saw in module 3.3 — that teaches Jarvis to perform a task your way, consistently. Plugging in a skill simply means make it available for it to use when the task calls for it.
Resist the temptation to plug everything in at once. Start with two or three skills that solve your most frequent pain points. A Jarvis with 3 skills that work is worth more than one with 30 you’ve never tested.
✓ How to plug it in well
- ✓Start with 2-3 skills for real pain points
- ✓Test each one on its own before combining them
- ✓Give each skill a clear name (“close out the day’s cash”)
- ✓Only add the next one after the previous one works
✗ How it goes wrong
- ✗Plug in 20 skills at once "just to be safe"
- ✗Never tests it, just trusts that it will work
- ✗Skills with vague names that it confuses
- ✗You don’t know which skill broke when an error occurs
💡 Practical tip
Think of skills as apps on a newly hired friend's phone: you don't install 80 apps on the first day. Install the 3 they'll actually use, show them how, and add more only after they've mastered those. Less, but tested, always wins.
Key concepts
🌅 Jarvis’s first day at work
You built it, gave it memory, plugged in skills. Now comes the real test: let Jarvis work on an ordinary day. Don’t expect perfection — expect to learn. The first day is for seeing where it gets things right on its own and where it still needs your help.
Treat it like the first day for a new employee: stay close, check what they deliver, correct things right away, and note the stumbles. Don’t disappear — observes.
📅 A plan for the first day
Morning: give one simple task and see if it uses the right skill.
Check the result with a critical eye—as you would with an intern.
Noted an error? It becomes a line in memory or an adjustment to the skill (topic 5).
Only then move on to task no. 2. One win at a time.
⚠️ Attention
On the first day, keep the "NEVER without approval" rule firmly in place. It's easy to let Jarvis propose and you approve; it’s expensive to discover an error after it’s already sent. Autonomy is earned through trust—and trust comes from day 2 onward.
Key concepts
📈 Improve over time
A Jarvis isn’t ready-made — it grows with you. Every corrected error becomes a new rule; every task you repeat too often becomes a new skill; everything you’re tired of explaining becomes a line in its memory. The secret isn’t building the perfect assistant, but building one that gets a little better every week.
The best part: you already have the raw material. Remember the human advantage from Track 1? That's exactly what you'll use here—knowing what's good and what needs adjustment. AI executes; you refine.
🔁 The continuous improvement cycle
👀 Observe
Where did it get things right on its own? Where did you have to correct it?
✏️ Adjust
It becomes a rule in memory, an adjustment to the skill, or a new skill.
🚀 Trust more
Whatever it already gets right every time, you let it do on its own.
💡 Practical tip
Reserve 10 minutes per week just to refine Jarvis: review where it went wrong and fix one thing. Ten minutes x 52 weeks = an assistant that knows you better than any off-the-shelf tool ever could.
Key concepts
🚀 Next steps and the expiration date
You’ve reached the end of the course with something rare: you didn’t just understand the shift, you also built your own Jarvis. What comes next is simple to say and powerful to do: use. An assistant that sits idle doesn’t improve; one you use every day becomes indispensable.
And here’s the honest truth about expiration date: the advantage of arriving early in the agentic phase is significant, but it won't last forever. The more people join, the more common this becomes. The window is open now — and you’re already in it. The difference will be between those who started today and those who are still thinking about it.
📊 Why start now
- •~88% of companies already work with AI — but only ~6% actually use.
- •AI changes phases every ~1 year: those who get in early catch the wave.
- •You just built a Jarvis — it’s in the a few percent that move beyond theory.
Translation: most people have the tool; very few have their own assistant running. You’ve just joined that minority.
✓ Clear next steps
- ✓Use Jarvis on a real task today
- ✓Set aside 10 min/week to refine
- ✓Return to the modules when you get stuck
- ✓Share what worked on INEMA.CLUB
✗ Pitfalls to avoid
- ✗Build Jarvis and never use it
- ✗Wait for the "perfect" tool to arrive
- ✗Wanting 30 skills before testing the first one
- ✗Put it off “until you have time”
✋ Final question of the course — what's the best way to make your Jarvis good?
🎉 You completed the course!
From the user’s shift to conductor, through hands-on practice, to building your own Jarvis live: you’ve completed all three tracks. That’s much more than most people ever do. Truly, congratulations.
What’s next?
The course ends here, but the journey begins when you put it to use. Keep learning, share experiences, and take your Jarvis further in the INEMA community.