AISOFT LLC Agentic Engineering

Lesson 5.3 · Where to go next

Where this gets you: you’ll pick your next project, one step harder than the one you just finished, and know which skill it’ll push on.

The idea

You built a real thing, shipped it, and added evals, no-slop review, design standards, and a second brain. From here, go deeper.

There’s more in Claude Code than this course used:

  • Skills. Reusable multi-step workflows you invoke by name, so a routine becomes one command.
  • MCP servers. Connections to external tools. GitHub, a database, Slack. The agent reaches past your files into the systems your work lives in.
  • Hooks. Scripts that run before or after agent actions. A test on every edit, a reminder on every deploy.
  • Agent SDK. For building your own agent, in Python or TypeScript, instead of using one.

Don’t go learn them. Pick a project that needs one, and you’ll learn it on the way.

Do it

Here’s the ladder. Each rung leans hardest on one course skill.

  1. An internal tool, well-bounded. No users to break. You get to practice clean evals with nothing on fire.
  2. A greenfield product with a small surface. Now DESIGN.md earns its place. Ship it to five real people.
  3. A feature inside an existing codebase. Now the second brain matters. The agent has to learn a system it didn’t build.
  4. A multi-stream build under time pressure. A hackathon, or a weekend you box yourself into. Now orchestration decides whether you finish.

The rung tells you what will hurt. Say you jump to rung three, a feature in a codebase you’ve worked in for a year. The agent writes a clean, well-tested handler and puts it in the wrong place, because nobody told it this repo routes everything through one dispatcher. You send it back four times. That’s not the agent failing. That’s your second brain being empty. Better to learn that here than on a hackathon deadline.

Your exercise

Pick your next project, one rung up from the one you just finished. Built an internal tool? Do the greenfield product. Did greenfield? Go into an existing codebase.

Then name, in one sentence, which skill from this course it’ll stress most.

You’re done when your next project is chosen and you’ve named the skill it’ll push hardest.

Build on it: a CLI that reads your repo’s git log and prints the five files changed most often — the map a new agent needs before it touches anything.

Why this matters

This is a transition, not a tool install. The first weeks feel slower, because you’re writing briefs and checks and proof instead of code. Keep going until it stops feeling like overhead and starts feeling like how you work. The way there is the next project, then the next.

Previous: Lesson 5.2 · Review gates and shipping · Next: you’ve finished the course. Go build.

Proving your skills and working closer to the models: see Certifications and frontier AI partnerships — cloud certs, portfolio-first credentialing, and partner programs with Anthropic, OpenAI, Google, Microsoft, and AWS.