Industries / Industrial equipment and services / Industrial services
CMMS and asset management software
CMMS and asset management software, meaning computerized maintenance management systems and the larger enterprise asset management (EAM) systems, keeps the record of every machine an organization maintains and of the work done on it: what each item is, which jobs are due and done, which spare parts it uses and how it has failed before.
1963
The US Navy began its Maintenance and Material Management (3-M) System. The Chief of Naval Operations set up a project group in January 1963, and the program was put in place "In March of 1963". It gave ships standard planned-maintenance schedules and a standard form for every maintenance action. From 1 August 1964 a Maintenance Support Office in Mechanicsburg, Pennsylvania, served as "the central data processing facility for the system", by Harvey Oelkers's 1968 master's thesis for George Washington University.
34
Milestones achieved across 7 domains.
9
To watch for: announced, not yet confirmed outside.
14
Open deliverables. Nobody has shipped them.
In one minute
Where it started
Maintenance records began as paper cards and forms. In 1963 the US Navy's 3-M System gave each ship printed cards for every planned task, quarterly schedules and one form per job, and from 1964 the reports were processed centrally. Software for single companies followed: Maximo 2.0, released in 1985, handled work orders and inventory on one IBM PC AT, by IBM's account.
Where it stands
Most systems now run in a web browser and on phones. The weak point is the record itself. The NTSB found that the database of Pacific Gas and Electric listed the pipe that burst at San Bruno in 2010 as made without a welded seam when it had one, and GAO found F-35 squadrons tracking data by hand because the logistics system's parts records were wrong or missing. Since 2025, vendors have added AI that reads nameplates, drafts maintenance plans from manuals, suggests failure codes and answers questions about the data. None of these features had a published accuracy figure in the sources found for this file.
What is open
Most open jobs are outside tests and shared data. The AI features now shipping in these products, from nameplate reading to reporting chatbots, have no published outside measurement, since the public tests that exist score research systems, and the field lacks open sets of work orders, parts descriptions and plant sensor names because companies keep their records private. Other open jobs connect sources the software still cannot read: the machine maker's digital description of a machine, and the contractor's documents handed over with a new building.
Domains
AchievedWatch forOpen
Domain
Progress
Last achieved
Watch for
Open
Asset records
Asset records are the list of every piece of equipment an organization maintains, with its make, model, serial number, location, place among the systems it belongs to, and the documents that describe it.
4 · 1 · 2
Last achievedA code linked to data2022
Watch forNameplates read by phone
2
openWork orders
A work order is the record of one maintenance job: what was asked for, who did it, what they found, what they replaced and how long it took.
4 · 2 · 2
Last achievedAI picks failure codes2023
Watch forFailure codes filled by AI
2
openPreventive maintenance
Preventive maintenance is work done on a schedule, by calendar or by hours of use, before equipment fails; the software holds each task, its interval and its due date.
4 · 1 · 2
Last achievedElectrical maintenance becomes required2023
Watch forPlans drafted from manuals
2
openSpare parts
Spare parts work keeps the right replacement parts in stock: naming and numbering each part, deciding how many to hold, and tracking where each one goes.
6 · 1 · 2
Last achievedPrinted spares at scale2025
Watch forAI finds duplicate parts
2
openUse on mobile
Mobile use puts the record in the technician's hands at the machine: scanning tags, reading instructions, recording work and getting help from far away.
5 · 1 · 2
Last achievedPhone guidance tested in noise2026
Watch forWork requests by voice
2
openAnalysis
Analysis turns the maintenance record into decisions: which machines fail, how often and why, and whether the maintenance program is working.
5 · 1 · 2
Last achievedA public test for agents2026
Watch forReports asked for in words
2
openConnecting to machines
Connecting to machines feeds the record from the equipment itself, with readings from sensors and controllers and the maker's own digital description of the machine, so work can be started by condition rather than by calendar.
6 · 2 · 2
Last achievedModels tested on failure signs2025
Watch forSensors catch a failing pump
2
open
Advances to build with
2 research notes

Claude Fable 5.2
Three working devices
Anthropic says its AI model Claude Fable 5.2 designed three prototype devices that worked without electrical modifications, with no outside test reported.

Cursor Agent SDK
Findings that cite notes
Cursor released a kit for building AI agents, and its example, a code reviewer that cites the note behind each finding, is known only from Cursor's account.
Open in cmms and asset management software
3 of 14 open deliverables
Open · Asset records
A public nameplate test
A public set of at least 1,000 photos of equipment nameplates (motors, pumps, compressors, boilers, air handlers, transformers), each labelled by two people with make, model, serial number, rating and date. Score each nameplate feature and general AI model on how often it reads each field correctly.
Built with what exists
phone cameras; Label Studio, an open-source labelling tool; the nameplate features of shipping CMMS products; general multimodal models such as Gemini and Claude, asked for each field; Gemini Decision 1, which gives probabilities for fixed-answer questions, to flag uncertain readings (see the Note "A thousand questions, one pass").
Open · Work orders
Score the failure codes
An open test of AI failure coding. Take at least 1,000 real work orders, have two reliability engineers each code them separately with ISO 14224 failure mode, mechanism and cause, then have several AI systems code them. Publish the AI's agreement with the engineers beside the engineers' agreement with each other.
Built with what exists
the MaintIE set of maintenance texts, released under the MIT license; the ISO 14224 code lists; Gemini Decision 1 for code probabilities (see the Note "A thousand questions, one pass"); a small model tuned on one graphics chip, such as Liquid LFM3 (see the Note "One GPU to fine-tune"); Cohen's kappa, the usual measure of agreement between two coders.
Open · Preventive maintenance
Test plans drafted from manuals
A public comparison over 30 manufacturers' manuals for common equipment. For each manual, two experienced planners write a preventive maintenance plan, each AI tool drafts one, and every task is scored: missed, wrong interval, or not in the manual at all.
Built with what exists
manuals that makers publish online; the plan builders in shipping CMMS products; Claude Fable 5.2 (see the Note "Three working devices") or Gemini, told to cite a page for every task; a shared scoring sheet.
Where it sits
Broader