The agent architecture
The map of the ecosystem — how the playbook’s Every section, the orchestration layer, and the per-area specialists fit together as one operating system. From the human strategy layer at the top to the systems of record at the bottom, every layer earns its place: humans own judgment, orchestration agents route signals, execution agents do the work, the data backbone captures every action, and the systems of record hold the long-term truth.
YOU DO NOT NEED EVERY AGENT
This is a reference architecture, not a shopping list. The right agent system for your company is a tailored subset — chosen against your tech stack, your motion, your team capacity, and your current bottleneck. Best-in-class teams start with three to five agents that close their biggest pipeline or productivity gap, run them clean for 90 days, then add the next two. The full architecture below shows what’s available; the Tech-Stack readiness selector tells you which subset is unblocked right now in your environment.
HOW TO READ THIS PAGE
The five layers are a top-down view of the marketing operating model. Read them in order: Layer 1 (Human Strategy) sits above the agents and tells you where humans make the calls. Layer 2 (Orchestration) shows the 8 cross-pillar agents that route signals across areas — the spine. Layer 3 (Agent Execution) shows the 27 per-area specialists doing the actual work. Layer 4 (CDP/Data Backbone) is the event-capture system that makes the upstream layers measurable. Layer 5 (Systems of Record) is the platform stack underneath all of it.
No agent is allowed to make a customer-facing decision without a human in the loop. No human is allowed to publish customer-facing content without a Brand Voice Agent pass. The architecture enforces the discipline.
Judgment, relationships, business-case ownership. Where humans — not agents — make the calls.
The principle: every agent below reports to a named human at this layer. Personhood and accountability prevent the “the agent did it” failure mode.
The spine. Cross-pillar agents that connect signals across every section — the layer most marketing functions never build.
CROSS-PILLAR
Ingests signals from every module and routes each to the right downstream agent or human owner. The central nervous system.
CROSS-PILLAR
Maps every marketing activity to the pipeline, expansion, and retention outcomes that traced from it. Closes the loop for the CFO.
CROSS-PILLAR
Aggregates per-account signals from every source into a single 360 view. The AE’s one-page brief in 10 seconds.
CROSS-PILLAR
Indexes every customer story, case study, reference, and quote. Surfaces the three best-matched proof points in seconds.
CROSS-PILLAR
Reads every draft output from every agent BEFORE it ships. Blocks publication if voice/forbidden-language thresholds fail.
CROSS-PILLAR
Runs eval suites against every agent’s output on cadence. Flags drift in agent quality before it ships at scale.
CROSS-PILLAR
Watches every outbound channel and enforces send-limits per recipient. Prevents nurture-overload failure modes.
CROSS-PILLAR
Reads every other agent’s output for updates that should propagate back to the Operator Brief. Keeps the Brief from going stale.
All 8 orchestration agents are specified in the AI Operating Model — where the operating discipline of the entire ecosystem lives.
27 per-area specialists. The agents that do the actual work inside each discipline.
Watches firmographic, CRM, and analyst data quarterly. Re-runs TAM/SAM/SOM. Drafts the CFO defense memo.
Scores every account against the ICP scoring model. Surfaces tier-1 fits weekly.
Simulates the buying committee personas reacting to every major message draft.
Monitors propensity signals for every account weekly. Triggers Hot-band alerts to sales.
Watches CRM and product analytics. Re-runs customer-language synthesis quarterly. Stress-tests messaging.
Crawls competitor sites, hiring pages, changelogs, funding news. Updates competitive-context.md weekly.
Monitors competitor pricing weekly. Tracks discount-rate trends. Flags ASP erosion.
Drafts personalized interview scripts. Codes transcripts. Produces quarterly leadership readout.
Owns Brief Section 6 (Brand). Refreshes voice rules from customer-language updates.
Watches quarterly scorecard inputs. Flags CMO-trap drift. Drafts the self-audit memo.
the · CANONICAL
Website optimization, conversion rate, content updates. The first agent every marketing function should ship.
the · CANONICAL
Paid channel optimization, budget allocation, ad creative drafting. The demand-engine specialist.
the · CANONICAL
Event coordination, GTM execution, attendee outreach. The field-and-community specialist.
Runs the content calendar cadence. Topic-cluster maintenance, comparison content, AI-Overviews optimization.
