The Agentic Hub pattern: one place to plan, route, and review agent work
AI-assisted research. This note was prepared with the assistance of AI tools. It may contain errors and is not professional advice. See our Legal Disclosures.
Agentic marketing has moved past the single assistant. Programs now run several specialist agents across research, content, paid media, lifecycle, and analytics, each with its own tools and owners. In our research, the hard problem at that stage is keeping all of them pointed at the same plan, more than improving any one agent.
Teams that start with one or two agents rarely need a coordination layer. Each agent has an owner, a workflow, and a few tools. Once a program reaches eight or ten agents, things change. Agents share audiences, depend on the same facts, hand work to each other, and compete for the same reviewers.
Coordination that once lived in people's heads and in scattered settings becomes the main source of failure. The Agentic Hub pattern is our answer: a single layer that plans, routes, holds shared context, enforces approvals, and keeps a run log. People set direction, and specialist agents do the work.
Why coordination breaks first
At this stage, we expect most failures to be coordination failures rather than capability failures. The content agent writes a strong email for an offer that changed yesterday. The paid media agent targets a segment the lifecycle agent is already messaging. Two agents ask the same reviewer to approve near-identical drafts an hour apart.
None of these agents did its own job badly. Each worked from a private copy of the plan, a private list of facts, or a private idea of who approves what. The mistakes live in the gaps between agents.
Adding more agents widens those gaps. Each new agent multiplies the number of handoffs and the number of places a stale fact can hide. A program can grow its capability each quarter and still feel less reliable each time.
Three layers and a feedback loop
The hub pattern divides a program into three layers, shown in our Agentic Hub architecture diagram. Each layer has a distinct job, and the boundaries matter as much as the jobs.
People set direction. Marketers and leaders define goals and plans, write guardrails and policy, and own approvals and review. They decide what the program is for and where its limits are.
The hub coordinates. It turns goals into plans, routes work, keeps shared context, enforces approvals, and records each run. It coordinates all agents and people against one plan.
Specialist agents act. Research, content, demand, lifecycle, and analytics agents do the actual work. They reach the CRM, the CMS, email, ad platforms, and analytics only through governed tool connectors with scoped permissions, hard guardrails, and logging on each call.
The loop closes when outcomes and learning flow back into plans. A campaign result, a rejected draft, or a guardrail trigger is information the next plan should use.
What the hub does
The hub has five jobs. Each one replaces a coordination task that would otherwise be repeated, inconsistently, inside each agent.
- Plan: hold campaign goals, plans, and constraints in one place, so agents read the current plan rather than keeping their own. See the planner-executor pattern.
- Route: receive each request and task, and send each to the right specialist agent or person. See the router agent.
- Hold shared context: store approved facts, audience definitions, brand rules, and campaign decisions once, and provide them to each agent that needs them.
- Enforce approvals: track the state of each piece of work, and make sure nothing reaches a customer without the approvals people have required. See state machines for campaign orchestration.
- Keep a run log: record each run, decision, approval, and outcome in one place, so evaluation, audit, and improvement all draw on the same record.
Guardrails deserve a note. People write them, and the hub carries them into each plan and routing decision. The hard limits on spend, audiences, and data access are also enforced at the tool connectors, so an agent cannot bypass them even if a plan is wrong. We prefer that double layer to trusting either check alone.
What the hub does not do
The hub does not write content, analyze campaigns, or decide strategy. Its job is coordination. A hub that starts producing work becomes a second, larger agent with unclear accountability.
It also does not replace people. Goals, guardrails, and approval rules come from the people who own the program, and the hub enforces them rather than inventing them. When the hub cannot route a task or a plan conflicts with a guardrail, it should stop and ask. Our post on supervisor agents and their limits covers why we keep judgment calls with people.
Keeping the hub narrow is what makes it dependable. It should be the most boring, most predictable part of the system.
A worked example
Consider a hypothetical retailer running a weekend promotion across email, paid social, and the site. A merchandiser updates the plan in the hub: 20% off outerwear, Friday to Sunday, excluding two clearance lines.
The hub records the change and routes three tasks. The content agent drafts email and site copy from the shared offer record. The demand agent prepares ad variations against the approved audience definition. The lifecycle agent schedules sends, checking the run log so it does not message customers already in an active winback sequence.
Each draft passes through the approval state the team configured. Copy with a price claim goes to a human reviewer, and routine ad variations go to a sampled review. The connectors log each call.
On Saturday morning, inventory runs low on one jacket. A person updates the plan to drop it. Because all three agents read from the same record, the change reaches email, ads, and the site in one step. Without a hub, the same change means three separate edits and a good chance one channel keeps promoting a sold-out item.
How to tell you need one
We suggest watching for a handful of signals. Any two of them together suggest coordination has become the bottleneck.
- Channel inconsistency: customers see different offers, dates, or claims depending on where they look.
- No clear answer to a simple question: nobody can say quickly what the agents are doing right now.
- Duplicated reference material: the same product facts or brand rules live in several agent instructions.
- Uneven approvals: the review process depends on which tool produced the work.
- Slow incident review: reconstructing what happened means pulling logs from several systems.
Teams with only a couple of agents can usually wait. They should still design each agent with clear inputs, clear outputs, and a log, so it can move under a hub later without being rebuilt.
What to measure
A hub earns its place by making the program more consistent and easier to govern. We recommend tracking four measures from the first month.
- Plan-to-channel lag: time from a plan change to all affected channels reflecting it.
- Consistency defects: cross-channel mismatches found in review or reported by customers.
- Approval cycle time: time work spends waiting for review, split by approval tier.
- Trace completeness: share of customer-facing outputs you can trace from goal to outcome in the run log.
If these do not improve within a quarter, look first at whether agents still keep private copies of plans or facts. That is the bypass we would check first.
Open questions
The pattern raises questions we are still studying. How much planning should the hub do itself, and how much should stay with a dedicated planner agent? How should one hub coordinate with another, such as a brand team's hub and a regional team's hub? And how fast should outcomes feed back into plans before the loop starts chasing noise?
We expect the answers to depend on program size and risk, and we will report what we find. The orchestration run diagram shows how a single piece of work moves through the hub today.
How to apply this: list where your campaign plans, approved facts, audience definitions, approval rules, and run logs live today. If the answer is several places, start a hub by moving the plan and the shared facts into one layer that all agents read.
Related research
Holiday campaign orchestration: a week-by-week agent plan
Holiday campaigns pack channels, offers, and deadlines into a short window. A week-by-week plan shows where agents help and where people stay in charge.
State machines for campaign orchestration: making approvals impossible to skip
Campaigns move through predictable states: draft, review, approved, live, paused, complete. Modeling them explicitly makes agent orchestration predictable.
Orchestrating across channels: one plan, many executors
Customers meet a brand across email, ads, social, sales, and the website. Orchestration keeps each channel's agents telling one story from one shared plan.
Put this research to work
Marketix turns these patterns into software: agents, automated workflows, and a hub that keeps people in control. Sign up for early access.
