Maximo Mobile 9.2 Put an Assistant in the Field That Only Works Online: The Offline Boundary and the Sync Budget

Maximo Mobile 9.2 adds permits, locations, timesheets, crew work orders and an assistant, and IBM's own documentation is explicit that the assistant works only in online mode. The features are real. The boundary is the part worth designing around.

Share

Maximo Mobile 9.2 Put an Assistant in the Field That Only Works Online: The Offline Boundary and the Sync Budget

Field software gets judged in places with no signal. A substation with a steel roof, a basement plant room, a mine pit below the rim. Every mobile programme eventually produces the same complaint: it worked in the yard, it did not work where the job was.

IBM's documentation for Maximo Mobile 9.2 is unusually candid about where that line falls for the newest capability on the list. The AI assistant is now available in the Maximo Mobile app, provided it is configured for Maximo Manage and the user holds the security permissions to reach it. The next sentence in the same bullet is the one that matters: currently, it works only in online mode.

That is one sentence, sitting in a list of twenty-odd improvements, and it is the most operationally significant line in the release notes for anyone running crews. It tells you that the flagship intelligence feature of the 9.2 mobile release is a connected feature, in an application whose entire selling proposition is that it keeps working when the connection does not. Everything else in the release makes the same point from a different direction: permits approved in the field, locations created at the asset, timesheets read from the server, crew labour captured on behalf of a team, meter readings refreshed individually when online. The platform is moving capability outward, toward the technician, while remaining honest that a good deal of it still depends on a round trip.

This article reads the 9.2 mobile documentation as a design brief rather than a feature list. It separates what genuinely works on the device from what needs the server, looks at why the assistant is online only by architecture rather than by oversight, examines the permit application that puts an approval surface in a technician's hands, and works through the sync budget: the download weight that every new app and every image and every doclink adds to a first login and to every subsequent window of coverage.

What 9.2 Actually Put on the Device

The technician-facing list in the 9.2 release notes is the longest in several releases, and it is worth sorting into three groups, because the three groups fail differently in the field.

The first group is record work that used to require a desk. The Locations app lets asset managers, field technicians and supervisors create and update locations from the device, view the location hierarchy to understand relationships in the field, and use the Maps option to navigate to the exact work location. IBM is explicit about one limit: updating the hierarchy by changing parent and child relationships is still done in the role-based application in Manage. The Asset and Location details pages gained a Related records view with a count per record type, an ability to launch the corresponding applications, and a create action for any supported record type associated with that asset or location. Those are small sentences with a large consequence: a technician who finds an unregistered pump house can now create the location where the work happens, rather than noting it for somebody in the office to enter next week.

The second group is money and time, which is where field applications usually earn or lose their reputation. The Material Request app is now a separate app in Maximo Mobile, able to create requests for any work order, show existing requests and carry descriptions of the materials ordered. Actual labour and materials details are available on the work order. The Timesheets app shows labour entries for today or yesterday with work order, asset and location detail, transaction types and total hours, expandable to duration, premium pay codes and task detail. It is read only, and IBM states that it always fetches the latest information from the server, which is a connected dependency hiding inside a feature that looks like a report.

The third group is team execution. Crew work order management lets technicians and crew supervisors view, accept, execute and complete work orders assigned to crews, with a dedicated schedule list separate from individual assignments. Any crew member or supervisor can start or stop travel and work timers on behalf of the crew, which automatically creates labour transactions for all active crew members, and when a crew work order is completed, all active labour transactions close. The whole feature is behind a security option, so it can be enabled or disabled per organisation. That is a sensible gate, because crew-level labour capture is the kind of capability that quietly changes how a timesheet dispute gets resolved.

Around those groups sit the usability changes: a mobile dashboard as the default entry point, which IBM notes uses already downloaded data to minimise additional synchronisation, an enhanced notification panel with deep links that open the related app or action, case-insensitive search, deep links built on business identifiers such as asset number and site rather than internal IDs, and biometric authentication for e-signatures using Face ID, Touch ID or a fingerprint scan with a password fallback when biometrics are unavailable or fail. Meter handling improved too: the most recent reading is displayed regardless of which app it was entered in, meter definitions are downloaded once, only the readings you need are updated, individual readings can be refreshed while online, and location meter readings entered offline sync when connectivity returns.