MQL nurture, lifecycle programs, deliverability discipline, newsletter cadence.
Founder LinkedIn drafts, exec voice cards, company-page cadence, comment-strategy targets.
Channel mix recommendations by ACV+motion, CPL ceilings, organic-first rule enforcement.
Tier-1 selection, intent-data sources, 1:1 playbook drafts, sales-marketing handoff orchestration.
Tier-1 flagship through Tier-4 webinars, event-lifecycle from T-12 months to T+30 days, sponsorship math.
G2/Capterra/peer review cadence, response discipline, Buyer Intent integration.
Tier-1 Champions tracking, case-study cadence, expansion playbook drafts, community programs.
Earned-media drafts, LLM-citation framing, the CFO defense for the PR budget.
Watches MQL/SQL/SAL math weekly. Flags coverage shortfalls. Drafts the weekly demand forecast.
Weekly model run, saturation alerts, board snapshot, sensitivity analysis.
Watches spend vs. plan by channel. Flags variance above 10%. Drafts monthly budget review.
Builds the Weekly Pipeline Call agenda. Aggregates Monthly Council inputs. Drafts the Quarterly Revenue Handbook.
Weekly ELT Dashboard drafts, Monthly State of Union, board snapshot, CEO 1:1 prep, CFO check-in.
Every agent action is an event. Every event is captured. Every record stays current. The data layer that makes the upstream layers measurable.
Every agent fires an event in real time when it acts: advocacy act logged, program enrollment updated, health score changed, touchpoint recorded, quote extracted, reference completed, review submitted, certification earned.
Single source of truth for all customer-engagement data. Customer Data Platform (Segment-class, mParticle-class) feeds Data Warehouse (Snowflake/Databricks/BigQuery-class) for long-term analysis.
CDP routes every event to the right system of record automatically. CRM gets enriched contacts; CS platform gets health-score updates; advocacy platform gets activity logs; rewards engine gets trigger thresholds.
The principle: the data backbone is what lets the function answer the CFO’s questions in hours, not weeks. Skip this layer and the upstream layers stay anecdotal.
The platforms that hold the long-term truth. Categorized by function, vendor-agnostic.
How the layers compound
The diagram above looks like a hierarchy. It isn’t — it’s a closed loop. Humans set strategy. Agents execute. Events get captured. Systems hold the truth. The truth flows back up — via the Revenue Attribution Engine, the Account Intel Hub, the Brief Sync Agent — to the humans, who adjust strategy. The cycle runs continuously.
The compounding happens because each layer makes the next one faster. A human deciding on a new bet without the layer below them is gambling. A human deciding on a new bet with the data backbone showing five quarters of attribution math, the Account Intel Hub showing the top-50 propensity-scored accounts, and the Brief Sync Agent showing which assumptions have drifted is making an informed call in 30 minutes — not 30 days.
The risk is also stacked. A weak Layer 1 (vague strategy, undocumented Brief) means every layer below it amplifies the vagueness. A weak Layer 4 (broken event ingestion) means every layer above it is operating on anecdote. The architecture is only as strong as its weakest layer — which is why the and the (Budget) and the all sit in the playbook as first-class pillars of the playbook.
THE GOVERNANCE PRINCIPLE
Every agent in Layer 3 reports to a named human in Layer 1. Every agent’s output passes through Brand Voice Agent before publish. Every signal generated by Layer 3 fires an event into Layer 4. Every record in Layer 5 is queryable by an agent in Layer 3 via the CDP in Layer 4. The architecture isn’t aspirational — it’s the operating discipline written down.
Building this is the work of a year, not a quarter. Most teams ship the first three canonical agents (Web Operations, Performance Marketing, Field Marketing) in their first 90 days, the orchestration layer over the following two quarters, the CDP/data backbone in year two, and the full systems-of-record consolidation as a multi-year journey. The architecture page tells you where you’re aiming. The Operator Brief page tells you how to get there.
What this unlocks
The companies that build all five layers have a structural advantage over the ones that ship agents in isolation. The agents themselves are commodities — anyone can prompt Claude. The orchestration layer, the CDP/data backbone, the systems of record, and the human-strategy layer that ties it all together — that’s where the moat lives. Build the architecture. The agents come next.