Patterns from the field.

The catalog. In one place.

Each card · the app (fictional), the industry, what the team would have done without governance, what AADM tools moved the conversation, and an illustrative hours-saved range. Read straight through, or skim for shapes that match your work.

• 01

  Media

  StudioPublish

  12–18 hrs saved

  Two exits, not one Back

  Producers drilling into an export sub-page kept landing in the wrong parent — campaign, QC queue, or library — because a single Back control ignored where they came from. ux_surface_pass, user_flow, and pressure_test modeled two intentional exits (return to origin vs return to hub) and named component tests before any header tweak shipped.

  Tools the agent called

  ux_surface_pass user_flow pressure_test
• 02

  Edtech

  Vertex Learning

  20–30 hrs saved

  Integration gap inventory

  Product committed to a full publish loop across slide, voice, video, and LMS vendors, but engineering only had partial async jobs and fourteen scattered tickets. ux_surface_pass captured the ten-step author journey; signal_stack_outline and journey_attest tagged what was proven vs latent on operator monitoring — producing one attested requirements brief instead of repeated cross-team workshops.

  Tools the agent called

  ux_surface_pass signal_stack_outline journey_attest
• 03

  Retail

  Northstar Retail

  30–45 hrs saved

  Slice the 4k monolith

  One merchant-admin screen (~4k LOC) shared URL state and cross-deck cache invalidation across three in-page decks — a split-without-plan would have broken purge auth. route_lane and plan_delivery sequenced hook attribution first; fetch_auth pulled tenancy controls in; architecture_lens pressure-tested purge vs archive. Named slices shipped instead of delete-and-pray.

  Tools the agent called

  route_lane plan_delivery fetch_auth architecture_lens
• 04

  Analytics

  Pulse Analytics

  16–24 hrs saved

  False-green dashboards

  Tenant lifecycle tests passed while finance disputed usage percentages — client DTOs fabricated fields and mocks used the wrong metric name (unattributed_pct vs unattributed_usage_pct). prism_frame split integrity vs measurement; sentinel_closure forced explicit pass/fail on schema seams — surfacing false-green dashboards separate from delete proofs.

  Tools the agent called

  prism_frame sentinel_closure
• 05

  Fintech

  LedgerFlow

  24–40 hrs saved

  Billing cycle math

  Month-end API credits overshot finance expectations because the usage query treated cycle end as inclusive when the contract said exclusive — and one webhook path skipped balance checks. debug_defect and prism_frame kept the thread evidence-first; journey_attest mapped reconcile nodes. Ledger fixes landed without dashboard band-aids or “upgrade the database” detours.

  Tools the agent called

  debug_defect prism_frame sentinel_closure journey_attest
• 06

  Insurance

  Harbor Insurance

  8–14 hrs saved

  CORS preflight trap

  Members saw empty account settings in production while Node integration tests passed — a custom client header was missing from the CORS allow-list. debug_defect chained DevTools proof → header setter → allow-list → contract test, halting timeout-padding guesses. Fix shipped in one attested path.

  Tools the agent called

  debug_defect
• 07

  Construction

  BridgeWorks

  10–16 hrs saved

  Missing projection row

  A new work package appeared in the spine but handoff preview returned 500 — partial projection lists skipped cold-start materialization. On-call proposed doubling query timeouts. triage_bug ruled out infra; digfind targeted materialization policy. Scoped fix, no rollback.

  Tools the agent called

  triage_bug digfind
• 08

  Regtech

  Atlas Compliance

  12–20 hrs saved

  Alert noise, not incident

  Monitoring fired 258 critical events in one day — all one self-healing backlog tick with duplicate worker emits and zero stale rows. Leadership wanted rollback. triage_bug and classify_bug downgraded to telemetry hygiene, not customer incident.

  Tools the agent called

  triage_bug classify_bug
• 09

  Logistics

  FleetDispatch

  25–35 hrs saved

  Three IDs, silent FK failures

  Mid-migration, org slug, depot UUID, and middleware scope coexisted while export routes still wrote slug into UUID FK columns. prism_frame documented partial ship; fetch_seam (SEAM-06) named contract drift; map_feature listed four product blockers. Cutover continued with a decision log — not depotId ?? orgSlug spread.

  Tools the agent called

  prism_frame route_lane fetch_seam map_feature
• 10

  Healthcare

  Meridian Health

  14–22 hrs saved

  CI cites ghost tests

  Proof-pack scripts referenced deleted 1,300-line monoliths while invariants register still claimed they ran — auditors assumed coverage that didn't exist. fix_pack required file-absence evidence before cutover; route_lane kept it Investigator hygiene. Registry matched disk.