Then there is the entry that this article opened with. The AI assistant is in the app, and it is online only. Set that against the rest of the list and a pattern appears. The features that change data on the device were designed for disconnection. The feature that reasons about data was not.

Why the Assistant Stops at the Edge of Coverage

The online-only limit is not arbitrary, and IBM's assistant architecture documentation explains why in about eight steps. The agentic Maximo Assistant is supported by a supervisory AI agent called the ALM agent, a set of AI tools, and the Maximo MCP server, all coordinated through AI Service. When a user submits a prompt in the chat window, it is passed to the ALM agent, which runs as the alm-agent pod in OpenShift. The ALM agent calls AI Service, which sends the request to the language model in watsonx.ai, with a prompt that includes schema metadata, business rules and entity relationships relevant to the request. The ALM agent then decides which tools are needed, calls those tools on the Maximo MCP server using JSON-RPC 2.0, and the MCP server routes each call to the appropriate Maximo Manage API. Results flow back through the server to the agent, and the agent produces the natural language answer.

Read that path against a phone in a basement. The reasoning happens in watsonx.ai. The tool execution routes through the MCP server to Manage. The only thing on the device is the chat window and the request. There is no local model in the documented architecture, no cached schema the app can reason over, and no tool broker on the handset. An offline assistant would have to be a different product with a different inference story, and IBM has not shipped that. The non-agentic assistant is not an alternative: it uses a single data retrieval tool that translates a natural language query into OSLC calls against Manage object structures, which is still a round trip.

Two details in the architecture documentation are worth carrying into field planning. The first is that the assistant operates within the same security, permission and approval rules that govern Manage and the suite, and that two users asking the same question can receive different answers depending on their security group permissions and access rights. That is the correct behaviour, and it means a crew supervisor and a technician standing at the same asset can legitimately get different answers to the same question. Training that assumes a single house answer will confuse people.

The second is the tool list. IBM documents data retrieval, condition insights, documentation search and schema retrieval, and notes that no tools or their related AI features require training, and that activating an AI configuration enables the tool and its feature. Documentation search is enabled by default, and IBM adds a caveat that responses using it typically contain links to IBM Documentation articles, and those links do not work in on-premises or air-gapped environments. That is a small trap with a specific symptom: an assistant that answers confidently with citations nobody can open.

The practical conclusion is not that the assistant is disappointing. It is that the assistant is a connected tool in a disconnected job, and the workflow has to be designed around that asymmetry. The technician who will spend the afternoon underground should ask the assistant before going in: show me the history on this asset, what the last three work orders found, what the torque spec is. The answers can be read on the device afterwards, because they arrived as text in a conversation. What cannot happen is asking at the asset, mid-job, with the panel open and no bars.

There is a second use for that asymmetry that teams tend to miss. Because the assistant is available only where the network is, it is naturally a planning and end-of-day tool rather than an execution tool. A supervisor can use it while dispatching, and a technician can use it in the van before the first job and at the end of the shift when the notes are being written. Designing the daily rhythm around those two windows costs nothing and removes most of the disappointment that comes from promising a field assistant that cannot answer in the field.

Permits Are Now Signed Where the Work Happens

The single most consequential addition in 9.2 is easy to miss because it is listed as one bullet among many. The Permit request mobile application allows work permit records to be created and updated in Maximo Mobile, and those records are integrated with IBM Maximo Health, Safety and Environment and IBM Maximo for Oil and Gas for real-time approval of workflows and documentation for high-risk activities. The feature-channel notes phrase the same capability as a Permit to work application that integrates with existing HSE and Oil and Gas business processes and workflows, ensuring that mobile-created permits align with safety processes, approval, validation and control mechanisms.

Think about what a permit is as an artefact. It is a record that asserts that a specific high-risk task was authorised by a specific person at a specific time, under specific conditions, with evidence of isolation, gas testing, or whatever the local safety case demands. It is the document that a regulator or an investigator reads after an incident. Moving its creation to the field is a genuine productivity gain, because the person who knows the conditions is the person at the asset, and it removes a transcription step that has always been a source of error.

Moving it to the field also collides with the offline boundary in a way the documentation does not resolve. The bullet describes integration for real-time approval, and real-time approval implies a connection. The documentation does not state which parts of the permit lifecycle work offline, whether a permit created on a device without coverage can be approved before it syncs, or what the record looks like if a sync fails after a permit was saved locally. Maximo Mobile has a well established offline pattern for work orders and inspections, and a permit is a different kind of object: it gates the start of work rather than recording work done.

