Most people picture an AI tool the same way: you open it, you type, it answers, you close it. That is one way an AI employee can work. It is not the only one, and for a lot of jobs it is the worst one, because it still depends on a human remembering to go use it.
A real AI employee can go to work in three ways. It can run on a schedule, deliver on demand, or fire on a trigger. Choosing the right one is the difference between an employee that quietly does its job and a tool that sits unused until someone remembers it exists.
This is a field note on those three deployment types, what each is for, and a one-line rule for picking the right one for any job in your business.
The three AI deployment types, defined
An AI employee’s deployment type is simply what causes it to start working. There are three.
- Scheduled. It runs on a clock, whether or not anyone asks. Every Friday at 7am, every morning before standup, the first of the month.
- On Demand. It delivers the moment a person asks, and only then. You hand it an input, it returns the output.
- Triggered. It acts when an event happens in your systems, with no human in the loop. A refund request lands, inventory drops below a threshold, a new lead hits the CRM.
That is the whole taxonomy. Every AI employee runs as one of these three, and some shift between them as they mature.
How to choose: ask what starts the work
Here is the rule that makes the choice obvious. Ask what starts the work.
If a clock starts it, schedule it. If a person starts it, make it on demand. If an event starts it, trigger it.
You are not choosing a personality for the AI. You are matching how it runs to how the work already arrives. Most jobs announce their own type the moment you ask that question.
Scheduled: work that should just happen
Scheduled employees are for recurring work with a predictable cadence, the things that should happen on a rhythm whether or not anyone has the bandwidth to initiate them.
The classic example is a weekly leadership memo. Every Friday, an employee gathers the week’s notes and numbers and writes a sharp summary of what moved, what stalled, and what needs a decision. Nobody has to remember to run it. It clocks in on its own.
Use Scheduled when the value is in consistency and the cost is usually that the task gets skipped when things get busy. Weekly reporting, recurring content, routine reconciliations, standing digests. The schedule is what guarantees it actually gets done.
On Demand: work that starts when you ask
On Demand employees are the closest to how people already think about AI. You give it an input, it returns an output, right now.
A sales call analyst is a good example. You finish a call, hand over the transcript, and get back a scored breakdown, the objections that surfaced, and a ready follow-up email. It does not run on a clock, because calls do not happen on a clock. It runs when you have one to feed it.
Use On Demand when the work is irregular and a human is the natural starting point. Drafting a proposal, analyzing a document, prepping for a specific meeting. The person decides when, the employee handles the how.
Triggered: work that starts on an event
Triggered employees are the ones most businesses never think to build, and often the highest leverage. They act when something happens in your systems, with no person involved.
A refund request hits the help desk and an employee drafts the response and flags the edge cases. Inventory on a hero product drops below a set level and an employee alerts the right person with a reorder recommendation. A high-value lead enters the CRM and an employee assembles the research brief before anyone has opened the record.
Use Triggered when the work is reactive and time-sensitive, and when waiting for a human to notice is itself the problem. The trigger removes the lag between something happening and something being done about it.
A quick comparison
| Type | What starts it | Best for | Example |
|---|---|---|---|
| Scheduled | A clock | Recurring work that gets skipped when busy | Weekly leadership memo |
| On Demand | A person | Irregular work a human initiates | Sales call analysis |
| Triggered | An event | Reactive, time-sensitive work | Refund triage on a new ticket |
Most employees evolve from one type to the next
Deployment type is not permanent. It is common for an employee to start as On Demand, prove itself, and then graduate to Scheduled or Triggered once you trust it to run without you watching.
That progression is the whole point of an owned system rather than a rented tool. You start it where the risk is lowest, with a human pressing the button every time, then move it to autonomous running once it has earned it. A tool you rent cannot make that climb. An employee you own can, because the four hours a week you spend training and reviewing it are what make it safe to let off the leash.
The reason this matters is the same reason most AI becomes shelfware. On Demand alone depends on a human remembering to open the tool. Schedules and triggers remove that dependency, which is exactly what MIT’s 2025 research pointed to: the AI that delivers value is the AI woven into real workflows, not the app someone has to remember to use.
Which type should your first AI employee be?
For a first build, the best deployment is usually the one tied to your most reliable pain. If there is a report you keep meaning to write, that is Scheduled. If there is a request type that keeps flooding your team, that is Triggered. If there is a high-skill task you do irregularly and want to do faster, that is On Demand.
You do not have to decide this alone or up front. Part of what a Map does is look across your operation and tell you not just what to build, but how each piece should run.
Frequently asked questions
What are the deployment types for an AI employee? There are three. Scheduled, which runs on a clock; On Demand, which delivers when a person asks; and Triggered, which acts automatically when an event happens in your systems. The type is defined by what causes the work to start.
How do I choose between Scheduled, On Demand, and Triggered? Ask what starts the work. If a clock starts it, schedule it. If a person starts it, make it on demand. If an event starts it, trigger it. The job usually tells you its own type.
What is a triggered AI employee? One that acts on an event with no human in the loop, such as drafting a response the moment a refund request arrives, or flagging a reorder when inventory drops below a threshold. It removes the lag between something happening and something being done about it.
Is on demand the same as a chatbot? It is similar in feel, since a person initiates it, but an on demand AI employee is scoped to a specific job, trained on your business, and able to act in your tools. A chatbot answers questions. An on demand employee produces a finished piece of work.
Can one AI employee use more than one deployment type? Yes. Many start On Demand so a human reviews every run, then graduate to Scheduled or Triggered once they have proven reliable. The ability to make that climb is a feature of owning the system rather than renting a tool.
Which deployment type delivers the most value? It depends on the job, but Triggered is the one most businesses overlook and often the highest leverage, because it acts on time-sensitive work the moment it appears instead of waiting for someone to notice.
How do I know which type my first AI employee should be? Tie it to your most reliable pain. A report you keep meaning to write is Scheduled. A request type that floods your team is Triggered. A high-skill task you do irregularly is On Demand. A Map will tell you what to build and how each piece should run.
Want to know which jobs in your business should run on a schedule, on demand, or on a trigger? Book a strategy call, or start with a Map and get a ranked plan in three weeks.
Lisa Hayford is the founder of The Lasting Method, where she designs and installs AI employees for DTC brands doing $1M to $10M in revenue. She has built and rolled out full AI implementations across entire operations, from marketing and support to operations and finance, with one rule: the business owns what gets built, not a vendor. She writes about what it actually takes to make AI stick inside a company instead of becoming shelfware.