Context graphs: Marketing as the tip of the spear in the enterprise

Every CMO I talk with wants the same thing from AI agents: to run more campaigns, faster, without sacrificing brand consistency or compliance.

The constraint isn’t LLMs — ‌we have plenty of those. It’s dynamic organizational knowledge.

When your best campaign succeeds, can you explain why it worked? When legal approves an exception, does that create precedent or was it truly one-off? When your star content strategist leaves, where does their expertise go?

Here’s the uncomfortable truth: It lived in Slack threads, side conversations, and someone’s head. That person might have left six months ago.

Your systems show you final deliverables‌ — ‌the brief, approved content, published assets. But the reasoning—, the why, who and how behind the campaign—is gone.

---

The industry just discovered what we’ve been building

Over the holidays, several viral threads hit on the same insight from different angles: enterprise AI is stuck at incremental productivity gains because we’re asking AI to scale our work while giving it nothing to reason from.

The industry is starting to call this missing piece the context graph.

We’ve been working on this problem for over a year. We called it the orchestration graph, and here’s why the market is finally catching up to what we saw coming.

Aaron Levie articulated what he calls Jevon’s Paradox for knowledge work: When you make work 10x cheaper through AI, demand doesn’t drop—it explodes. Organizations do 10-100x MORE work because there are more use cases than previously contemplated.

Andrej Karpathy crystallized the moment for developers: “I’ve never felt this much behind as a programmer.” An entirely new abstraction layer is forming—agents, subagents, contexts, memory, permissions, tools, workflows—and nobody has mastered it yet.

Jaya Gupta and Animesh Koratana illuminated the bottleneck: Enterprise systems were built to store records (data and state), not to capture decision logic as it unfolds (reasoning and context). That gap‌ — ‌the missing “why” behind every “what”—is now the constraint keeping organizations from safely scaling AI usage.

What they’re describing is exactly what we’ve been operationalizing at WRITER.

For this kind of context to work in the enterprise, we must think beyond the representation of the context to the application of it. This introduces an important set of new considerations around the enterprise context graph, including:

• Capturing context graphs at individual, team, department, and org-wide levels
• Laying on the right policies and guardrails for compliant capture and usage of the Context graph so legal precedents or communication chains aren’t misapplied.
• Offering auditability and traceability so teams can see and influence what’s in the context graph and how it’s applied.

Below we’ll explore how we’re operationalizing this first for marketing teams.

Why coding adapted first‌ — ‌and why marketing is next

Let’s acknowledge the obvious: programming has seen the most radical AI transformation.

Claude, Cursor, GitHub Copilot—they’re all over everyone’s timelines right now. Developers are experiencing what Karpathy described viscerally: code assistants have fundamentally changed what “programming” means. It’s shifted from writing code to specifying intent, orchestrating AI agents, and ensuring correct behavior.

Why did coding workflows adapt so much faster?

The infrastructure was already there. Code is inherently structured‌ — ‌every function has inputs, outputs, and testable behavior. Repositories capture version history. Pull requests document decision logic. The “context graph” for code practically builds itself from Git commits, code reviews, and CI/CD pipelines.

But programming is one function.

What about teams that work with customers, create campaigns, build brand equity, manage cross-functional initiatives? How does this transformation reach the rest of the organization beyond “do what you do today, but have Copilot draft it faster”?

This is where the gap becomes clear‌ — ‌and where marketing becomes the critical proving ground.

Unlike code, marketing decisions are inherently organizational. You can’t write on-brand messaging without understanding brand evolution. You can’t approve competitive claims without knowing legal precedent. You can’t scale campaign creation without capturing what worked and why.

Programming got its transformation because the context infrastructure existed. Marketing gets its transformation now because we’re building the context infrastructure it needs.

When you sit with 50 CMOs trying to scale marketing operations with AI, you hear the same pain points:

• “Our best campaigns aren’t repeatable because we can’t capture what made them work”
• “Every new hire takes 6 months to learn ‘how we do things here'”
• “Legal approves exceptions, but we never know if it sets precedent”
• “We’re running the same 4 campaigns because we don’t trust AI with more”

This isn’t theory. It’s what we hear every week from marketing leaders at Vanguard, Prudential, and Qualcomm.

The infrastructure breakthrough that enabled Cursor and Claude for developers? That’s what we’re building for knowledge work and GTM teams — for marketers to start — through our own approach to retrieving, storing, and applying the marketing context graph. 

Why this hits marketing first (after coding) 

As context graphs enter the mainstream conversation, we’re seeing approaches that capture observable activity patterns across systems‌ — ‌watching how work flows and inferring processes from repetition.

