Course 4.6 · Curriculum 2026.1
The multi-agent adversarial audit
Nine agent result blocks, 87 findings anchored to file and line, condensing to five root causes, all confirmed against live systems and fixed the same day.
- Estimated effort
- 5 hr
- Lessons
- 42
- Modules
- 7
- Material
- 5 hr
Estimated effort is the sum of the lesson and lab times in the course outline, for this version of the course.
This course is planned and not on sale. The curriculum below is the plan, and it may change before the course opens.
See the courses on sale01What you will learn
7 capabilities, stated as things you can do
Each one is what you should be able to do at the end, not what the course covers. If an outcome is not testable, it is not an outcome.
- Design an audit where several agents attack the same system from different angles.
- Require every finding to be anchored to a file and a line, or discard it.
- Condense many findings into root causes without losing the evidence trail.
- Confirm a finding against a live system before it becomes work.
- Recognise the two AI-specific failure classes this audit found.
- Build the heartbeat that would have caught a silently dead job.
- Report an audit so a sponsor can approve the fixes in one sitting.
02Curriculum
7 modules, 42 lessons
Every lesson shows its length and its type, from the course outline.
7 modules · 42 lessons · 5 hr of material
Several agents, several angles, one anchoring rule. A finding with no file and line is an opinion.
- VideoThe angles: security, correctness, performance, operationsPreview8 min
- VideoBriefing agents so their angles do not overlap6 min
- VideoThe anchoring rule: file and line, or it does not countAnchoring is what makes 87 findings triageable rather than a wall of prose. It also makes a false positive obvious in seconds.7 min
- ReadingWhy anchoring makes a false positive obvious in seconds5 min
- LabLab: brief three agents on three non-overlapping angles10 min
- QuizCheck: designing the audit8 min
Nine agent result blocks is a lot of output. The template is what keeps it triageable instead of a wall of prose.
- VideoRunning nine result blocks without drowning in them7 min
- VideoA finding template that forces file, line and evidence6 min
- ReadingWhat to do with a finding that has no anchor5 min
- LabLab: run a three-angle audit on a small repository11 min
- QuizCheck: running the audit8 min
- VideoKeeping nine result blocks from becoming one wall of prose6 min
87 findings became five root causes. The condensation is the analysis, and it is the part an agent does worst on its own.
- VideoClustering findings without losing the evidence7 min
- Video87 findings, five root causes, and how the condensation happened6 min
- ReadingThe part an agent does worst on its own5 min
- LabLab: condense a finding set into root causes11 min
- QuizCheck: from findings to root causes8 min
- VideoConfirming against live systems before committing work7 min
Two of the five root causes were AI-specific, and both were invisible until someone went looking on purpose.
- VideoThe agent that was silently dead for three weeksA model name baked into a serialized object at deploy time. Nothing errored loudly, so nothing alerted, so nobody looked.7 min
- VideoA model name baked into a serialized object, and why nothing errored6 min
- VideoThe digest that died with an auth error6 min
- ReadingNineteen days unnoticed, because no failure alert existed anywhereNo alert existed for the absence of a run. Build the heartbeat before you need it.5 min
- LabLab: find the silent failure in a sample system10 min
- QuizCheck: the two AI-specific findings8 min
An agent had been dead for three weeks. A digest failed unnoticed for nineteen days. Neither one made a sound.
- VideoSilence is not success, and the heartbeat that proves it7 min
- VideoBuilding the heartbeat6 min
- VideoAlerting on absence, not just on error6 min
- ReadingWhat a heartbeat should check, and how often5 min
- LabLab: build a heartbeat and trigger the alert10 min
- QuizCheck: silence is not success8 min
A finding becomes work only after it is confirmed against the live system it claims to describe.
- VideoConfirming a finding against a live system before it becomes work7 min
- VideoFrom confirmed finding to same-day fix6 min
- ReadingWhat “confirmed” has to mean before a sponsor sees it5 min
- LabLab: confirm three findings against a live system10 min
- QuizCheck: confirming and fixing8 min
- VideoWhy an unconfirmed finding stays a hypothesis, not a task6 min
The report is what turns 87 findings into a decision a sponsor can make in one sitting.
- ReadingReporting an audit a sponsor can approve in one sitting7 min
- VideoWhat belongs on one page, and what belongs in the appendix6 min
- VideoRanking five root causes by what they cost to leave unfixed6 min
- LabLab: write the one-page summary of a finding set10 min
- QuizCheck: the adversarial audit8 min
- VideoGetting sign-off in the meeting, not after it6 min
04Before you start
What you need first
Assumed knowledge and setup
- Courses 4.4 and 4.5.
- A codebase and an org you are allowed to audit.
- Read access to production logs, or the audit cannot confirm anything.
Before this course
Courses often taken next
- 4.7 · Automated backlog triage against a live org
- 4.8 · Deploying agents, and the footguns that cost three weeks
- 4.10 · Adversarial code review, and the defect five reviewers found
Each is a separate course, bought on its own.
05Who teaches it
Elliot Saha
Co-founder, Chief Technology Officer
Platform architecture, custom Apex and Lightning Web Components, AI systems, and the Academy platform itself.
06Reviews
No learner reviews yet
This course has not been taken by enough people to publish an honest rating, and we will not print invented quotes on a page that sells verification discipline. Here is what we can evidence instead.
- Assessment
- Lesson checks onlyThis course carries lesson checks and issues no certificate.
- Curriculum version
- 2026.1You enrol into a version. Content changes do not move you mid-course, and your required-lesson count is snapshotted at enrollment.
- Refunds
- 14 daysAn individual purchase can be refunded within 14 days if you have completed less than 20% of the lessons. Team seats are refundable before they are assigned. From the Academy terms.
Start 4.6
This course is in development and not yet open for enrolment. The courses open now are listed in the catalogue.
- 5 hr estimated effort, labs included
- A ShareCo certificate is not a Salesforce certification