Smart Mining 2026: The Agentic AI Stack Your FMS Doesn’t Provide

Your mine is fully wired. MineStar or Modular Mining tracks every truck. FMS and condition monitoring stream real-time machine data. But your dispatcher still reroutes haul routes by radio. Your geotech inspects the berm, documents it on WhatsApp, and the finding never reaches your CMMS. Your shift boss spends 45 minutes every shift-change speaking in circles. Knowledge transfer has no structure, no record, no system to act on it.

Your smart-mining stack is excellent at what it does. It just doesn’t do this part.

TL;DR

• Smart mining FMS platforms manage machine data. They do not capture dispatcher decisions, shift handovers, or inspection findings.
• Roughly 60 percent of operational reality at mines never reaches a system of record, flowing instead through radio, WhatsApp, and verbal handover.
• Quarrying and small-scale mining operations lack FMS infrastructure entirely, running on radio and paper.
• Processing plant downtime reasons are classified retrospectively and incorrectly after the shift ends.
• The global autonomous haul truck fleet grew 84 percent in 12 months, from 2,080 units (July 2024) to 3,832 (July 2025), yet the operational layer above the machines stays uncaptured.
• Ungoverned AI is a safety and regulatory liability on mine sites (MSHA, tailings, environmental reporting).

What Smart Mining Actually Means in 2026

Smart mining is the convergence of autonomy, FMS, sensors, and digital twins. OECD data shows mining productivity fell by roughly half from 1997 to 2023 while manufacturing productivity more than doubled in the same period, a gap confirmed by McKinsey analysis (Global Mining Review, December 2025). The smart-mining wave is the response. Every major operator is somewhere on this stack, but it solves one layer: the machine. It does not solve the operational layer where humans make calls, document findings, and handoff shifts.

Autonomy solves truck routing, not the shift log

Autonomous haulage platforms like Caterpillar MineStar and Komatsu Modular Mining Dispatch are exceptional. The global autonomous haul truck fleet grew from 2,080 to 3,832 units between July 2024 and July 2025, an 84 percent jump, with BHP operating 360 trucks and Rio Tinto 305 (GlobalData, 2025). They manage route optimization, collision avoidance, and real-time truck position. A dispatcher can adjust allocation rules by the hour. The system records every truck movement and tire temperature. What the system does not record is why the dispatcher made the adjustment. Was it a blast delay?, and a shovel breakdown? A geotechnical restriction? That logic lives in the dispatcher’s head or a 5-second radio call. The FMS captures the outcome, not the reasoning.

FMS, condition monitoring, drones, and digital twins

A full smart-mining stack includes several components. Hexagon Mining and Maptek handle fleet and asset optimization. ABB Ability and SAP offer condition monitoring and predictive maintenance, and drone networks generate surveying data. Digital twin models simulate pit geometry and equipment performance. Each is built for a specific job: machine telemetry, route planning, equipment diagnostics, geotechnical accuracy.

None of them was designed to capture what a geotech writes in a WhatsApp voice note. None structures shift-handover knowledge into actionable carryover risks. None closes the gap between the operator’s field reality and the system of record.

What each technology solves and what it does not

Smart-mining technologies excel at the machine layer. They answer: Where is the truck? What is its fuel temperature? When is the next preventive maintenance due? How does shovel output compare to planned dig volume?

They do not answer: Why did the dispatcher reroute three loads this shift? Which geotechnical clearances are current for this pit sector? What did the night shift discover that might affect today? Why was OEE below 75 percent yesterday? Recent surface-mining studies place average OEE at 25 percent for shovels and 38 percent for dump trucks, with low utilization the dominant driver (Resource Policy, 2026). The FMS reports the number; it does not explain it.

This is not a weakness in the vendors. It is an architectural boundary. The FMS was designed to manage machines. It was not designed to capture what the supervisor wrote down during shift change.

The Dark Data Gap Your Stack Doesn’t Close

Roughly 60 percent of operational reality at mines flows through radio calls, WhatsApp groups, and shift-handover whiteboards that never reach any system. The FMS sees the truck; it does not see the dispatcher’s rationale. Dark data is information that shaped the decision but never reached the system. Agentic data capture from radio and WhatsApp closes this gap.

Surface mining: blast plan drift and shovel-loader mismatches

A blast sequence shifts by four hours. The dispatcher reroutes haul allocation by radio. The FMS records the new route. It does not record the blast delay or its cascading impact on shift priorities. A shovel-loader matchup misalignment runs for two shifts before anyone reviews actuals versus plan. By then, the loader has burned extra fuel waiting idle. The reason for the mismatch was documented on a whiteboard that got erased at shift-change.

Underground: ventilation decisions and shift-change rituals

A nightshift geotech spots a cracked berm during underground inspection. She reports it via WhatsApp to the shift boss. The shift boss tells equipment operators to avoid the sector. The finding never reaches the CMMS or the ventilation system controls. Equipment movement continues without formal clearance. The next day, a different crew is unaware of the restriction.

