Everything you wanted to ask before signing up.

Software engineers who can already build software and now want to design agentic AI systems with real judgment — not just wire up a framework. If you can ship code but freeze when asked to design an agent that won't misfire in production, you're our target learner.

You should be comfortable as a working software engineer — APIs, databases, and at least one backend, plus having called an LLM API once or twice. You don't need to already know ReAct, RAG, tool-calling, or multi-agent orchestration. The graph teaches those from the decisions up.

Ten canonical agentic systems, each designed end-to-end: a customer-support agent, a knowledge/RAG assistant, a text-to-SQL agent, a coding agent, a deep-research agent, chat at scale, a memory agent, a multi-agent system, an actuation/tool-use agent, and evaluation & observability. Across them runs one reusable toolkit of ten cross-cutting decision axes.

For most users, a few months at ~30 min/day takes you from a cold start to designing all ten canonical systems with defensible trade-offs. Users with strong agent priors finish sooner; users new to LLM systems take longer. The dashboard tells you where you are — you don't have to guess.

No. Frameworks teach you their API. AgentGraph teaches the decisions underneath any framework — autonomy and control, context strategy, memory, tool design, model routing, control flow, evaluation. Learn those and you can pick or build the right tool, instead of being trapped inside one.

The web app works on phones for review and short drills. For the longer design drills — where you're writing out trade-offs and sketching control flow — you'll want a laptop.

$9 per month, or $90 per year (two months free). No hidden tiers, no per-feature upsells, no ads.

We do a 7-day money-back guarantee instead of a free trial. Subscribe, use everything without restrictions for a week, and if it didn't move the needle we'll refund you — no back-and-forth.

Yes, from the dashboard. Cancel ends your subscription at the next billing period. Pause freezes your spaced-review queue — useful if you're on vacation or heads-down on a ship.

Not yet. Try it as an individual first; once you've tried it, reach out via support and we'll work something out if your company wants to sponsor it.

All major credit/debit cards via Stripe. No PayPal, no crypto, no local bank transfer at this stage.

Agentic AI is a fast-moving target — a bottom-up glossary of definitions goes stale and never builds judgment. Top-down learning (play the whole game first, then descend) drops you into a real system, lets it fail, and teaches each fundamental exactly when a failure forces you to reach for it. You retain it because you needed it.

Courses optimize for coverage ("we taught everything") at the expense of retention. Mastery-based learning refuses to let you advance past a decision you can't yet defend. The result: the judgment you build, you keep.

The graph tracks when you last defended each decision. Decisions you've just learned resurface quickly; ones you've nailed several times come back at longer intervals. By the time you're designing an agent for real, nothing in the graph has had time to fade.

Each drill has a canonical answer and a grading rubric that lists the specific properties your response has to satisfy — did you name the alternatives, the criteria, your choice, and what you trade away? Your answer is scored against that rubric, you get feedback pointing at the exact gaps, and the ideal answer is revealed so you can compare.

It can be, especially on judgment-call trade-offs. When you disagree, you can flag a grade — every flag goes into a review queue where we tighten the rubric. Over time, disagreements become rarer as the rubrics get sharper.

Because 30 focused minutes of calibrated practice crushes 2 hours of passive reading. Short sessions are easier to commit to, keep your recall active between days, and fit into an actual working life. You can do more than one session if you want — diminishing returns kick in faster than you'd think.