Examples with real answers
We ran the mission prompts on a real AI (Claude Sonnet 5.5, on 06/11/2026, without web search) and checked the answers afterwards. One answer per prompt: this is not a benchmark. Another AI, or the same one on another day, may answer differently.
The lesson was not what we expected: the AI did well. Checking is still what turns an answer into evidence.
Mission 1: the cost report
Section titled “Mission 1: the cost report”Full answer (in Portuguese)
| What we checked | Result |
|---|---|
| Totals: 0.92 and 25.63 USD | Correct, they match the custos command |
| Biggest increase: NAT Gateway, +16.20 | Identified correctly |
| Ended with questions, without the ready-made conclusion | Yes, as the prompt asked |
| Claims beyond the file | “Lambda and Amplify probably stayed in the free tier”: that is a guess, and it said “probably” |
It becomes this proposal
- Topic: what weighs the most in a small AWS account was not what I imagined.
- Audience: people who just created their account.
- Problem: thinking the cost is always in the server.
- Goal: read a report and find the biggest increase in 1 hour.
- Evidence: the hypothesis, the AI’s answer, the output of
custosand the alert you created.
Mission 2: the options for publishing a site
Section titled “Mission 2: the options for publishing a site”Full answer (in Portuguese)
| What we checked | Result |
|---|---|
| Warned that a site on S3 requires Block Public Access to be off | Yes, and the documentation confirms it |
| Said that a site on S3 only has HTTP | Yes, and the documentation confirms it |
| 7 links cited | All of them opened. It warned that it had not opened them |
| Recommendation | S3 with CloudFront, mentioning Amplify as a simpler alternative |
Notice that the AI and the mission chose different paths, and both are defensible. CloudFront gives more control. Amplify has fewer pieces for people just getting started. Comparing and deciding is your job, and that decision makes a great paragraph for the article.
It becomes this proposal
- Topic: I published my portfolio without leaving a bucket public.
- Audience: people who have never published anything in the cloud.
- Problem: choosing between options without knowing what each one opens up.
- Goal: compare two options and publish one in 1 hour.
- Evidence: the comparison, the sources you opened, the site with HTTPS and the cleanup.
Mission 3: tutoring on the error
Section titled “Mission 3: tutoring on the error”Full answer (in Portuguese)
| What we checked | Result |
|---|---|
| Pointed to the right line | Yes |
| Held back the fix, as asked | Yes: it asked a question and waited |
Mission 3: the StatusCode 200
Section titled “Mission 3: the StatusCode 200”Full answer (in Portuguese)
| What we checked | Result |
|---|---|
| Said that 200 does not guarantee the function ran without errors | Correct, and it matched our real test |
Citation of FunctionError |
Matches the Invoke API |
202 for Event and 204 for DryRun |
Match the same page |
It becomes this proposal
- Topic: my Lambda answered 200 and it was broken.
- Audience: people learning to read logs.
- Problem: trusting only the status code.
- Goal: find an error from the log, without asking for the ready-made answer.
- Evidence: the hypothesis, the real output, the log and the documentation.
A complete example
Section titled “A complete example”Want to see it from start to finish, with hypotheses that fell apart and the finished article? See the queue example.
Independent community project by Fernando Azevedo, AWS Community Builder. Not official AWS material, and not endorsed by AWS. AWS and related names are trademarks of Amazon.com, Inc. or its affiliates. MIT license.