Glossary

Applied AI, in plain words.

The terms behind our perspectives, for chief executives, boards and investors. Each is defined once and linked to where we argue it in full.

A

Agent: Software with a job to do, like a team member with a named role. It prepares, drafts and chases; anything that commits the business waits for a person. See From tools to teammates.
Agent action: One piece of finished work by an agent, such as logging a call or drafting a follow-up. The unit piMonk prices by, with an allowance included and extra use capped at a ceiling you set. See The end of the seat.
AI-native: Software designed from the start for agents to do the work, rather than AI added to a system built for people to type into.
Applied AI: AI built around one business job, such as running a pipeline or answering customers, and delivered inside the software where that work happens. The opposite of a general tool that sits beside the work. See The bridge to applied AI.
Applied-AI grid: Kintoro's map for adoption: business functions down the side, the stages Land, Expand and Compound across the top. Each square is one team, one job and a measured result. See The applied-AI grid.
Approval: The moment a named person accepts, edits or rejects an agent's proposed work before it reaches a customer, a price or a payment. See Judgement is the scarce resource.

Baseline: What a job costs and how long it takes before agents start, agreed in writing. Without it, no return can be shown. See Where the return comes from.

Ceiling: The monthly limit a customer sets on extra agent work. Nothing is billed above it, and extra use is off until switched on.
Compound: The third stage of adoption, months three to twelve: managers supervise agents day to day, and what one function learns starts to help the next.
Customer zero: Kintoro moved its own company onto piMonk before offering it to anyone else. Our software run rate fell 83 per cent in three months, counting staff time as well as licences (Kintoro internal figures, 2026). See customer zero.

Decision rights: Who may decide what, written down. When agents do more of the work, explicit decision rights are what keep a company in control. See The company of 2030.

Expand: The second stage of adoption, days 30 to 90: every team in a function, and the next job.

Human in the loop: A way of working in which agents do the routine work and a person decides anything that commits the business. Our short form: agents do the work, people keep the decisions.

Judgement: The ability to decide well when the answer is not in the data. As answers become cheap, judgement becomes the scarce resource. See Judgement is the scarce resource.

Land: The first stage of adoption, the first 30 days: connect the data, name an owner, and get one team doing one job with agents.
Large general AI model: A very large AI model trained to do almost anything. Powerful, and the most expensive to run for routine work. See Sovereign intelligence.
Leapfrog: Skipping a stage of development, as many economies skipped landlines for mobile phones. Growing companies can skip the expensive middle stage of business software. See The leapfrog economies.

Operating leverage: Growing revenue faster than costs. When agents carry the routine work, a company can grow without its headcount rising in step. See Where the return comes from.

Payback: How long it takes for what a change returns to cover what it cost. The simplest test an AI programme should pass.
Pilot: One team, one app, 30 days of real work, measured against a baseline, with the day-30 decision (expand, extend or stop) agreed before it starts. See how a piMonk pilot works.

Run rate: What a company pays each month to run its software, including the staff time spent administering and maintaining it, not only licences. The first return from applied AI to show up in the accounts.

Seat pricing: Charging for software by the number of people with a login. It suited software people used; it fits poorly when software does the work. See The end of the seat.
Small AI model: An AI model trained for one job. Cheaper and faster to run for routine work, and able to run on servers a company controls. See Sovereign intelligence.
Sovereign intelligence: Owning your data and what is learned from it, even when you rent the thinking. See Sovereign intelligence.
System of record: Software that stores what happened, such as a traditional CRM. People do the work; the system records it.
System of work: Software that does the work, with people deciding. The next generation of business software. See From tools to teammates.

Time to value: How long from signing to the first measured result. With applied AI it should be counted in days, not quarters.

Usage pricing: Paying for the work done rather than the people logged in. Fair when the unit is visible, the allowance covers normal use, and the customer controls the ceiling.

Value creation plan: The plan an investor agrees with a company it backs to raise that company's value. Applied AI belongs in it as run rate, capacity and speed, each measured. See Where the return comes from.

The terms in practice

Approval rules, agent actions and pricing, in piMonk's own glossary.