01 — The seat
Autonomy,
built from the desk.
This is the view from my chair. One screen, one AI agent, and the whole chain on the other side of the glass: the parts we engineer, the suppliers we buy from, the floor that builds, and the network that delivers.
Scroll to lean in · move to look around
02 — The agent
Intent in.
Motion out.
I describe the outcome. The agent works out which systems, suppliers and machines it touches, simulates the change, and only then commits it.
- Plan natural-language goals into task graphs across PLM, ERP and MES
- Simulate parts, cells and supply plans in their digital twins
- Verify safety envelopes, tolerances, costs and dates automatically
- Deploy to PLCs, robot controllers, CNCs and supplier portals
03 — Engineering
Parts that
design themselves.
Requirements come straight out of the customer's package. Geometry grows from the load case, gets checked against the machines we actually own, and is programmed and inspection-planned before anyone opens a CAD seat.
- Intake drawings, specs and emails into structured, traced requirements
- Generate topology-optimised geometry under real load cases
- Check DFM against our machines, tool library and process capability
- Program CAM toolpaths simulated on the target machine
- Inspect FAI plan and CMM program generated from the model's PMI
04 — The floor
Robots where they earn their place.
Conveyors carry the parts. Arms only do what a belt can't: vision-guided bin picking, guarded welding and end-of-line palletizing. 5-axis CNC cells along the wall receive toolpaths minutes after a design locks.
- Dispatch work orders released to cells as material and tools are ready
- Instruct station instructions generated for the exact revision being built
- Trace every lot, machine, program and operator linked as the part is made
- Control SPC on live gauge data, with offsets corrected before drift becomes scrap
05 — Intralogistics
Floors that drive.
Ceilings that fly.
Autonomous mobile robots route material in real time while drones count inventory overhead, so the stock the planners see is the stock on the racks.
06 — The backbone
Nothing runs itself
without the plumbing.
Every mission reads and writes real systems, from sensors (L0) to ERP (L4). Each cell publishes to one namespace, every high-risk change waits for a named person, and the agent only reaches the zones it's allowed into.
- Connect OPC UA and MQTT into a unified namespace, with one master record per part
- Retrofit clamp-on sensors and edge gateways for machines older than the network
- Govern sign-off by risk, a full audit trail, and a rollback point for every change
- Defend IEC 62443 zones and conduits, least-privilege agent credentials, OT anomaly detection
07 — The network
The supply chain
is the factory too.
Most lead time isn't spent on the floor. It's spent waiting on quotes, confirmations, certificates and containers. The agent watches every supplier and lane, and acts before a delay reaches the dock.
- Source RFQs out, quotes normalised, award recommended with trade-offs
- Confirm POs acknowledged and promise dates written back to ERP
- Sense port, carrier and capacity signals on every lane
- Reroute qualified alternates switched in before a stock-out
08 — The business
The rest of the company
runs on it too.
A part isn't done when it ships. It's done when the cash is in, the books are closed, and nobody got hurt making it.
- Order POs read from any format, promise dates from live capacity, S&OP scenarios ready
- Close cash applied daily, books closed in two days, variances explained
- Protect near-misses reported by voice, LOTO and permits tracked to closure
- Measure energy and Scope 1–3 carbon per part, loads moved off the peak
- Serve RMAs and warranty claims traced to the lot and fed to quality
- People rosters built on skills, and a plan for every role the agent changes
09 — Automate the manual first
The first things to automate aren't robots.
They're the emails, spreadsheets and re-keying between systems. Here is every manual hand-off across the organisation, from the customer's PO to the closed books, scored by the hours it eats, the lead time it adds and the errors it causes, against how hard it is to automate.
Baseline: a representative 200-person discrete manufacturer with about 40 active suppliers. The numbers are typical ranges, not a measurement of any one plant. Measured data from a real plant would replace them, and the ranking would move with it.
10 — Asset economics
Run the floor like a portfolio.
Every machine on the floor has a failure curve, a book value, parts that are going out of production, and a cost of keeping it one more year. Change the volume, the cost of capital or the budget and see how capacity, quality and reliability move, what to maintain, what to overhaul and replace, and what the capital plan returns.
The register mirrors the floor in the scene: three lines, each with a bin-picking arm, a two-arm weld cell, a vision QC gate, a case packer and a palletizer, plus the machine shop and the intralogistics fleet. Costs, failure rates and lifecycle dates are an illustrative baseline. With a real plant's fixed-asset register, CMMS history and sensor data, every number would follow.
11 — The people
Every role trained before the machines arrive.
The arms, AMRs, printers and the agent only pay off if the people around them know how to run them, fix them, check them and stop them. Each role climbs the same four levels, rehearses in the digital twin first, and is signed off on the real equipment.
Staying qualified
Hours are typical for a plant with this equipment mix and include the shared safety core. Standards named are the usual US references; OEM courses and the plant's own skills matrix set the final curriculum.
12 — Say hello
Thinking about the same problems?
This desk is a working idea of how a whole manufacturer could run from one screen. If you have a question, see something that's wrong, or are working on the same thing, I'd like to hear from you.
Get in touch