That gap is not a criticism of the product. It is a gap in the documentation that a rollout team has to close by testing, because the answer determines a safety procedure. If a permit can be created and saved offline, and approved only after sync, then a crew in a dead zone is waiting for coverage before starting a high-risk task, which is unacceptable and will be worked around. If the approval path requires an approver who is also online, then the procedure needs a named fallback that the safety case already recognises, whether that is a paper permit, a radio authorisation with a recorded time, or a second device with connectivity.

The design questions to put in front of a safety lead are short and they are not technical. Who is allowed to approve a permit that was created on a device with no coverage when the approver is remote? Does the approval timestamp reflect the moment of authorisation or the moment of sync? If a technician creates a permit, loses the device, and the permit never synced, what does the audit trail say? Which fields of the permit are validated on the device and which are validated against the server, so that a permit which looks complete in the field is rejected later?

There is also a smaller, more immediate question about who the permit app is for. A permit to work is usually the interest of a supervisor or a permit coordinator rather than the person turning the wrench, and the mobile app ships under the technician-facing list. Teams should check which security groups see the app before they announce it, because a permit capability that appears for people who should not be creating permits is a governance problem dressed as a feature.

The right posture is neither to disable the app nor to assume it behaves like an inspection form. Test it in a real dead zone, write down what happens at each step of the permit lifecycle, and make the fallback explicit in the procedure. The value of a mobile permit is that it records what actually happened at the asset. That value survives only if the procedure around it is honest about where the network ends.

The Sync Budget Got Bigger and the Controls Got Finer

Every feature added to a field application is weight. The first login has to download configuration, applications, work lists, supporting data, job plans, safety plans, inspection forms and, increasingly, images and documents. Subsequent syncs carry changes in both directions. The 9.2 documentation reads like a release that finally admitted this, because most of the administrator-facing changes are about controlling weight rather than adding capability.

Start with doclinks. IBM documents that Maximo Mobile and Maximo Manage now use separate relationships for mobile apps and browser access, with MOBILE_DOCLINKS carrying a simplified query that downloads data faster to devices while MANAGE_DOCLINKS handles browser access, and each can be customised independently without affecting the other. That is a precise admission that mobile was paying for browser-shaped queries. It is also a maintenance obligation: two relationships to keep in step, and a customisation to a browser relationship that will no longer propagate to devices by accident.

Images are the second front. Record image downloads can now be controlled per application in the mobile offline data settings and container settings, and globally through Maximo Manage mobile properties, so that image weight can be managed at the level of the whole deployment or a single app. Image upload compression is configurable, with an enable or disable switch, a default upload quality applied when users attach images, and an option allowing users to override and upload full resolution when the job requires it. A maximum number of attachments per upload action can be configured, because Maximo Mobile previously enforced file size limits only, which allowed bulk uploads on constrained networks that ended in failures or memory pressure. All of that is the platform saying, in configuration, what mobile teams have been saying in meetings for years: an eight megabyte photo of a corroded flange is not free.

Login and background load are the third front. Parallel data downloads let configuration data, system information and images arrive at the same time, and module update checks are consolidated into a single request. Background fetch allows the app to download data automatically, scheduled for specific user groups so that data is ready before a shift starts. Preloaded databases can be configured in the Mobile Configuration application, with status visible and the option to disable or delete them. A running generation can be stopped, and IBM states that a stopped generation cannot be resumed, with a new one starting on the schedule defined in the mobiledbCrontask cron task. A property called maximo.mobile.ignore.multipurpose excludes named business objects from work list data generation during supporting data preload, for objects that serve as both supporting data and work list data.

Then the controls that live in the hands of the technician and the administrator mid-shift. Refresh data options changed to Reload all data, Update recent changes, or Do not download. Individual records such as work orders and assets can be refreshed one at a time, and a Sync all records option synchronises every record associated with the current app in one action, so nobody has to visit the cloud sync page or update app by app. An active record synchronisation can be stopped from an inline notification on pages that use card groups, with downloaded pages remaining available and the sync restartable, and IBM notes that stop-sync support in this release applies only to card group components, with the underlying API available to application teams for other areas. The app now persistently indicates when it is disconnected, covering no internet, a VPN problem or a server outage, and saves offline work for automatic synchronisation on reconnect.

