There is no doubt that asset management will look very different in the next one to four years, whether we want it or not.
A good infrastructure decision draws on a lot of information: population growth, demographic trends, council meetings, community plans, the asset register, condition assessments, capital projects, current and historical work orders, and much more. All of it matters. Connecting it today takes a tremendous amount of time, yet it is essential for good asset stewardship.
I think we will see AI agents doing much of that connecting, and helping us make better decisions with it. This post covers what that looks like in practice today, the piece of technology that makes it possible, and the part we have not solved yet: how much to trust an agent to act on its own.
The short answer: The most useful thing AI does in asset management today is connect information that already exists, across your asset system and outside it, and turn it into a first draft: an answer, a ranked list, a report. People still check it and make the call.
Here is what this post covers:
- Practical ways to use AI in asset management today
- What MCP (Model Context Protocol) is and why it matters
- A worked example: population growth meets the renewal plan
- How much autonomy to give an AI agent, and the controls for each level
- A safe way to start, one question at a time
How can AI be used in asset management?
Forget the predictions for a moment. These are things asset managers can do now with an AI assistant connected to their asset data:
- Ask the portfolio questions. "Which assets in Poor or Critical condition have had more than three repairs in five years?" is one sentence instead of three reports and a pivot table.
- Draft the documents. First drafts of asset management plan sections, state of infrastructure summaries and council reports, built from your live data rather than last year's spreadsheet.
- Rank renewals. Combine condition, risk, level of service and the replacement plan into a shortlist with the reasoning written out.
- Find patterns in work orders. Repeat failures, assets that cost more to keep than to replace, and repairs that cluster by season or area.
- Clean up the data. Find assets with no install date, no replacement value or a condition score that has not been updated in years. See our guide to asset data quality.
- Research requirements. One customer had an assistant find the city and provincial requirements for his systems and set them up as compliance records, after he approved the list.
Why is connecting the data the hard part?
Because the information behind a single decision lives in many places, and only some of them are in your asset system.
Inside your asset system
- The asset register: what you own, where it is, when it went in
- Condition assessments and risk scores
- Current and historical work orders, and what they cost
- Capital projects and the long-term replacement forecast
Outside it
- Population growth and demographic trends
- Official community plans and development applications
- Council meeting minutes, motions and strategic priorities
- Funding programs and their reporting requirements
Today a person carries information between these by hand: an export here, a PDF there, a spreadsheet that ties them together for one council meeting and is out of date by the next. It is slow, so it gets done less often than it should. An AI assistant can read across all of them in one conversation, if it can reach them.
What is MCP, and why does it matter for asset management?
MCP, the Model Context Protocol, is an open standard that lets an AI assistant connect to another system, see what it can do there, and use it. When I first heard about it, I did not realize how powerful it could be when connected to the right context.
A general-purpose assistant knows a lot about asset management in general and nothing about your watermains. Connected over MCP, the same assistant can read your asset register, condition scores, work orders and capital plan, and, if you allow it, create and update records. The rest of the context, such as a community plan or council minutes, you can hand it as a document or let it search for.
Because MCP is an open standard, you are not tied to one AI vendor. Claude, ChatGPT, Microsoft Copilot and other MCP clients connect the same way. For the technical detail, see how MCP and the REST API work in a CMMS and our developer page.
Worked example: population growth meets the renewal plan
Here is the kind of question that takes days to answer by hand, because it crosses a planning document, the asset register, the work order history and the capital plan.
- 1Reads the community plan you attached
- 2Pulls watermains and roads in the north end, with condition and install year
- 3Counts repair work orders on each over 10 years
- 4Checks the replacement plan and level of service targets
- 5Proposes a shortlist with its reasons, and waits
| Segment | Condition | Repairs (10 yr) | Planned | Suggested |
|---|---|---|---|---|
| Watermain, Ridge Rd | 38 Critical | 6 | 2033 | 2028 |
| Watermain, North Park Dr | 47 Poor | 3 | 2031 | 2029 |
| Road, Ridge Rd | 52 Poor | 2 | 2030 | 2028, with the main |
Notice what the assistant did not do. It did not move anything in the capital plan. It showed its reasoning and stopped, because bringing a project forward is a decision with a budget, a council and a community behind it. That is the right shape for this kind of work today: the agent does the gathering and the first draft, and a person decides.
