Study plan - AIP-C01
This is a gap-find-and-close plan, not a study-from-zero plan. The founder picks the cadence; phases below are ordered, not scheduled.
This week: take the AWS Certification Official Pretest cold, through the Skill Builder exam prep plan. Then tag each miss A, B or C and fill in the pretest log below. The rollback rule decides the next step. Skip the rest of this file until then.
Founder profile and the assumption under test
- Passed: Anthropic CCA-F (850, 2026-08-17), Snowflake GES-C02 (854, 2026-09-28), Anthropic CCDV-F (926, 2026-10-30).
- Daily hands-on Claude agent and harness work: hooks, MCP, skills, RAG, guardrails, evals, context engineering. Strong Python, SQL, data engineering, Snowflake Cortex.
- Assumption (not stated by him): GenAI design is strong and AWS service mechanics are the gap - Bedrock APIs and features, IAM, Step Functions, AppConfig, cross-Region inference, CloudWatch and X-Ray, SageMaker AI touchpoints. Phase 1 tests this before any study time is spent.
- AWS's target candidate has 2+ years building production apps on AWS or open source, plus 1 year hands-on GenAI (exam guide).
Phase 1 (week 1): official pretest, cold
Take the AWS Certification Official Pretest before reading anything in this folder. AWS's prep plan places it as the step that finds gaps in AWS knowledge (cert page). Access is through the Skill Builder exam-prep page linked from the cert page (see resources.md). Pretest length, cost and whether it reports a per-domain breakdown are (unverified).
Do not look up answers during the pretest. The point is a clean read of where he stands.
Scoring the pretest (do this right after, from the explanations)
Tag every missed question with exactly one bucket. Use the question's explanation, not memory of the stem.
| Bucket | Meaning | Examples |
|---|---|---|
| A - AWS foundations | The miss came from general AWS knowledge that is not GenAI-specific | IAM policy evaluation, VPC endpoints, Lambda vs ECS vs EC2 choice, S3 lifecycle, SQS vs EventBridge vs SNS, basic CloudWatch or CloudTrail roles |
| B - AWS GenAI service mechanics | He knew the concept but not the Bedrock or SageMaker AI feature, API or setting | Guardrails filter types, Knowledge Bases chunking options, cross-Region inference profiles, provisioned throughput, invocation logging, Prompt Management versions |
| C - GenAI design judgment | The miss is about the design choice itself, independent of AWS | Chunking strategy choice, eval metric choice, agent stopping condition, prompt structure |
If a miss fits more than one bucket, pick the one whose fix would have made him answer correctly. If still unsure, tag it B.
Record in the log below: raw percent correct, total misses, count per bucket, and the per-domain breakdown if the pretest reports one.
Rollback rule (pre-registered 2026-10-30)
Thresholds are Ray-chosen. The founder can change them before taking the pretest, not after.
Let score = pretest percent correct (raw, not scaled) and A_share = bucket-A misses divided by total misses.
- If
score≥ 60%: stay on AIP-C01. Use bucket B and C counts to order Phase 2. - If
score< 60% andA_share≥ 40%: roll back to AWS Certified AI Practitioner. The misses cluster in AWS basics, not GenAI design, so the professional exam is premature. - If
score< 60% andA_share< 40%: stay on AIP-C01, with a longer Phase 2. The gaps are Bedrock-specific or design-specific, and both are what this exam teaches.
Every input to the rule is a count he writes down, so anyone can recompute the outcome from the log. Note: the 60% raw threshold is not the pass mark. The real exam reports a 100-1,000 scaled score with 750 to pass (exam guide), and AWS says passing scores are set by statistical analysis and does not publish how many correct answers a pass takes (FAQs).
AWS lists AI Practitioner among the certs that can help before AIP-C01, alongside Solutions Architect - Associate, ML Engineer - Associate and Data Engineer - Associate (cert page). Rollback to AI Practitioner is the founder's chosen fallback.
Pretest log
| Field | Value |
|---|---|
| Date taken | |
| Raw score (% correct) | |
| Total questions / total misses | (unverified: question count) / |
| Bucket A / B / C misses | / / |
| A_share | |
| Per-domain breakdown (if shown) | D1: · D2: · D3: · D4: · D5: |
| Rule outcome (1, 2 or 3) | |
| Exam date set |
Phase 2: Domains 1-3 first (77% of scored content)
Domain weights: D1 31%, D2 26%, D3 20%, D4 12%, D5 11% (exam guide). Within Phase 2, order by bucket-B miss count from the pretest; default order is D1, D2, D3.
For each task statement, the tag in study-guide-summary.md sets the treatment:
[likely-known]: skim the skill list in the guide, learn the AWS names, move on.[mixed]: read the AWS doc page for each named feature, then answer: which service, which setting, why not the distractor.[AWS-gap]: do the lab below. Reading alone does not stick for configuration questions.