  Tools the agent called

  fix_pack route_lane
• 0B

  HR tech

  ClearPath Talent

  18–28 hrs saved

  Skills map demo widget

  A static three-row chip map proved nothing; engineering proposed client-side D3 with invented weights while server constellation APIs already existed. ux_surface_pass and map_feature tied Expand, filters, and clustering to authoritative read models before wireframes.

  Tools the agent called

  ux_surface_pass map_feature user_flow
• 11

  Media

  StudioPublish

  40–60 hrs saved

  JSONB drift 500

  Publish dashboard returned 500 when stored queue schema lagged reader expectations. Round 1 fixed the throw, but Round 2 found taxonomy saves not invalidating projections — UI would lie “ready” on stale verdicts. debug_defect plus journey_attest closed both classes in one train.

  Tools the agent called

  debug_defect journey_attest prism_frame
• 12

  Regtech

  Atlas Compliance

  32–48 hrs saved

  Computing forever

  Audit summaries hung for 20+ minutes — ten concurrent recompiles stampeded advisory locks at 8s pooler timeout while SQL ran in 5ms. Team wanted global timeout increases. debug_defect classified missing in-flight dedupe and listed forbidden fixes.

  Tools the agent called

  debug_defect route_lane fetch_seam
• 13

  Edtech

  Vertex Learning

  35–50 hrs saved

  Empty after save

  Lesson rows persisted with a success toast, but the plan panel stayed empty — the UI gated on a nullable hint while projections 500'd on cold miss. route_lane and fetch_seam rejected warm-all-first (60–90s hangs) and client-only fallback before multi-slice work.

  Tools the agent called

  route_lane fetch_seam fetch_auth architecture_lens
• 14

  Construction

  BridgeWorks

  28–40 hrs saved

  Phantom 409

  First brief save returned version mismatch — timestamp OCC tokens lost microsecond precision through JavaScript Date — and recovery copy sent users to re-import workflows. route_lane, coding_principles, and plan_delivery drove integer revision counters and lane-specific conflict codes in one attested cutover.

  Tools the agent called

  route_lane coding_principles plan_delivery
• 15

  Insurance

  Harbor Insurance

  24–36 hrs saved

  Release transparency

  Underwriters saw pass/fail gates with no artifact transparency; a seven-tile “what we're shipping” strip risked merge without proof. journey_attest returned ready_with_residual (21 nodes, 0 blocking); sentinel_closure signed horizontal seams. Shipped with residuals explicit.

  Tools the agent called

  journey_attest sentinel_closure architecture_lens
• 16

  HR tech

  ClearPath Talent

  45–70 hrs saved

  One assistant brain

  Recruiting, onboarding, learning, and analytics each hardcoded different assistant actions — MCP and an assignment console would have added a fifth brain. map_feature, plan_delivery, and grill_questions locked one resolver, drop-reason codes, and CI guardrails before UI sprawl.

  Tools the agent called

  map_feature plan_delivery grill_questions
• 17

  Media

  StudioPublish

  30–45 hrs saved

  Deletion test wins

  A spec still required a four-verb dropdown while a per-slot rerun strip already shipped the same backend — building the menu would introduce a second client resolver with zero new capability. grill_questions and architecture_lens failed the menu on the deletion test; doc cleanup only.

  Tools the agent called

  grill_questions architecture_lens
• 18

  Edtech

  Vertex Learning

  16–24 hrs saved

  Structure brain first

  A headless cron worker was ready to wire Stage 2 outline generation beside the orchestrator — risk of divergent module plans and mismatched recipes. grill_questions audit_mode CONFIRM GO blocked Stage 2 until a pedagogy resolver landed.

  Tools the agent called

  grill_questions route_lane architecture_lens
• 19

  Regtech

  Atlas Compliance

  50–80 hrs saved

  Targeted test recovery

  Remediation touched 50+ spec files; each guess triggered ~90-minute full CI. journey_attest smoke marked proven vs latent nodes; fix_pack anchored evidence per cluster. Recovery ran targeted bundles instead of shotgun CI.

  Tools the agent called

  journey_attest fix_pack pressure_test
• 20

  Logistics

  FleetDispatch

  20–30 hrs saved

  Product decision menu

  Engineering was ready for org→depot cutover but product hadn't decided catalog scope, scope columns, assistant rail, or agent API safety — and coders started assuming answers. slice_issues, grill_questions, and map_feature produced a numbered decision menu in one agent session.

  Tools the agent called

  slice_issues grill_questions map_feature prism_frame
• Field patterns

  21

  vignettes · fictional apps

  Delivery

  6

  Defect

  6

  Triage

  4

  Assurance

  5

  509–775 hrs saved (illustrative)

  wrong path vs governed agent session

Hour ranges are method · typical wrong path (meetings, wrong fixes, full CI, parallel UI) vs attested path (one governed agent session). A senior engineer + agent pair. Not a guaranteed SLA. Not a compliance claim. Your repository's tests and CI still decide pass/fail.