Connecting the data by hand vs with an AI assistant
| Attribute | By hand | With an assistant over MCP |
|---|---|---|
| Gathering the data | Exports from several systems, then a spreadsheet | Read live from the asset system in the conversation |
| Outside context | Someone reads the community plan and remembers it | The plan is read alongside the asset data |
| How often it happens | Once a year, for the budget | Whenever the question comes up |
| First draft of the report | Days of formatting | Minutes, then reviewed |
| Trail | Spreadsheet versions in someone’s inbox | Every call recorded in the audit log |
| Who decides | People | Still people |
The same decision, before and after the information can be connected in one conversation.
How much autonomy should an AI agent have?
This is the part I do not think we have fully solved. AI agents can be incredibly helpful, but also dangerous when given too much autonomy. An agent that can create a work order can create a hundred of them. An agent that can update a condition score can quietly change the numbers a capital plan is built on.
The technology is mostly here. What we are still figuring out is the governance and controls around it. A useful way to think about it is in levels, each with its own control:
| Level | What the agent does | Control | When |
|---|---|---|---|
| Ask | Reads data and answers questions | Read-only key | Start here |
| Draft | Prepares records, reports and changes for review | A person approves every write | Ready today |
| Act within limits | Writes routine records on its own | Write scopes for named resources only | Once it has earned it |
| Autonomous | Decides and acts across systems unattended | Not solved yet | Not for capital decisions |
Whatever level you choose, a few controls should be non-negotiable:
- Its own credential. The agent connects with a key made for it, not a person's login, so you can see what it did and switch it off.
- Least privilege. Read-only first. Add write permissions one resource at a time, and only where the agent has earned them.
- Approval before writes. Ask the agent to propose changes and wait for a yes, as in the example above.
- An audit log. Every call recorded under the key that made it.
- Check the numbers. AI summarizes confidently, including when the data underneath is incomplete. Anything that goes to council gets checked against the source.
- Know where the data goes. Your organization's AI policy, and your vendor's list of where each service runs, should both be read before the first connection.
Rule of thumb: Give an agent the least autonomy that gets the job done, and widen it only after you have watched it work.
How to start using AI in asset management
1. Pick one recurring question
Choose something you answer every quarter and dread: the state of infrastructure summary, the renewal shortlist, the deferred maintenance number. One question with a known right answer is the best way to judge whether the assistant is any good.
2. Look at your data first
AI does not fix bad data; it summarizes it. Ask the assistant to find the gaps as its first task: missing install dates, missing replacement values, stale condition scores. That is a useful result in itself.
3. Connect read-only for a few weeks
Give the assistant a read-only key and use it for questions and drafts. Compare its answers with what you would have produced by hand. You will learn quickly where it is reliable and where it needs checking.
4. Add write access where it saves real time
Logging work orders from a site walk, creating PM schedules, setting up compliance records: routine, reviewable writes are where an agent pays off first. Keep capital and condition data under human control for longer.
5. Write down your rules
Which data the agent can see, which records it can change, what always needs approval, and who owns the key. It does not need to be long. It does need to exist before the second person connects an assistant.
Where AssetLab fits
AssetLab has an MCP server, so our clients connect their own AI assistants and carry out work in AssetLab from a chat window. Here is what that looks like the week before a budget meeting:
list_site_fci_historyFCI by site, this year vs lastlist_asset_condition_assessmentsAssessments at the arena since Januarylist_asset_replacement_plansPlanned years for the assets that dropped
The arena is the only site that moved much: FCI went from 0.08 to 0.14. Three assessments account for most of it, and the roof is not planned for replacement until 2030.
| Asset | Now | Last year | Planned |
|---|---|---|---|
| Roof membrane | 34 Critical | 58 Fair | 2030 |
| Dehumidifier DH-1 | 44 Poor | 61 Fair | None |
| Ice plant compressor 2 | 52 Poor | 66 Fair | 2029 |
update_asset_replacement_planRoof membrane moved from 2030 to 2027create_work_orderRoof inspection and leak survey, high prioritycreate_asset_replacement_planDH-1 replacement added for 2028
That is an afternoon of exports and spreadsheets done in a few minutes, and the follow-up work is already in the system instead of on a to-do list. Every step was one of the controls above. The key was created for the assistant in Settings > API Keys. The assistant could only write because that key has write scopes for replacement plans and work orders, it showed the changes before making them, and all six calls are in the audit log under that key.
