Skip to content

AB-730 deep review — September 27, 2026

The complete guide was read and 41 published October objectives in eight groups were mapped. The July 22 accepted baseline remains separate from October 20 preparation. The new fourth domain requires applying agents and supervising Cowork; the earlier construction group remains useful for the July baseline without standing in for the new scope. The credential confirms 45 minutes, 13 languages and the October revision; the official instructor-led course remains one day.

Learning changes

Four original examples cover notebook references that outlive a sanitized summary, gross versus net growth, missing meeting capture and the scope of an action approval. The guide now includes ten labs and 44 answered checks, with explicit preparation for model selection, external connectors, visuals, meeting tools, prebuilt agents, skills and supervised automation.

Current documentation supports these practical distinctions:

  • Model choice, source permission and factual correctness are separate checks.
  • Indexed and live connector retrieval have different freshness boundaries.
  • Sharing a notebook can invite recipients to its linked files.
  • Saved memories, history/work inferences, instructions and retention differ.
  • Private catch-up, shared Facilitator notes and intelligent recap serve different purposes and depend on capture, access, licenses and supported meeting types.
  • A Cowork approval can cover later session actions; automated tasks can carry separate preauthorization. Cancellation does not promise rollback.
  • Skill evaluation is not certification of business correctness. Shared updates require review and a new-session test; browser tasks have their own identity and supported-surface requirements.

Eight synthetic arithmetic/set assertions checked the worked growth and capture figures, reference-set difference and course-duration sum. No Microsoft 365 tenant, sharing, agent, model, meeting, browser, skill or automation lab was executed. All ten labs remain learner exercises with unavailable steps recorded explicitly.

The blueprint requires filtering chats by agent, but the reviewed public history article does not establish a universal filter location. The guide provides a rollout-qualified practice task and fallback instead of inventing exact menu steps. Tenant-level verification remains necessary.

Resources and blog decisions

The linked Learn path now lists four modules under Enhance business workflows with AI. The earlier six-module/4h31 statement was removed as a current claim; individual modules and the revised duration were not independently reviewed. Pluralsight lists two Vlad Catrinescu courses totaling 3h01, rounded to three path hours, with the second dated September 11. It remains in production.

O'Reilly's public Renaldi Gondosubroto agenda totals 180 minutes plus breaks and still teaches three domains. MeasureUp still lists 110 June questions, an unrelated generic Azure-AI description and older three-domain AB-730 details. Neither page establishes complete October coverage. Udemy blocked retrieval, so its earlier metadata is explicitly historical. No paid questions, lessons, video runtime, signed-in Practice Assessment or partner event seats were verified.

Two September 25 Microsoft articles were read and linked with exercises:

  • Jared Spataro's Home, Code and Autopilot announcement supports a comparison of context, action and supervision needs, while preserving staged Frontier/private previews and the distinction between product news and exam objectives.
  • James Oleinik's Work IQ business-context article supports a renewal-evidence and approval table. Its September 30 preview start, October rollout and planned late-October Finance/Operations support remain future observations as of review.

Videos, linked demonstrations, customer deployments and performance claims were not independently validated. New research checkpoints cover business-app grounding, Copilot interface rollout and announced early-October federated connector writes.

ADLC_Docs/operations/2026-09-27-ab-730-deep-review.json preserves prior records, guide/objective hashes, all objective mappings, source observations, findings, blog decisions and publication checks. All 35 unique guide links are registered: 34 were reachable, one was access-blocked and none was missing/error. Research and repair used the same AI context; independent and human review remain pending. Publication requires unit tests, repository validation, a strict site build, generated-site validation and a diff check. The Microsoft tracker records the remaining program.