Hands-on Bedrock labs for [AWS-gap] and colleague-flagged items
Run in a personal AWS account, never a client account. Turn things off after each lab. All doc URLs were checked on 2026-10-30.
| # | Lab | Task statements | Doc |
|---|---|---|---|
| L1 | Call one model through the Converse API, then with streaming. Note the request JSON shape | 1.3, 2.4 | https://docs.aws.amazon.com/bedrock/latest/userguide/conversation-inference.html |
| L2 | Invoke a model through a cross-Region inference profile; compare Geographic and Global profiles and when data residency rules out Global | 1.2 | https://docs.aws.amazon.com/bedrock/latest/userguide/cross-region-inference.html and https://docs.aws.amazon.com/bedrock/latest/userguide/inference-profiles.html |
| L3 | Store a model ID in AppConfig, read it from a Lambda function, switch models without a redeploy | 1.2 | https://docs.aws.amazon.com/appconfig/latest/userguide/what-is-appconfig.html |
| L4 | Build a Step Functions state machine with a retry, a timeout, a fallback model branch (circuit breaker) and a human-approval step | 1.2, 2.1, 2.4 | https://docs.aws.amazon.com/step-functions/latest/dg/welcome.html |
| L5 | Build a Knowledge Base on S3 docs; ingest the same docs with fixed-size, hierarchical, semantic and no chunking; compare retrievals; add a reranker | 1.4, 1.5, 5.2 | https://docs.aws.amazon.com/bedrock/latest/userguide/knowledge-base.html, https://docs.aws.amazon.com/bedrock/latest/userguide/kb-chunking.html, https://docs.aws.amazon.com/bedrock/latest/userguide/rerank.html |
| L6 | Create a guardrail with content filters, a denied topic, a word filter, PII masking and contextual grounding; attach it to a model call and to a Knowledge Base query | 3.1, 3.2, 3.4 | https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails.html |
| L7 | Write an IAM policy that allows one model and denies others; call Bedrock through an interface VPC endpoint | 2.3, 3.2 | https://docs.aws.amazon.com/bedrock/latest/userguide/security-iam.html and https://docs.aws.amazon.com/bedrock/latest/userguide/vpc-interface-endpoints.html |
| L8 | Version a prompt in Prompt Management; chain two prompts in Flows (the guide's "Prompt Flows") | 1.6, 2.5 | https://docs.aws.amazon.com/bedrock/latest/userguide/prompt-management.html and https://docs.aws.amazon.com/bedrock/latest/userguide/flows.html |
| L9 | Read a SageMaker AI Model Registry walkthrough; know when a fine-tuned or LoRA model goes to SageMaker AI vs Bedrock, and what provisioned throughput buys | 1.2, 2.2 | https://docs.aws.amazon.com/sagemaker/latest/dg/model-registry.html and https://docs.aws.amazon.com/bedrock/latest/userguide/prov-throughput.html |
| L10 | Read the Bedrock agents and AgentCore overviews; map them to his own harness concepts (memory, tools, tracing) | 2.1, 3.4 | https://docs.aws.amazon.com/bedrock/latest/userguide/agents.html, https://docs.aws.amazon.com/bedrock/latest/userguide/trace-events.html, https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/what-is-bedrock-agentcore.html |
AWS's own hands-on options (Builder Labs, Cloud Quest, AWS Jam, SimuLearn) are named on the cert page and can replace any lab above.
Phase 3: Domains 4-5 (23% of scored content)
Mostly [likely-known] and [mixed]. One [AWS-gap] task (4.3) gets a lab:
| # | Lab | Task statements | Doc |
|---|---|---|---|
| L11 | Turn on model invocation logging (it is off by default), send logs to CloudWatch Logs, query them with Logs Insights, and trace one call with X-Ray | 4.3, 2.5, 5.2 | https://docs.aws.amazon.com/bedrock/latest/userguide/model-invocation-logging.html, https://docs.aws.amazon.com/bedrock/latest/userguide/monitoring.html, https://docs.aws.amazon.com/xray/latest/devguide/aws-xray.html |
| L12 | Run one Bedrock evaluation job (LLM-as-a-judge or RAG evaluation) on the L5 Knowledge Base | 5.1, 3.4 | https://docs.aws.amazon.com/bedrock/latest/userguide/evaluation.html |
| L13 | Read-only: prompt caching, latency-optimized inference and batch inference. Know which one answers "cheaper", "faster first token" and "large offline job" | 4.1, 4.2 | https://docs.aws.amazon.com/bedrock/latest/userguide/prompt-caching.html, https://docs.aws.amazon.com/bedrock/latest/userguide/latency-optimized-inference.html, https://docs.aws.amazon.com/bedrock/latest/userguide/batch-inference.html |
Phase 4: official practice, then timed run
- AWS Certification Official Practice Question Set. Review the explanation for every question, including ones he got right (Ryan Leonard's advice, from the founder's Glean pack).
- Re-tag misses with the A/B/C buckets. Any bucket-B miss sends him back to the matching lab.
- AWS Certification Official Practice Exam, timed. The real exam is 75 questions in 180 minutes, about 2.4 minutes per question (cert page). The cert page lists this as step 4 of the prep plan; the Glean pack did not mention it.
- Final review: optimization (D4), evaluation and troubleshooting (D5), which Glean's sequence also called out.
Third-party practice tests are optional and carry a "dated" warning (see resources.md).
Running alongside Claude Architect Professional
Claude Architect Professional is on 12-22; its new exam guide lands 11-17 and its content changes 12-07. AIP-C01 runs alongside it.
- Expect overlap on agents and tools (2.1), safety controls (3.1) and evaluation (5.1). Whether the Architect guide covers these is
(unverified)until 11-17. - When the Architect guide lands, mark any AIP-C01 task statement it also covers, so one study pass counts twice.
- The AWS-gap labs (L2, L3, L4, L7, L9, L11) have no Architect overlap. They are the natural items to pause or keep going around 12-22. His call.
Exam date
TBD. Set it after the pretest and the rollback rule outcome. Retake wait after a fail is 14 days, and each attempt costs the full 300 USD fee (FAQs, cert page).