Shift-change rituals are verbal handovers between incoming and outgoing crews. Critical information: open work orders, pending repairs, geotechnical restrictions, recent near-misses, and all communicated aloud. When the incoming shift boss is distracted or the outgoing boss leaves early, carryover risks are missed. Multi-channel status capture from field teams turns these verbal rituals into structured records.

Quarrying: operating with no FMS at all

Smaller aggregates and quarrying operations lack FMS infrastructure entirely. Operations run on radio, a spreadsheet, and the site manager’s memory, and dig plans exist on paper. Shovel dispatch is by eye. Equipment maintenance is reactive, run it until it breaks, and production tracking happens manually post-shift. The dark data gap is not reduced; it is total.

An agentic layer here delivers proportionally greater value because there is no digital system at all to start with.

Processing plants: downtime classification after the fact

A recovery rate excursion happens at 02:00. The cause is discussed on the floor and never formally classified. The downtime reason code is filled in retrospectively on the next day shift by someone who was not present. The code is wrong because the memory is incomplete.

Real-time downtime recording prevents this. When the operator flags a halt, the system asks: what happened? The answer is captured while the condition is fresh. Retrospective classification introduces delay and error.

Where Agentic AI Fits the Mining Stack

Smart-mining platforms are excellent, but the question is what layer sits on top. The answer is a three-layer architecture: SCADA/FMS generate machine data, EquipmentOS unifies it with human-channel data, and Agent Builder workflows act on top. This architecture is used across field-driven industrial operations, mining, ports, logistics, manufacturing, wherever unstructured operational data is the blocker.

The three-layer architecture: SCADA, data backbone, Agent Builder

Layer one (bottom) is SCADA, PLCs, OEM telemetry, FMS, and condition monitoring. These generate machine data and handle autonomy. This article is not an argument against them. They are excellent at what they do.

Layer two (middle) is EquipmentOS. It ingests machine telemetry from the FMS layer. It also ingests human-channel data from radio, WhatsApp, email, Teams. It creates a unified event record across equipment activity and human activity. It acts as the integration layer over SAP, Maximo, MainPac, and legacy systems.

Layer three (top) is Agent Builder. Operations users describe what they need in plain language. AI agents build it in a governed staging environment. IT and HSE review and approve before anything reaches production.

The FMS vendor cannot ship a shift-handover summarizer in a quarter. Agent Builder ships it in weeks on the infrastructure the mine already has. For a deeper primer on agentic AI before diving into the stack, see the industrial IT guide to understanding agentic AI.

Opsima as an overlay, not a replacement

This is critical: Opsima does not replace MineStar, Modular Mining, Wenco, or ABB Ability. Each agentic workflow is built on top of your existing FMS and CMMS stack. The agent reads from and writes to Maximo, SAP, or whatever system your site already runs. Opsima is an overlay and enrichment layer, not a competing product to the FMS vendors.

How EquipmentOS bridges sensor data and human data

EquipmentOS is the operational data backbone that merges two data streams. Stream one: machine telemetry from sensors, SCADA, and FMS. Stream two: human data from radio, WhatsApp, email, supervisor notes. By merging them, EquipmentOS gives the mine a unified real-time view. A blast delay is not just a timestamp in the FMS. It is a correlated event across truck movement, dispatcher decisions, and plan variance. An inspection finding is not just a WhatsApp message. It is a tagged asset, a defect category, a severity level, and a work order trigger.

This unified layer enables Agent Builder workflows to act with full context.

Six Agentic Workflows That Extend Your FMS

Each workflow is built on top of your existing FMS and CMMS stack. The agent reads from and writes to Maximo, SAP, or whatever system your site already runs. Opsima is an overlay and enrichment layer, not a competitor to MineStar or Wenco. Here are six workflows that operators have asked to build:

Shift handover summarizer

The outgoing supervisor’s verbal or written handover is captured via WhatsApp, Teams, or email. The agent structures carryover risks and open work orders. It flags items requiring immediate incoming-shift attention. It creates a timestamped record in the CMMS. Result: a structured 5-minute review replaces a 30 to 45-minute verbal ritual. Knowledge loss drops from routine to zero.

Inspection-to-work-order agent

A pit inspector photographs a leaking hydraulic line and sends a voice note on WhatsApp. The agent extracts asset ID, defect category, and severity. It creates a structured work order in Maximo and routes it to the right crew. Result: inspection findings become work orders within minutes, not hours or never.

Plan adherence and blast-drift tracker

The agent compares real dispatcher decisions (captured from radio exchanges and FMS logs) against the blast plan and dig-plan targets, and it surfaces drift in near-real-time. The FMS shows the plan; Agent Builder tracks adherence and flags variance. Result: deviations are visible hours before impact, not days after.

Downtime classification agent

When equipment halts, the operator or supervisor flags the reason in real time via WhatsApp or radio. The agent structures the response: equipment, asset class, downtime category, root cause. It logs it to the CMMS immediately. Result: root-cause analysis is current, not retrospective. Automated OEE and MTTR calculations are accurate from the moment the event happens.