MCP access is part of the Enterprise plan, or the API keys add-on on other plans. The Trust Center lists where the MCP server and every other supporting service run.
AI and MCP in AssetLab
Your assistant of choice, connected to your real asset data, under controls you set.
Any MCP assistant
Claude, ChatGPT, Microsoft Copilot or any MCP client.
Your whole portfolio
Assets, condition, risk, work orders, projects and the capital plan.
Scoped keys
Read and write permissions per resource; read-only by default.
Audited
Every call recorded in the audit log under the key that made it.
Revocable
Revoke a key and every connected assistant stops at once.
REST API too
The same keys and scopes for integrations without an assistant.
Frequently Asked Questions
How can AI be used in asset management?
Today the most useful work is connecting information that already exists: answering questions across the asset register, condition assessments, work order history and the capital plan, drafting asset management plan sections and council reports, ranking renewals by condition, risk and level of service, finding gaps and inconsistencies in asset data, and researching the regulations that apply to your assets. The AI does the gathering and the first draft; people check it and make the decision.
What is MCP (Model Context Protocol) in asset management?
MCP is an open standard that lets an AI assistant such as Claude, ChatGPT or Microsoft Copilot connect to another system, see what it can do there, and use it. When your asset management software has an MCP server, the assistant can read your asset data and, if you allow it, create or update records, from a chat window. It is the piece that turns a general-purpose assistant into one that works with your actual portfolio.
Will AI replace asset managers?
No. The hard parts of asset management are judgement calls: which level of service a community can afford, which risk is acceptable, which project goes first when the money does not cover everything. AI is good at the hours of gathering and cross-referencing that come before those decisions. It changes how much of an asset manager’s week goes into finding and formatting information, not who is accountable for the decision.
Is it safe to give an AI agent access to asset management data?
It can be, with the right controls. Start with read-only access, give the assistant its own credential with only the permissions it needs, require a person to approve anything it writes, keep an audit log of every call, and make sure access can be revoked instantly. The bigger risk is not the connection itself but giving an agent more autonomy than the task needs before you have seen how it behaves.
What data do I need before using AI for asset management?
A reasonably complete asset register with locations, install dates and replacement values is the minimum. Condition scores, work order history and a capital plan make the answers much more useful. AI does not fix bad data: it summarizes it confidently. Cleaning up the register first, or asking the assistant to find the gaps as a first task, pays off quickly.
Which AI assistants work with AssetLab?
Any assistant that supports MCP, including Claude, ChatGPT and Microsoft Copilot. An Administrator creates a scoped API key in Settings > API Keys and connects it to the assistant. MCP access is included in the Enterprise plan and available as the API keys add-on on other plans.
We are still early
The pieces needed for AI agents to connect the information behind infrastructure decisions are mostly here. The governance is not, and that is the work for the next few years: deciding what an agent may see, what it may change, and what always needs a person. There is a lot to figure out. But the possibilities are pretty exciting.
Pick one question you answer every quarter. Connect an assistant read-only. See what it gets right.
If you want to see an AI assistant working with real asset data in AssetLab, we can walk you through it in about 20 minutes.
Related Articles
AI-Powered CMMS: How MCP & REST API Transform Maintenance
Connect Claude or ChatGPT to your CMMS. AssetLab's read-write MCP Server and REST API let AI assistants query assets, create work orders, and manage maintenance in natural language.
Facility Risk Assessment: Plan for Failure Before It Happens
Most facility risk lives in one person’s head. Which building systems to assess first, the four failure scenario questions to ask of each, a worked boiler example and a one-page response plan template you can copy.
Facility Compliance Tracking, and a Use We Never Designed For
Compliance in AssetLab is earned by completed work orders, not typed into a status column. How the score works, what inspectors see on a placard, and how one customer had Claude research his city requirements and create the compliance records against his asset registry.
