Agentic Engineering

A hands-on course for engineers who want to move from first agent session to shipped work they can review, prove, and hand off.

Pick one small, real project. Carry it through the lessons. Finish with a shipped slice, proof commands, and the habits behind the work.

For teams, AISOFT runs Agentic Engineering Day as the in-person version: shared lab, live review, and a first-week rollout plan.

How the course works

• 35 lessons, 7 parts. Short to read, longer to do. Take them in order if you’re new to agents; jump around if you already are.
• 7 milestones. If you want the guided route, start with the Milestone path. It tells you what to read, what to do, and what proof you need before moving on.
• One project, the whole way through. You pick it in Lesson 3.1. Every lesson after that puts the new skill to work on your project. The project compounds; the lessons make sense because they land on it.
• Every lesson ends with an exercise and a plain “you’re done when.” Reading without doing won’t move you, but the doing is the satisfying part. That’s where the skill actually appears.
• We teach Claude Code first, then translate. Codex CLI, Gemini CLI, and Snowflake Coco get variant notes where the workflow changes. See CLI variants.
• Where this leads. The role this course builds toward — sit inside a customer’s team, ship agentic systems into their stack, train their people, hand off clean — is where the industry is now placing its biggest bets. Microsoft’s $2.5B Frontier unit (6,000 embedded engineers), and parallel direct-delivery pushes from Anthropic, Google, and Amazon, are all deploying people into customer teams because the model, not the demo, is where AI actually lands. That embedded role is the Forward Deployed Engineer. If you want to take it that far, the Two-week FDE ramp is a structured day-by-day plan. FDE is a destination, not a prerequisite for starting.

Who can follow it

This is beginner-first, not beginner-only.

For facilitation, use the practice run guide.

For instructors and mentors, use Teaching agentic engineering. It captures the teaching loop, external references, workshop formats, and the common failure modes to watch for.

For concrete projects, use the Data + AI practice labs: 30 exercises across SDLC, data engineering, big data, analytics, governance, data science, ML/MLOps, AI apps, backend, fresh-grad portfolio, team lead adoption, and agentic workflows.

For job-focused learners, use Job pathways + AISOFT offerings. It maps pathways to roles, portfolio proof, interview stories, and AISOFT service lines.

For page, slide, handout, or workshop copy, use the AISOFT Agentic Engineering brand system.

Fresh graduates and career switchers should keep Beginner prep: Git, terminal, and language basics open while starting the milestone path.

Milestone route

Use this route if you want someone to go from zero to useful, with visible proof at every step.

Open the full Milestone path for the step-by-step version with commands.

Before you start

You need three things, nothing more:

1. A computer with a terminal — Mac, Linux, or Windows with WSL.
2. One AI coding tool with a paid plan — Claude Code is the default; Codex CLI or Gemini CLI work too. You install it in Lesson 2, so don’t set it up yet.
3. Enough coding comfort to read a diff — you don’t need to be an expert. You need to look at a change and judge whether it’s right.

New to Git, the terminal, or a programming language? Spend an hour with Beginner prep first, then come back. That is the only prerequisite reading. Everything else is taught inside the lessons.

Ignore certifications and the FDE track for now. They are not prerequisites for anything. Start the lessons, ship something, then see the After you finish section for credentials and next steps. Keeping them out of the way is the point.

The path

Part 1 · Foundations

The conceptual stack you walk in with, so the workflow lessons land cleanly. Senior full-stack engineers from any background can skim quickly; nothing here assumes prior AI work.

• A. The AI map
• B. LLMs: just enough to be dangerous
• C. What makes an agent
• D. Multimodality
• E. The model zoo

Part 2 · Get set up

From nothing installed to your first agent session.

1. What is agentic engineering?
2. Install your tool
3. Your first session
4. Staying in control

Part 3 · Build something real

Pick a project and take it all the way to shipped.

1. Pick your first project
2. Plan before you build
3. Build it, step by step
4. Ship it

Part 4 · The core concepts

The discipline that makes agent work hold up under real users.

1. The brief
2. Evals
3. Context and the second brain
4. The no-slop standard
5. Design discipline

Part 5 · Scale up

From one agent to a way of working.

1. Orchestration
2. Review gates and shipping
3. Where to go next

Part 6 · Operating in the real world

The envelope you walk out with. The eleven lessons that turn a working agent engineer into a Forward Deployed Engineer.

• 1. The harness wars
• 1. The application taxonomy
• 1. Coordinating with agents and humans
• 1. Staying current: the intel-watch pattern
• 1. The team shape
• 1. The problems in every layer
• 1. Discovery and scoping the engagement
• 1. Communicating to non-engineers
• 1. Observability and cost discipline
• 1. Security and compliance for AI products
• 1. The handoff playbook

Part 7 · Building on the platforms

From driving an agent to building on the raw platform. These three lessons take you under the CLI to the API, the agent SDKs, and the tool protocols that connect agents to the world. Tool-agnostic by design, and they map one-to-one onto the four domains of the Claude Certified Architect - Foundations exam (Claude Code, Claude API, Agent SDK, MCP).

• 1. Building on the API
• 1. Programmatic agents (the SDKs)
• 1. Tool protocols: MCP and connectors

What you’ll walk away with

• A real project, built, shipped, and yours. Not a tutorial toy.
• The scaffolding installed on it and actually used: a CLAUDE.md (or AGENTS.md), a DESIGN.md, an eval suite, a no-slop review pass, a HANDOFF.md, and a decisions/ log.
• The habits: a brief before code, evals before you call it done, parallel work when the pieces are independent, and proof before you say “shipped.”
• A tool-agnostic workflow you can run in Claude Code, Codex CLI, Gemini CLI, or Snowflake Coco.
• A working personal intel-watch so you stay current on the people whose signal matters, without doom-scrolling.
• The operating mode of a Forward Deployed Engineer: walking into a customer’s stack, scoping, building, shipping with proof, and handing off cleanly.
• A teaching model you can reuse for mentoring, workshops, team enablement, or hiring filters.

After you finish

You’ve shipped a real slice with the habits behind it. Everything below is optional and off the critical path. Do the ones that serve you, in whatever order:

1. Go deeper on the craft — Lesson 16 Where to go next, and if you want customer-facing work, the Two-week FDE ramp.
2. Get a credential — this curriculum maps to the Claude Certified Architect – Foundations exam. The Certifications and frontier partnerships page has how to enroll and the lesson→exam crosswalk (plus cloud-provider certs and partner programs). Portfolio first, cert second.
3. Turn it into a job — Job pathways + AISOFT offerings.

None of these are prerequisites for anything. This is where credentials and partnerships live, so the lessons themselves stay focused on shipping.

The artifacts

Three forkable starters back the course. You install each one during its lesson:

• no-slop-skill/: a review pass the agent runs against its own output (Lesson 4.4).
• templates/DESIGN.md: a design-quality spec for anything with a surface (Lesson 4.5).
• second-brain-starter/: a CLAUDE.md skeleton and memory structure (Lesson 4.3).

Maintained by Ravinder Jilkapally · AISOFT · Mentoring: book a free 30-min session