Demonstrated Routine (Show-and-Teach)
Show-and-teach describes training AI agents by demonstration: the agent observes a single execution, stores steps, rules and exceptions as a routine and then runs it independently.
A demonstrated routine is created when an agent watches a human perform a workflow and derives a reusable automation from it — instead of the workflow being modelled up front.
Explanation
Classic automation starts with a process diagram somebody has to maintain. Demonstrated routines invert the order: the work happens first, the automation follows. That lowers the barrier to entry considerably but shifts the quality question. The agent learns exactly what was shown — including shortcuts, undocumented edge cases and mistakes. Robust practice therefore means demonstrating a clean run, deliberately showing exceptions, reviewing the derived routine like code and logging its runs. Grok Bot made the pattern broadly visible in 2026, but the principle also appears in other vendors' computer-use agents.
Marketing Relevance
For marketing operations this is the fastest route out of recurring manual work: assembling reports, chasing approvals, maintaining data. Value only appears with review though — a demonstrated routine is a draft, not a finished process.
Example
A campaign manager shows an agent the weekly reporting process once. From the following week the agent delivers the draft itself and only flags deviations it cannot classify.
Common Pitfalls
If a faulty or untypical run is demonstrated, the error becomes the rule. Without versioning and logs you cannot reconstruct later why a run deviated. Routines with write actions additionally need approval thresholds.
Origin & History
Programming by demonstration is known from robotics. Computer-use agents made it transferable to screen work from 2025; in 2026 it became the standard onboarding path in agent products such as Grok Bot.
Marketing Use Cases
Weekly report assembly across several interfaces.
Chasing open approvals and reminding owners.
Data maintenance in systems without APIs such as partner or billing portals.
Onboarding checklists with many small manual steps.
Frequently Asked Questions
Is this the same as RPA?
Related but more flexible. RPA follows rigid selectors, demonstrated routines use a model that interprets interfaces and bridges small changes — at the cost of being less deterministic.
How many demonstrations does an agent need?
Usually one clean run covers the normal case. Extra runs pay off for exceptions, otherwise the agent only knows the happy path.
How do you stay in control?
Through reviews of the stored routine, logging of every run and mandatory approval for actions that change data or leave the organisation.
Related Services
Go deeper: Agentic AI Hub · Governance & compliance