Building Coding Agents with Tool Execution
by DeepLearning.AI (with E2B)
Stop defining tools one by one — let the agent write and run code, safely, in a sandboxed microVM.
Overview
This is a DeepLearning.AI short course built with E2B and taught by Tereza Tizkova and Francesco Zuppichini, both from E2B — 1 hour 21 minutes of video across nine lessons with five runnable code examples, a hands-on project, and a graded quiz for PRO members who want the certificate. The premise is the distinction between an agent restricted to whatever functions you remembered to define and one that writes code, which gives it Python's entire ecosystem instead. Lesson 2 opens up a coding agent's internals and Lesson 3 builds a minimal one with an iterative feedback loop that reads errors and retries. Lesson 4 compares execution environments — running code locally, in a container, and in a sandboxed microVM — and is effectively the security lesson: an agent executing model-generated code on your own machine is a remote-code-execution path you built yourself. Lesson 5 moves execution into E2B's cloud sandboxes, which are Firecracker microVMs giving each run its own kernel rather than a shared one behind a container boundary. Lesson 6 builds a data-analyst agent that explores CSV datasets with pandas, produces summaries and visualisations, and answers questions through a Gradio chat interface. Lesson 7 builds a full-stack agent that edits multiple files and generates a working Next.js web application. E2B's own infrastructure is Apache-2.0 and self-hostable via Terraform.
At a Glance
- Topic
- Agentic
- Level
- Intermediate
- Format
- Course
- Cost
- Freemium
- Duration
- 1 hour 21 minutes of video across 9 lessons, plus 5 code examples, a hands-on project and a graded quiz
- Provider
- DeepLearning.AI (with E2B)
- Hands-on
- Yes — code/exercises
- Certificate
- Available
What You’ll Learn
- ✓Build a coding agent that writes, runs and debugs Python through an iterative feedback loop
- ✓Compare local execution, containers and microVMs on isolation strength and startup cost
- ✓Run untrusted model-generated code in a Firecracker microVM sandbox with its own kernel
- ✓Manage files, install packages and persist state inside a remote sandbox session
- ✓Build a data analyst agent that explores CSVs and generates pandas visualisations on request
- ✓Wrap an agent in a Gradio chat interface so non-engineers can ask it questions
- ✓Deploy a full-stack agent that edits multiple files and generates a Next.js application
Highlights
- •Treats sandbox isolation as a first-class lesson rather than a footnote — rare in agent courses
- •Taught by E2B engineers who build the sandbox infrastructure, not by generalist instructors
- •Two complete working agents by the end, not fragments: a data analyst and a full-stack generator
- •Short enough to finish in an evening at 1 hour 21 minutes, with runnable notebooks throughout
- •The underlying E2B infrastructure is Apache-2.0 and self-hostable, so nothing here locks you in
Who It’s For
Best For
- ✓Python developers who have built tool-calling agents and hit the limits of predefined functions
- ✓Engineers responsible for the security review of an agent that executes generated code
- ✓Builders prototyping data-analysis or code-generation products on top of an LLM
Prerequisites
- •Working Python and familiarity with calling an LLM API
- •Basic understanding of agents and tool calling is helpful but not strictly required
- •An E2B API key to run the sandbox lessons yourself rather than watching them
FAQ
What is Building Coding Agents with Tool Execution?
A short course from DeepLearning.AI built with E2B that moves an agent from calling a fixed set of predefined functions to writing and executing arbitrary code. It covers the security tradeoffs between local, container and microVM execution, then builds two working agents: a data analyst that explores CSVs with pandas, and a full-stack agent that generates a complete Next.js application inside a sandbox.
Is Building Coding Agents with Tool Execution free?
Building Coding Agents with Tool Execution offers free content, with paid options for certificates or premium features.
What level is Building Coding Agents with Tool Execution for?
Building Coding Agents with Tool Execution is aimed at a intermediate audience. Recommended background: Working Python and familiarity with calling an LLM API, Basic understanding of agents and tool calling is helpful but not strictly required, An E2B API key to run the sandbox lessons yourself rather than watching them.
How long does Building Coding Agents with Tool Execution take?
Expect roughly 1 hour 21 minutes of video across 9 lessons, plus 5 code examples, a hands-on project and a graded quiz. Most learners work through it at their own pace.
What will I learn from Building Coding Agents with Tool Execution?
You'll learn: Build a coding agent that writes, runs and debugs Python through an iterative feedback loop; Compare local execution, containers and microVMs on isolation strength and startup cost; Run untrusted model-generated code in a Firecracker microVM sandbox with its own kernel; Manage files, install packages and persist state inside a remote sandbox session; Build a data analyst agent that explores CSVs and generates pandas visualisations on request; Wrap an agent in a Gradio chat interface so non-engineers can ask it questions; Deploy a full-stack agent that edits multiple files and generates a Next.js application.
Topics
Sources
This page was written from 3 sources, 2 on domains other than deeplearning.ai.