The 2026 AI Agent Engineering Roadmap

A 6-phase, 17-week guide to building, harnessing, evaluating, and shipping production agent systems. Primary sources only. ~8,000 words.

📖 Read it: ROADMAP.md

---

Give it to your agent

Two ways to use this with Claude Code, Codex, Cursor, OpenClaw, or any agent that can fetch URLs.

Option A — Just read the roadmap

Paste this into your agent:

Read https://raw.githubusercontent.com/codejunkie99/agent-roadmap-2026/main/ROADMAP.md and use it as my reference for agent engineering.

The agent fetches the full roadmap as context. Done.

Option B — Personalize it for me (recommended)

Paste this into your agent:

Follow the protocol in https://raw.githubusercontent.com/codejunkie99/agent-roadmap-2026/main/AGENT.md to build me a personalized agent engineering roadmap.

The agent will:

1. Load the canonical roadmap.
2. Ask you 5 questions (level, hours/week, stack, provider, goal).
3. Write a tailored MY_ROADMAP.md to your project — adjusted timeline, filtered resources, checklist of deliverables.
4. Write an AGENTS.md so future agent sessions in that folder pick up where you left off automatically.

You end up with a working plan in your repo, not a tab full of notes.

---

Files

File	Purpose
ROADMAP.md	The full canonical roadmap.
AGENT.md	Protocol an agent follows to personalize the roadmap for a user.
README.md	This file. Entry point and the two prompts above.

---

Why this exists

Anthropic's own number: same model, different harness, 78% vs 42% on CORE. That gap is harness engineering, and the field is bottlenecked on engineers who can build it. This roadmap is the path from "I've called an LLM API" to "I ship production agents that don't regress."

If you find a broken link or want to suggest a better resource, open an issue or PR.