Skip to content
PROOF // Industrial & MRO Distribution Multi-warehouse industrial & MRO parts distributor · 30,000+ active SKUs · thin single-digit margins
← All Proof

CHAPTER 03 · BUILD

Agents Shipped Inside the Existing Stack

The first system, and the one right behind it, shipped inside the existing ERP — no new platform, no rip-and-replace. Each one is real and running.


THE AGENTS · 3 FOR THIS WORKFLOW

This network needed three. The count isn't a package — it's however many distinct jobs the work splits into. Each agent is a separate identity with its own scoped tools and permissions.

01

Demand-Driven Reorder Recommendation

Status Runs every morning
Impact Manual spreadsheet review → reviewed recommendation
Team Automated, 1 human review
02

Supplier Risk Monitoring

Status Sanctions & delay signals parsed
Impact Manual news scanning → click-through review
Team Automated, 1 human review
03

Warehouse Slotting & Pick-Path Optimization

Status Wave 2 — in build
Impact Pick-path efficiency
Team Automated, plus a solver

IT DOESN'T STOP AT A REPORT

The Agent Takes the Action

This one barely looks like a report at all — it runs the reorder. Most of what these agents do is take the action inside the ERP, with a buyer approving anything that spends money.

agent inbox — this morning
DATA VIEW

Run the demand plan

Re-forecasts every SKU against real orders each morning — nobody rebuilds a spreadsheet.

Ran automatically

Create the purchase order

When a SKU crosses its live reorder point, the agent drafts the PO — supplier, quantity, terms — for the buyer to approve.

Send the PO to the supplier

On approval it goes out through the existing EDI or email path automatically.

Flag the dead stock

Opens a markdown-or-return recommendation on slow movers before the cash is stranded.

Ran automatically

PRODUCTION TRACE

Demand-Driven Reorder Recommendation

Every step below runs against the plant's real data — the exact sequence, in order.

live.internal — this morning
DATA VIEW

142

Reorder recommendations generated

121

Recommendations accepted

5

Supplier risk flags reviewed

8

Slow-mover alerts carried forward

6 items need your judgment

B-201 SR-063 lead time extended 40% — reorder quantity needs your call
High
B-202 SR-204 sanctions name-match resolved as false positive
Medium
B-203 Slow-mover review: 14 SKUs flagged for markdown
Medium

Ingest

Automated

Live sales and inventory data pulled from the ERP every morning across all three DCs — no new sensors, no new data entry.

Fan out by DC

Automated

One data pull becomes three reorder recommendation sets — one per distribution center, each grounded in that DC's own demand pattern.

Ground every claim

Automated

Every recommended reorder cites the exact demand trend or supplier lead time behind it, not a general opinion.

Buyer review

Reviewer input

Each recommended order is one-click accept or reject. Rejections become training signal for tomorrow's recommendation.

Tomorrow's diff

Automated

The next morning's list opens with a diff against today's accepted orders — did it work, what changed.

REAL OUTPUT · SUPPLIER RISK MONITORING

One Example, Fully Processed

When a job calls for a written artifact, this is what the system produces — everything it finds, structured for the action it triggers next.

Supplier Risk Scan — Bearings & Fasteners Category

30 monitored suppliers · scanned against OFAC, Federal Register, and delivery-performance data · updated hourly

DATA VIEW

Supplier Risk Signals Detected

Supplier Signal Source Detail Confidence
Supplier SR-118 Delivery delay pattern Historical EDI 3 consecutive late shipments on the bearings line 93%
Supplier SR-204 Sanctions list name match OFAC Partial name match — resolved as false positive on manual review 71%
Supplier SR-092 Price volatility Federal Register + FRED Raw-material tariff change flagged upstream 88%
Supplier SR-140 Financial distress signal Public filing scan Late-payment pattern on trade references 82%
Supplier SR-063 Capacity constraint Delivery performance Lead time extended 40% over trailing quarter 90%

Supplier Risk Summary — for Procurement

Five suppliers carry a flagged signal this week. One is a resolved false positive (name-match only); the other four warrant a closer look before the next PO cycle — particularly SR-063, where lead time has grown 40% in a quarter with no communicated cause.

Suppliers to Watch This Week

SR-063: lead time trend needs a call before the next order — don't let it become a stockout.
SR-204: sanctions match was a false positive, but log the resolution for the audit trail.
SR-140: financial distress signal is early-stage — worth a payment-terms conversation now, not after a missed shipment.

WAVE 2 — IN BUILD

Warehouse Slotting & Pick-Path Optimization — What's Next

High-velocity SKUs are slotted far from the packing stations in two of the three DCs — a layout that made sense two inventory cycles ago and hasn't been revisited since.

The agent proposes a re-slotting plan; a pick-path solver checks the math against real order-mix data. Nothing ships on an LLM's arithmetic — the solver's answer is the answer, the agent just explains it.

Worked example the solver evaluates today in testing

DC2's top 50 SKUs by pick frequency are compared against their current slot distance from packing. Four re-slotting approaches scored by projected daily pick-time savings:

DATA VIEW
4 No change — current slotting 0 min/day saved
3 Re-slot top 10 SKUs only +38 min/day saved
2 Re-slot by velocity tier (ABC analysis) +71 min/day saved
1 Re-slot by velocity tier + co-pick affinity +104 min/day saved

WHAT CHANGED

Time returned

~85% of manual reorder review time

Buyers got the time back for supplier negotiation and exception handling, not for a longer spreadsheet.

Operating complexity

A once-set-and-forgotten min/max rule became a reviewed, one-click-accept recommendation

Where this stands

First system live in 30 days

CHAPTER 03 OF 5

UP NEXT · TEAM VIEW

See what actually changed for the people doing the work, role by role.

Continue to Team View