Put the controls next to the features and the arithmetic becomes visible. A rollout that adds the Locations app, the Material Request app, the Permit app, crew work orders and Timesheets is adding records to the sync set, and every one of those records competes with images and doclinks for the same window of coverage. The controls exist because the right balance is per role, not per deployment: an inspector needs forms and images, a storeroom clerk needs item and bin data, a crew supervisor needs the crew's assignments and labour state. The 9.2 tooling finally allows that balance to be expressed precisely, which means the failure mode shifts from the platform to the configuration. A sync set that is too broad produces slow logins and full devices. One that is too narrow produces technicians who cannot see the asset history they need. Neither problem announces itself as a configuration error.

What to Configure Before the Trucks Roll

The administrator surface in 9.2 is broad enough that a rollout can spend weeks in settings and still miss something that matters. The order below is not a substitution for IBM's documentation, but it is a reasonable sequence for a first pass.

Start with identity, because everything else depends on it. Configure a default authentication method for Maximo Mobile from local authentication, LDAP, OIDC or SAML, and note that a specific method can be selected directly from the URL. Test the chosen flow on the actual device models in the fleet, not on a desktop browser, because identity redirects on mobile handsets are where first-day support tickets come from. Then check the security groups against the app list. The SELFSERVICE security group has new access permissions for Service Request mobile users, crew work orders sit behind a security option, and the permit application needs a security review before it appears for anyone.

Set the data scope before the first login, not after the first complaint. Configure preloaded databases for the roles that need a head start, schedule the mobiledbCrontask generation sensibly, and treat a stopped generation as a deliberate act rather than a pause, because it cannot be resumed. Configure background fetch for shift patterns so the download happens on a charger rather than in a van. Decide whether record images download at all for each app, set the upload compression default, and decide whether technicians may override it for high resolution.

Then verify like a technician rather than an administrator. Log in with a production identity on a production device and confirm which apps appear. Pull a work order down, put the device in flight mode, complete the work, capture a photo, record labour, and reconnect. Open the assistant while offline and record what it says, so that the support desk knows the expected answer rather than inventing one. Create a permit in a dead zone and follow it through to approval, writing down each state the record passes through. Check that the offline indicator appears and that it clears, and check the notification deep links on a device that has never seen the record, where IBM's documented behaviour is a server-side search and automatic download if the device is online.

Finally, capture the baseline. If the deployment uses the agentic assistant, the agent's restart log lists which tools survived its filter, and that count is worth recording for the same reason you record a firewall ruleset. It tells you what the assistant can actually do, as opposed to what the integration diagram says it can do.

Practical Implications

For field service leads, the 9.2 release is a genuine step forward that should be planned around one hard boundary. Everything that records work has become better on the device. Everything that reasons about work still needs the server. Build the daily rhythm around that: assistant in the van and at the end of the shift, execution on the device in between. Promise the crew the second and mention the first as a bonus, and the rollout will not inherit the disappointment that a connected assistant in a disconnected job usually creates.

For mobile administrators, the centre of gravity has moved from deployment to data scope. The Mobile Configuration application, preloaded databases, background fetch, per-app image controls, doclink relationships and the refresh options are now the levers that decide whether the app feels fast or slow. The most useful artifact you can produce is a per-role data contract: which apps, which records, which attachments, which refresh behaviour. Without it, every performance complaint becomes a debate about the network.

For safety and compliance owners, the permit application deserves a review before it reaches a crew. The integration with HSE and Oil and Gas is the right architecture, and the offline behaviour of the permit lifecycle is not something to assume from the work order behaviour. Test it, document the fallback, and make sure the people who can create permits are the people who should be creating them.

Bottom Line

Maximo Mobile 9.2 is a strong field release with one honest limitation printed in the release notes. The assistant works only in online mode, because the reasoning happens in watsonx.ai and the tools route through the MCP server to Manage, so the handset is a window rather than an engine. Everything else moved in the other direction: locations, related records, permits, crew work orders, timesheets and meter readings all became more capable on the device, and the administrator controls for images, doclinks, preload and refresh became precise enough to express a different data scope for every role.

That combination asks for a specific kind of planning. Treat the connected features as connected, and design the shift around when the technician is online. Treat the offline features as a data problem, because that is what they now are. And test the permit path in a dead zone before a crew does, because a permit is the one mobile record whose failure mode is measured in safety rather than in rework.