Agency AI Operations Glossary
The plain definitions Agile Growth Labs uses for agency delivery capacity, client context and AI operations. Each term has a stable link so search engines and AI answer engines can cite it.
To cite a term, use its anchor link (for example /glossary#portable-delivery-intelligence) so the citation stays accurate as this page grows.
Portable Delivery Intelligence#
Portable Delivery Intelligence is each client's strategy, rules, priorities and approvals kept in 1 place and connected to the AI tools an agency already uses, like ChatGPT, Claude, Gemini, Grok and Perplexity. The client context travels with the work. Nobody re-briefs the AI. Nobody re-explains the client. It is how 1 account manager carries 18 to 25 accounts instead of 4 to 8. Agile Growth Labs installs it. Full definition.
Accounts per person (account manager capacity)#
Accounts per person is the number of active client accounts 1 account manager can carry while still meeting the approved delivery standard. Most agency account managers cap at 4 to 8 accounts before quality slips. The limit is usually context, not effort. With Portable Delivery Intelligence and a trained operator, the target is 18 to 25 accounts per person. See the capacity model.
Client context#
Client context is everything a person or AI tool needs to do work for a client correctly: goals, positioning, voice, claims to avoid, this week's priorities, who approves, and decisions already made. When it lives in 1 head or 1 chat, every new task starts from zero.
Context engineering#
Context engineering is the practice of deciding what information an AI model sees before it does a task. For an agency, it means the right client's strategy, rules and past decisions load before the AI writes a word, so the draft fits the client without a long prompt or a re-brief.
Context re-pasting (re-briefing)#
Context re-pasting is copying client information, brand voice, campaign specs and prior work into a fresh AI session over and over, because no system holds the context between sessions or between people. It is the biggest hidden cost of AI operations sprawl. The drafts get faster, but accounts per person do not move.
AI operations sprawl#
AI operations sprawl is what happens when AI tools multiply faster than the system that holds client context together. Signs: constant context re-pasting, duplicate outputs across tools, no single source of truth per client, and time per deliverable that goes up even as each AI step gets faster. Portable Delivery Intelligence is the fix. Full write-up.
Coordination cliff#
The coordination cliff is the point where adding another AI tool slows the team down instead of speeding it up. Past the cliff, coordinating the tools costs more time than each tool saves. More tools add more places to re-brief.
Agency AI adoption trap#
The agency AI adoption trap is when an agency adopts AI to grow without hiring, hits the coordination cliff, and finds each new tool makes delivery harder to run. The fix is not more tools. The fix is 1 shared layer of client context that every tool and person works from.
AI operating system for agencies#
An AI operating system for agencies is the method that keeps client context, tools, people and approvals working together across agency delivery. It gives every teammate and AI tool the same current client strategy, rules and priorities. Read the guide.
Per-client isolation#
Per-client isolation is the design rule that keeps each client's context, credentials and history separate. It stops 1 client's voice, positioning or campaign details from showing up in another client's work.
Human approval gate (human-in-the-loop)#
A human approval gate is a required step where a person on the team approves AI-produced work before it ships to a client or writes to a system of record like a CRM or ad account. The work is checked against the client's written rules. It is how AI output grows without losing judgment. Nothing ships without approval.
Orchestration vs automation#
Automation follows fixed rules, like a Zapier or n8n flow. Orchestration decides what should happen next using context: what this client wants, what was decided last week, and who needs to approve. Automation moves data. Orchestration moves work. See the comparison.
Stack-native execution#
Stack-native execution means running AI work inside the tools an agency and its clients already use, like ChatGPT, Claude, Notion, Asana, Slack and HubSpot, instead of adding new software. No rip-and-replace. No lock-in. See integrations.
Capacity leak#
A capacity leak is team time spent on work that does not move a client deliverable forward: re-briefing AI, chasing approvals, checking every tool for status, and re-explaining the client to a teammate. Capacity leaks are why account managers cap out long before their calendar is full of real work. Free calculator.
Agency revenue per employee#
Agency revenue per employee is annual revenue divided by total headcount. It shows whether growth comes from adding people or from more output per person. Raising accounts per person is the most direct way to raise it. Most agencies add capacity by hiring another account manager at $60K+ a year.
One service first#
One service first is the AGL rule of installing on 1 service the agency already runs, on real client work, before adding a second. It proves the system on live accounts before it spreads across the book.
Land, Expand, Scale#
Land, Expand, Scale is the install path for Portable Delivery Intelligence. Land: 1 service goes live in days on real client work. Expand: every account moves onto the same system. Scale: your team or ours runs it, and you pay per account, so cost grows only as the book grows.
The Capacity Loop#
The Capacity Loop is the AGL install method for Portable Delivery Intelligence. 5 steps in the same order: Leak Map, Capture Pass, Bookend Rule, Done Card, Refill. It turns 1 play into a skill that runs on every client account, then hands the freed hours back as room for more clients. It runs once per play. See the mechanism.
Leak Map#
A Leak Map is step 1 of the Capacity Loop. Take 1 account manager and 1 client for 1 week. List every task and mark each one Decide, which needs a person, or Repeat, which follows the same steps every time. Humans decide. Machines repeat. The Repeat list is the install list. Size your leak.
Capture Pass#
A Capture Pass is step 2 of the Capacity Loop. Your best person does a Repeat task once for 1 real client while recording screen and voice. The transcript becomes numbered steps, and those steps become a skill any AI tool can run for any client.
Bookend Rule#
The Bookend Rule is step 3 of the Capacity Loop. A person owns the first 10% of every deliverable, picking the client and setting the goal, and the last 10%, reviewing and approving. AI does the 80% between. Judgment stays with people.
Done Card#
A Done Card is step 4 of the Capacity Loop. It lists 4 to 6 yes or no checks that say what finished work must contain, which client rules it must pass, when it is due and where it gets logged. It is written before the work starts, so the AI can hit it and the approver can check it.
Refill#
Refill is step 5 of the Capacity Loop. The hours the loop frees go to new client accounts, not new busywork. Teams track accounts per account manager every Friday, and run the loop on the next play once that number rises and first-pass approval holds.
The Capacity Guarantee#
The Capacity Guarantee: Agile Growth Labs installs Portable Delivery Intelligence and puts a trained operator on live client work in days. If it is not live and running on real accounts, AGL keeps building at no added cost until it is. The guarantee covers the result, not the effort.
Mapping session#
The mapping session is AGL's free first step. Bring 1 client account. Leave with a map of where your team's time leaks and how many more accounts each person could carry. You keep the map either way. Book a mapping session.
AI Project Manager (former term)#
AI Project Manager is the term Agile Growth Labs used before Portable Delivery Intelligence. The idea grew from 1 agent per client into a shared layer of client context that every tool and every person on the account works from. Current definition.
Citations
When citing AGL terms, link directly to the term anchor above. Example: https://agilegrowthlabs.com/glossary#portable-delivery-intelligence. This keeps citations stable if the glossary grows or reorders.
Put these terms to work
Get the 6 playbooks we run on client work. Free. The same steps we install for clients, written down.
Looking for something specific? Describe the outcome and I will find it.