MES without rip-and-replace means the execution system deploys alongside what your plant already runs: the ERP stays, the PLCs stay, the spreadsheets get ingested, the paper gets digitized. Nothing is torn out to make room. Value starts on one line in weeks, and old tools retire only when they have nothing left to do.
No rip-and-replace is Harmony AI's line, and this post is the honest explanation of what it means, what it does not mean, and why the alternative fails often enough to have earned its own reputation. If you have ever watched a systems cutover freeze a plant for a weekend and haunt it for a quarter, you already know why this matters.
Why do rip-and-replace projects go wrong so often?
Because they concentrate all the risk into a single moment. A cutover project spends months building a replacement in parallel, then picks a weekend to switch the plant's nervous system. Everything must work at once: every interface, every workflow, every operator's new habits. Anything that was missed, and something is always missed, becomes a production problem on Monday morning.
The deeper cost is what gets thrown away. Legacy systems and processes are ugly, but they are ugly the way a well-worn jig is ugly: shaped by years of real problems. The spreadsheet with the weird changeover matrix encodes somebody's hard-won knowledge. The paper form has a checkbox that exists because of an incident nobody wrote up but everyone remembers. Rip it all out and you discard institutional memory you did not know you had, then spend a year rediscovering it as defects and downtime. That is a big part of why traditional MES implementations fail, and why the failure usually surfaces after go-live, not before.
And there is a quieter failure mode: the rip-and-replace that never ships. Multi-quarter replacement projects are fragile against reorganizations, budget cycles, and champion turnover. A project that cannot show value until everything is finished is one executive departure away from being shelved.
What does running alongside actually look like?
It looks like your plant on Tuesday, plus a new layer that sees everything. The ERP keeps doing what it is good at: orders, inventory, cost. Your SCADA or historian keeps collecting what it collects. The PLCs keep running the machines. Harmony AI connects to each of them as-is, reads what they know, and adds the layer none of them provide: digital capture at the point of work, live visibility, downtime intelligence, automated reporting, and plain-English search across all of it. The full connection surface, machines, software, paperwork, and tribal knowledge, is mapped in what an AI-native MES connects to.
The practical consequence is that nothing about your operation has to stop, and no one has to bet the plant on a cutover weekend. Operators keep making product. The only visible change on day one is that a form that used to be paper is now on a tablet at the same workstation, laid out the way the paper was. That familiarity is deliberate; it is the difference between adoption in days and resistance for months.
What stays, what changes, and what retires on its own?
| System | Day one | Over time |
|---|---|---|
| ERP | Stays, connected for orders and items | Stays; it is the commercial record |
| PLCs, SCADA, historians | Stay, become data sources | Stay; the machines still need them |
| QMS / WMS | Stay, connected where useful | Stay, with cleaner data flowing in |
| Paper forms | Digitized in the same structure | Paper retires; the workflow lives on |
| Tracking spreadsheets | Ingested as sources | Retire naturally as the layer covers them |
| Morning report ritual | Still available while trust builds | Replaced by automated reporting |
| Legacy MES, if one exists | Runs in parallel, unthreatened | Retired deliberately; see below |
The pattern in that table is the point: retirement happens by attrition, not decree. A spreadsheet dies when the last person stops opening it, because everything it held now lives somewhere better. That is a far safer end-of-life than a memo announcing its deletion. If you do run a legacy MES today, the same alongside logic applies with one extra chapter at the end, covered honestly in replacing a legacy MES.
Doesn't adding a layer just create another silo?
It would, if the layer were another point tool. A plant that buys a downtime app, a checklist app, and a reporting app has three new data silos with better fonts. The alongside model only works because the layer is one system connected to everything: machine signals, ERP context, digitized capture, and documents land in the same place, so a downtime event, the work order it interrupted, and the quality check that followed are one connected record instead of three exports. That is the architectural difference between an AI-native MES and a stack of floor apps, and it is what makes the layer the system of record for execution rather than one more tab.
How do you deploy alongside in practice?
- Map what exists, in person. Deployment starts with discovery on your floor: which systems hold what, which paper captures what, and which people know what. Harmony AI does this white-glove, with engineers at your machines, because the map has to reflect the plant as it runs, not as the org chart says it runs.
- Digitize one line's capture. The line keeps producing. Its forms go digital in the structure operators already know. This is the first alongside win and it costs the old systems nothing.
- Connect the readers. Machine signals and ERP context flow in through supported interfaces. No existing system is modified beyond granting read access; the timeline stays in weeks because there is no rebuild.
- Run in parallel and compare. For a few weeks, the old ritual and the new layer coexist. The comparison builds trust and produces your before-and-after baseline, the raw material of the ROI case.
- Expand and let attrition work. More lines, more capabilities. Each old tool retires when, and only when, nobody needs it anymore.
Why coexistence is technically routine
- The ANSI/ISA-95 standard exists precisely because enterprise and floor systems are expected to coexist and exchange data across a defined boundary; alongside deployment works with that grain rather than against it.
- OPC UA (OPC Foundation) and gateway support for legacy protocols like Modbus mean most equipment can be read without modifying it, which is what makes connect rather than replace a real engineering option and not a slogan.
- Independent surveys of large software and systems projects have reported elevated failure and overrun rates for big-bang replacements for decades; the honest takeaway is directional rather than a precise percentage, and incremental go-lives exist specifically to avoid joining that sample.
What did alongside look like in a real plant?
At CLS, a specialty decoration and labeling manufacturer in Chattanooga, nothing was ripped out. Paper-based production logging became digital capture at the point of work. The data that had always existed, accurate, thorough, and trapped until end of shift, became visible in real time. Morning reports that took real staff effort to compile became automated output. Decades of documentation stayed exactly where knowledge should stay, except now anyone on the floor can search it in seconds. The plant that existed before Harmony AI still exists. It just stopped being blind between shifts.
If you are weighing this model against a replacement project, run both timelines through the ROI calculator, including the internal hours and the quarters of waiting. The alongside case usually wins before the software features even enter the argument.