PM compliance agent for mixed fleets

Equipment comes from multiple OEMs with different PM schedules. Caterpillar haul trucks, Komatsu loaders, generic crushers. The agent tracks meter hours and runtime for each unit. It surfaces upcoming PM windows by equipment class. It routes maintenance alerts to the right crew. Result: predictive maintenance works across mixed fleets without manual spreadsheet tracking.

Ventilation and geotech permit agent

Underground operations require permitting for restricted zones. Before any equipment enters, the agent checks permit status against the current ventilation schedule and geotech clearance log. It blocks the movement order if clearance is missing, and it alerts the shift boss. Result: a safety workflow executes in software, not via radio chatter or trust in memory.

Why Mining Demands Tighter AI Governance

An agentic workflow that triggers equipment movement without verifying ventilation permits or geotechnical clearance is a safety event, not a software bug. MSHA citations, environmental monitoring requirements, and tailings management regulations mean every deployed workflow must pass risk assessment before production. This is non-negotiable.

MSHA, environmental reporting, and tailings risk

Mine safety is regulated. MSHA requires documented controls for underground operations. Environmental agencies track tailings management and water discharge. A poorly designed AI workflow that clears equipment movement when ventilation data is stale is a citation waiting to happen. An agentic system that misclassifies downtime and misses recurring failure patterns is a safety incident.

Ungoverned AI is a safety liability

Consumer vibe-coding tools produce ungoverned scripts with no staging environment, no risk assessment, and no audit trail. A script that looks right in a demo and fails during a blast sequence or an underground equipment movement permit check is a liability no mine operator accepts.

Staging-first protects production

The five-agent architecture exists because mine operations are high-risk. Environment Setup connects to SAP, Maximo, and legacy systems. Discovery Agent interviews ops users and generates specs. Execution Agent builds in staging using Claude Code. Risk Assessment Agent checks for vulnerabilities and governance gaps. IT Admin System delivers to IT and HSE for review and approval before production rollout.

This is the governance architecture that makes agentic AI deployable on a mine site, not just in a pilot.

Build vs Buy: The Weeks-Not-Quarters Benchmark

A typical SI engagement for a custom shift-handover capture and structuring workflow: 3 to 6-month scoping, 6 to 12-month build, $30K to $50K per consultant per month. The mine still does not have it in production a year later. This is the status quo you are comparing against.

What an SI quote actually costs

An SI scoping engagement starts with interviews, and then requirements. Then RFP responses, and then contract negotiation. Then team mobilization. At $40K per month, three months of scoping is $120K spent before a single line of code is written. A six-month build is another $240K, and total: $360K for one workflow. And that timeline assumes no scope creep or change orders mid-project.

When SI engagements make sense

SIs make sense for large, one-time migrations, and a new TOS. A full ERP rollout. Infrastructure modernization that spans 18 months and rewires the entire operation.

SIs do not make sense for the 30 to 50 operational workflow improvements ops teams need every year. That is the budget Opsima competes for: not other software, but custom IT development and SI hours. Real-world agentic AI deployments across heavy industry have shipped in weeks, not 6-month projects.

The permanent capability vs consultant

Agent Builder delivers a permanent in-house capability. The next workflow ops needs goes through the same weeks-not-quarters pipeline, and no new SI. No new vendor, and no new 12-month wait. That is the architectural difference: a capability that compounds versus a consultant engagement that ends. Customers pay only when the software delivers value, with the first risk on Opsima rather than the buyer.

Where to Start: Pick the Highest-Friction Workflow

The highest-friction dark-data source at most mines is shift handover. It takes 20 to 45 minutes per shift. Most knowledge transfer is verbal. Carryover risks are routinely missed. Every missed event is a potential unplanned downtime or safety incident. This is the fastest path to a visible win that ops and IT can both see.

Identifying the darkest dark-data source on your site

Walk through a shift-change at your mine. How long does the handover take? Is it documented? Do the incoming crew members have to ask repeated questions? Do carryover risks get missed?

Next, sit with your dispatcher during a shift, and how many reroutes happen? Are they recorded in the FMS or just spoken on radio? Can you trace why a reroute happened?

The darkest data source is the one that repeats every shift and is never recorded. It impacts operations every time it is missed.

The 48-hour bootcamp: proof on your real data

The 48-hour bootcamp is not a demo. It is a working agent running on the mine’s real data, and the actual shift handover format. The actual CMMS, and the actual radio channel feed. The output is a structured shift record and carryover report, not a PowerPoint.

95 percent of enterprise AI pilots never reach production. The bootcamp proves this one does.

Expanding from one workflow to full capability

The VP of Operations or Director of Maintenance identifies the highest-friction workflow, and brings Opsima to IT. IT governs the staging and approval pipeline.

Ops leads demand, and IT approves deployment.

The first workflow surfaces the next five to ten workflow candidates that ops teams have waited years for IT to address. Now they have a 48-hour path to get them built. For more context on how agentic AI applies across field-driven operations, see the mining industry primer.

Your smart-mining stack solves machine management. The operational layer above it, shift handovers, radio chatter, inspection findings, downtime classification, still runs on tribal knowledge. That is where the next 30 percent of operational improvement lives. To stop dark data from bleeding throughput and margin, see how Opsima captures it in weeks, not quarters on your real data.