Products · agents
Your best engineer can review a few hundred readings in a day. An agent will review every one of them, the same way every time, and bring that engineer the few that need a decision.
What an agent is
One job
A single goal, set by a person
For example: check every protection reading against its criterion. Not a general assistant with opinions, one job done the same way every time.
The data it reads
Named sources, on the record
Which datasets it uses, how often they arrive, and when data last came in. If a source went quiet, you can see that it did.
The pipe it covers
Its scope, stated
Which systems, lines and miles. It works across all of them every time, not a sample and not the ones someone remembered to check.
The rules it uses
Yours, with their source
The operator's own criteria and thresholds, each carrying where it came from. Change a threshold and the agent works to the new one.
What it hands back
A finished result to review
What it found, where, the evidence behind it, and a suggested next step for an engineer to accept, change or reject.
What it never does
It never makes the decision
It flags and recommends. A person decides, and every decision is recorded with who made it and when.
Plainly
Answers questions across the platform, on the operator's own record. It is there today and people use it.
The six below are a prototype: named agents with goals, data and scope. The rules behind them run today as studies, which is how we know they work. The standing watch is not built yet.
The named agents
So your skilled people spend their time on judgment, analysis and decisions.
| Agent | Its one job | Who gets the result | Reads from |
|---|---|---|---|
| CP criterion agent | Watches every test station and rectifier against its criterion, continuously, and flags what moved | Corrosion engineers | Core |
| Survey comparison | Compares each survey with the one before it and flags readings that break from their own history or their neighbors | Corrosion engineers | Core, Decision Support |
| Run re-grade | Re-grades anomalies run over run and ranks the ones to confirm by dig or at the next run | Integrity engineers | Decision Support |
| Data check | Flags data that is missing, stale or wrong before anyone relies on it | Whoever owns the dataset | Core |
| Compliance clock | Tracks every obligation and its deadline against the asset it binds | Compliance leads | Compliance & Assurance |
| Capital planning | Ranks spend by risk and cost across the system | Finance, capital planning | Financial Planning |
The more packages an operator runs, the more an agent can read, because an agent reads across every package on the record.
How a day will look, once the standing watch is live
Overnight
Agents run
Across every line they cover, against the latest data that arrived.
Morning
One screen
What needs a look, worst first, each with its reason and the evidence behind it.
Review
The engineer decides
Reads the finding and what sits near it, then calls it: no action, watch, or propose work.
Work
It becomes an order
Proposed work goes to the field app and comes back as the record.
Start with the studies that the agents are built from.