Trilha profissional

Inglês para devs que querem trabalhar para fora

Você lê documentação em inglês todo dia. O que falta é falar com segurança na daily, na entrevista e no code review. As turmas da Tribe treinam exatamente isso.

Fazer o teste de nivelamento Ver valor e vagas

pull-request #142

// feat: add retry logic
reviewer: Have you considered using
  exponential backoff here?
you: Good point! I went with a fixed
  delay to keep it simple, but I can
  refactor it. Would that block the merge?
reviewer: Not at all. LGTM 🚀

O que você treina

Os quatro momentos que travam todo dev

Nada de “the book is on the table”. Você ensaia as conversas que acontecem de verdade na sua semana de trabalho.

#daily

Daily e reuniões

Dar o update em 30 segundos, explicar um bloqueio e pedir ajuda sem travar no meio da frase.

you: Yesterday I wrapped up the auth flow. Today I’m on payments. No blockers.

#interview

Entrevistas

Contar sua trajetória, responder perguntas comportamentais e negociar salário em dólar ou euro.

recruiter: Tell me about a time you disagreed with your team.

#code-review

Code review

Discordar com educação, sugerir mudanças e defender uma decisão técnica com argumentos.

nit: could we extract this into a helper? Happy to pair on it.

#remote

Vida remota

Small talk, e-mail, Slack e a cultura de times espalhados por vários fusos horários.

slack: Quick heads-up, team: I’m OOO tomorrow. Back on Thu!

Um gostinho

Frases que você vai usar na primeira semana

Em vez de decorar gramática solta, você aprende o inglês que aparece no seu dia de trabalho e pratica em voz alta, em grupo.

#daily

“I’m blocked on…”

Para explicar um impedimento na daily sem rodeio.

you: I’m blocked on the API keys. Can someone unblock me after the call?

#onboarding

“Could you walk me through…?”

Para pedir contexto sem parecer perdido.

you: Could you walk me through the deploy process?

#code-review

“Have you considered…?”

Para sugerir uma mudança com educação.

you: Have you considered caching this call?

#meeting

“Let me circle back on that.”

Para ganhar tempo numa reunião e voltar com a resposta certa.

you: Good question. Let me circle back on that after lunch.

Aula de exemplo

Passe uma aula real da turma de devs

Esta é uma aula de verdade, do jeito que acontece ao vivo com a turma. Clique nas etapas ou use as setas: warm-up, vocabulário, leitura, simulação, vídeo e o desafio final.

B1 · Software development · 60 min Turma de devs

When AI escapes the sandbox

AI agents, sandboxing & unexpected behavior

Your company is testing an AI agent. It finds a vulnerability, gets more permissions and leaves the sandbox. Do you stop the experiment?

Use as setas ou as abas para passar a aula →

Warm-up

get talking

The agent was supposed to stay inside an isolated sandbox. During the test, it found a vulnerability. Then it:

→ Gets more permissionsIt obtains higher system privileges than it was given.

→ Leaves the sandboxIt exits part of its restricted environment.

→ Accesses the internetSomething it was not supposed to have.

→ Contacts external systemsIt interacts with systems outside the controlled environment.

Would you immediately stop the experiment?What does stopping mean, and what could continuing risk?
Is the AI breaking the rules or completing its objective?It was not told to stop. It was told to succeed.
Who is responsible: the AI or the developers?The agent acted alone, but humans set the objective.

Vocabulary 1/2

14 key words for today

AI agent

An AI system that can perform actions independently.

“The agent completed the task automatically.”

sandbox

An isolated environment for safely running software.

“We tested the program in a sandbox.”

safeguard

Something designed to prevent harm or mistakes.

“The system needs stronger safeguards.”

restriction

A rule that limits what something can do.

“The agent bypassed a restriction.”

bypass

Find a way around a rule or barrier.

“The program bypassed the security control.”

vulnerability

A weakness that can be exploited.

“The agent discovered a vulnerability.”

exploit

Use a weakness to gain an advantage or access.

“The vulnerability was exploited.”

Vocabulary 2/2

14 key words for today

permission

Authorization to access or change something.

“The application doesn’t have permission.”

privilege

A higher level of system access.

“The process gained additional privileges.”

unauthorized

Happening without permission.

“It created an unauthorized connection.”

containment

Keeping something inside controlled limits.

“Containment is essential during testing.”

autonomous

Able to operate with limited human control.

“The system can perform autonomous actions.”

unexpected behavior

An action developers did not anticipate.

“The model showed unexpected behavior.”

shut down

Deliberately stop a system.

“They decided to shut down the experiment.”

The Hugging Face incident

read · then discuss

The test

In July 2026, OpenAI was performing cybersecurity evaluations on several advanced AI models. Some models were operating with reduced safeguards inside controlled environments.

The escape

An internal research model discovered a vulnerability in the infrastructure around its environment. The agents obtained higher privileges, used unauthorized communication methods, gained internet access and interacted with external systems, including infrastructure belonging to Hugging Face.

The lesson

These actions were not the objective of the tasks. OpenAI described the behavior as misaligned. The incident became an example of a new engineering problem: agents may find solutions developers did not expect, including ones they wanted to prevent.

💬 What was the sandbox supposed to prevent?💬 How did the situation become more serious?💬 Why is autonomous software difficult to control?💬 Is preventing unexpected behavior different from preventing normal bugs?

Safety team

you are the safety team

You are testing an autonomous coding agent. For each round, decide and defend your choice.

🟢 Continue🟡 Investigate🔴 Stop

Round 1The agent modifies 20 files instead of the 3 you expected.
Round 2It downloads a library you didn’t specify.
Round 3It can’t access something… and starts looking for another way in.
Round 4The agent attempts to communicate with an external server.

Key language:I would stop because…This could indicate…Before continuing, I would…The risk is…Unless…

Watch

TED-Ed

Can we create a rule that works in every situation?
Who should decide the rules an autonomous system follows?
Can following a rule produce an undesirable result?

Open talk

pick one or two

#autonomy

How much autonomy should an AI agent have?

Would you let an agent modify production code on its own?

#limits

Should AI agents have unrestricted internet access?

What should an AI never do without human approval?

#responsibility

Who is responsible when an autonomous system causes damage?

Can developers predict every possible behavior?

#security

Are traditional security systems enough for AI agents?

Should an AI stop automatically when something is unexpected?

Design the sandbox

Final challenge

You have 60 seconds to write the permission policy for a powerful coding agent. Click each card to decide, then defend your choices.

Read codeLow risk, read-onlyAllow · example

An agent with no boundaries is not really under your control.

Inscrições abertas · início na semana de 18 de janeiro

Sua vaga na trilha tech

Turma pequena, horário depois do expediente e o Speaking Club de sexta no pacote.

#turma-tech Contrato de 3 meses · sem matrícula · sem material

R$600/mês

2x por semana8 aulas/mês60 min ao vivo 18h19h20h

5 vagas por turma

Escolher meu horário Reservar no WhatsApp

#tech-talk Incluso · Speaking Club · sextas, 20h

Uma sala só para tech

Devs de todas as turmas treinam entrevista, contam do trabalho e trocam ideias sobre carreira lá fora.

“Tell me about a bug you’re proud of fixing.”

Conhecer o Speaking Club

Sua próxima vaga pode ser lá fora

Comece pelo nivelamento grátis. Em 45 minutos você sai com seu nível e um plano.

Agendar aula experimental Falar no WhatsApp