AI Architect Academy sample module · v0.8 plan
Your placement (from diagnostic)
AWS cloud engineer → builder-architect track. You already own IAM, VPC, Lambda, CloudWatch and cost, so we're skipping Track A (the bridge). Target: screen-ready for AWS-partner GenAI shops like Automat-it (Bedrock AgentCore, agentic). This is Track B · Module 2.

The Agentic Loop & Tool Use CCA-F Domain 1 · Agentic Architecture (highest-weighted) · ~25 min

Maps to: CCA-F D1 · AWS GenAI Dev Pro (Bedrock agents) · real-work: the harness you own
01
Objectives
02
Lesson
03
Retrieval
04
Mastery
05
Coach
OBJECTIVE MASTERY
0%
Backward design · what you'll be able to do

Objectives

Every objective below maps to a question in the retrieval set and to the mastery gate. You don't advance on time spent — you advance when you can do these.

  • LO1Trace one full iteration of the agentic loop and name exactly what ends it.
  • LO2Explain why you branch on stop_reason instead of reading the model's prose for tool intent.
  • LO3Tell syntactic (malformed) from semantic (valid-but-wrong) tool errors, and say where each is caught.
  • LO4Bound a loop — budgets, done-conditions, escalation — to prevent runaway behavior and excessive agency.
  • LO5Reframe a senior's "make it deterministic" instinct into evals + guardrails.
recall > reread — there's a question for each of these
Lesson · framed as deltas from what you already know

Your code stops being the decision-maker

As a senior, your instinct is to own control flow: if this, call that. In an agent, you hand the branching decision to the model and your code becomes the loop harness around it. That single inversion is the whole mindset shift — everything else follows from it.

The loop, precisely

# the harness you write — not the intelligence, the plumbing
while True:
    resp = model.send(messages, tools)
    if resp.stop_reason == "tool_use":        # model decided to act
        result = run_tool(resp.tool_calls)     # YOU execute + validate
        messages.append(tool_result(result))  # feed it back
        continue                              # loop again
    if resp.stop_reason == "end_turn":        # model is done
        break

Read it as a sentence: send → check stop_reason → if it wants a tool, run it and feed the result back → repeat → stop when it's done. The model never touches your systems directly; it requests, your harness executes. That gap is where all your leverage (and all the risk) lives.

Map it to what you know
It's an event loop / a REPL / a retry-with-side-effects loop — except the branch decision each turn comes from the model, not your if statements. You've written this control structure a hundred times; only the decider changed.

The four footguns (where seniors actually get burned)

  • Reading the prose, not the structure. Don't scan the reply text for "I'll call search()". The response tells you structurally via stop_reason + tool_use blocks. Parsing prose is guessing.
  • Assuming schema-valid = correct. A typed tool schema stops malformed calls (syntactic), not wrong-but-well-formed ones (semantic). A valid account_id can still be the wrong account. You validate values in the tool.
  • No budget = runaway. Without a turn/tool cap and a done-condition, a loop can spin 40 times burning tokens — or take an irreversible action it shouldn't. This is OWASP LLM06 Excessive Agency: bound it, gate high-risk actions, keep a kill switch.
  • Fighting non-determinism. Same input can take a different tool path. Don't chase a deterministic path — judge outcomes with evals and constrain behavior with guardrails.
Why this is your job, builder-architect
At an AWS-partner shop like Automat-it, this loop is the Bedrock AgentCore agent. The harness, tool boundaries, stop conditions, cost controls, and safety gates all live here — and they're exactly the production instincts you already have from running systems. This module is you porting those instincts onto a probabilistic decider.
Retrieval practice · scenario items, CCA-F style
Question 1 of 5
pick the option you'd defend in a code review
Mastery gate · advance on competence, not time

Mastery check

0/5

Your spaced-repetition schedule

Items you missed resurface sooner; items you nailed return on a longer interval. That's the forgetting curve being managed for you — recall right before you'd forget.

Objective / itemResultNext review
The coach · scalable 1:1 tutoring (the lever, §3)

A real corrective turn

Illustrative transcript: a learner picks "lower the temperature" to stop a runaway loop (a very common senior conflation). A good tutor doesn't say "wrong" — it finds the misconception, ties it to what you already know, and hands the loop back.

Learner
For the runaway-loop one I went with lowering temperature — fewer random detours should mean it stops looping, right?
Coach
Quick gut-check first: what does temperature actually control? It biases token sampling — how adventurous the next-token pick is. It has zero authority over how many times your loop runs. So the runaway isn't a sampling problem; it's a control-flow problem — your harness never told the loop when to stop.

The fix lives in your code, not the model's decoding params: a turn/tool-call budget, an explicit done-condition, and a human escalation when you hit the cap. You already do exactly this with rate-limiters and circuit-breakers — same instinct, new decider. Want to redo that item with the control-flow lens?
What just happened (by design): diagnosed the specific misconception (temperature ≠ loop control) · connected it to existing senior knowledge (rate-limiters / circuit-breakers — adult learning) · gave the correct model, not just the verdict · prompted a retry so the item re-enters the mastery + spaced-repetition cycle.
How this 25 minutes maps to the gold-standard pedagogy (§1c)
Backward design
Measurable objectives (LO1–LO5) defined first; every item and the gate trace back to one.
Retrieval practice
You recalled under scenario pressure instead of re-reading — the testing effect is the core mechanic.
Spaced repetition
The schedule table: missed items return on short intervals, mastered ones stretch out.
Mastery gating
≥4/5 to advance — competence, not completion.
1:1 tutoring
The coach turn diagnoses a misconception and adapts — the 2-sigma lever, at scale.
Adult / expert
Deltas from known mental models, scaffolding faded fast, no beginner pacing (expertise-reversal aware).
Diagnostic path
Placement banner: an AWS engineer skips the bridge and targets Automat-it's actual stack.