Georgia State University — J. Mack Robinson College of Business PATH — Pathways for AI Training & Hiring CIS 4394 Agentic AI  ·  Fall 2026  ·  Dr. Xinyu Fu
Week 2 · Learning from the best

Unpacking coding agents, then building your own.

Reverse-engineer Codex. Discover that the same architecture explains Claude Code. Then transfer the pattern into a completely different domain: a Job Search Agent that thinks before it acts.

CourseGSU CIS 4394 Agentic AI
InstructorDr. Xinyu Fu
TermFall 2026
FormatHands-on studio class
📡 Agent Radar · this weekAgents are getting real powerThree stories from the past few days: physical power, economic power, and a containment incident. Think → Act → Coordinate → Spend → Affect the physical world. →
📝 After classWeek 2 SummaryScavenger-hunt answer key (nine layers → real Codex files) · join your GA1 group on iCollege · Opal / Fleet / ADK tutorials with videos · Quiz 1 readings + five sample questions. →
See it move · inside a coding agent

The harness does more than the model

When you type “fix the failing test” into Codex or Claude Code, the model never touches your files. A harness assembles the context, runs the tools, asks you before anything risky, and decides when to stop. Press Play and watch one repair from prompt to passing tests, then the same task with a bloated context.

Try this: count how many times the model is called in the first run. Then look at the system-prompt page: which of the nine layers did you just watch?

Go further (new, October 2026): Harness², Recursive Agent Harnessing for an Open World ↗ from Google Research is the strongest recent evidence for this page’s claim. It improves an agent by editing its harness while the model weights stay fixed: a Propose–Probe–Compose step tries harness edits on the task without ground-truth checks and keeps the ones that help, and a second level turns those edits into lessons for future tasks. The gains it reports (up to 11.5 points on WorkBuddy-Bench) are the authors’ own numbers. Code on GitHub (Apache-2.0) ↗; the local demo runs without credentials, model-backed runs need a Google Cloud project.

Start here

The path through today

Work through the four parts in order. Do Prep before you come to class; the session starts with building, not with account recovery.

Discussion questionIf Codex had the exact same model, but no access to files, shell commands, tools, state, or permissions, would it still be Codex?
No: you would have a chatbot that talks about code. Everything that makes it Codex (reading files, running tests, applying patches, pausing for approval) lives in the harness. The model supplies reasoning; the product is the loop built around it. Part 1 takes this answer apart layer by layer.
Today’s arc

Reverse-engineer → compare → transfer → build

Unpack Codex

What architectural ingredients turn a language model into a coding agent?

Compare Claude Code

Does the same architecture explain a competing product, or is it product-specific?

Transfer & build

Keep the reasoning pattern; swap the goal, environment, tools, and safety boundary.