For many workflows, that makes sense. Standardized processes like incident escalation or contract approvals benefit from pattern inference. After observing 100 incidents, you understand the escalation path.

But marketing is fundamentally different.

Here’s why we saw this problem crystallize in marketing:

Volume creates the richest learning dataset. Marketing generates hundreds of decision traces weekly—every campaign, content piece, channel test creates decisions about positioning, messaging, claims, tone. Legal reviews contracts monthly. Finance approves budgets quarterly. Marketing’s constant micro-decisions make it the ideal environment for proving context graphs actually work.

Inherent ambiguity demands explicit context. Sales has MEDDICC frameworks. Finance has policies. Legal has compliance rules. But when you ask “Is this on-brand?” the answer depends on evolving context and standards that live in people’s heads, not systems.

Every campaign is unique. You can’t learn “how to run a product launch” from observing 100 launches. The competitive landscape shifts. The audience evolves. The brand positioning iterates. Pattern inference doesn’t work when every execution requires fresh judgment.

The reasoning already exists—it’s just scattered. The “why” isn’t trapped in people’s heads. It’s documented in creative briefs, legal review threads, post-campaign analyses, brand guideline discussions. The problem isn’t missing documentation‌ — ‌it’s that this reasoning lives across Google Docs, Slack threads, Asana comments, and disappears when people leave.

Consider your company’s last major campaign‌ — ‌the kind that touches multiple teams and requires cross-functional coordination.

Can you answer these questions by looking at your systems?

• Why did legal approve that product claim when it’s usually pushback territory?
• What precedent was referenced from the last similar campaign?
• How did messaging for enterprise vs. mid-market get decided?
• Who approved the tone shift from the original brief, and why?

You can’t.

This is the enterprise bottleneck that’s keeping marketing stuck at incremental AI gains instead of the 10-100x expansion Levie describes.

What marketing leaders are seeing (and where they’re hitting the ceiling) 

With WRITER’s current platform, our customers have achieved impressive results: Vanguard saw 57% faster time to market for their first client-facing AI agent. Prudential drove 70% adoption across their global marketing team with a 40% boost in creative capacity. Qualcomm teams are saving 2,400 hours per month with 85% weekly agent usage.

But here’s what they tell us: Even with AI making individual tasks faster, they’re hitting a ceiling. Every new campaign still starts from scratch. When someone leaves, their expertise disappears. Legal exceptions don’t create searchable precedent. The reasoning behind success remains trapped in scattered documents.

That’s the problem we’re solving for.

How we’re building this at WRITER: Agents do the work, an enterprise brain connects the dots

We don’t try to read minds or infer intent from activity logs alone. We capture what marketing teams already document‌ in agent-mediated workflows — ‌and make it queryable, reusable, and compounding.

WRITER Agent sits in the execution path‌ — ‌where decisions actually commit. Not observing from the outside, but participating in:

• Campaign briefs that explain positioning decisions
• Legal review workflows that document approval reasoning
• Post-campaign analyses that connect outcomes to strategy
• Brand guideline applications that show how standards evolve
• Creative feedback loops that capture what worked and why
• Competitive intelligence that shaped messaging choices

And critically, it works across your entire stack. Your best campaigns aren’t decided in one system. The customer insight lives in your CRM. The performance data is in marketing automation. The creative brief is in your project tool. The approval happens in Slack. Real marketing decisions synthesize across all of these—some systems needs to see the full picture.

We’re calling this system an enterprise brain, and we’re building it now to make that scattered reasoning work for you:

As work happens, the enterprise brain captures decision traces‌ — ‌the reusable context like roles, definitions, and preferences and connects them to actions as they’re taken . Not every action, but meaningful decisions that shape outcomes: what customer insights informed your positioning, what competitive pressure shaped your messaging, what past campaign performance guided your channel mix, what brand precedent you referenced.

This is fundamentally different from:

• Knowledge graphs (which store facts)
• MCP (which enables access to stateful data and tools)
• Activity logs (which capture events)
• Audit trails (which track who did what)
• Pattern inference (which learns from repetitive processes)

Most enterprise AI investments focus on stateless interactions‌ — ‌each query starts fresh. An enterprise brain is fundamentally stateful‌ —it remembers, connects, and compounds over time.

Think of it this way: Your CRM knows you ran a campaign. Your marketing automation knows it got a 3.2% conversion rate. Your project management tool knows Legal approved it on Tuesday.

But none of them know why you emphasized security over innovation in the messaging, why Legal approved a competitor comparison this time when they usually don’t, or why that particular customer quote resonated with enterprise buyers but fell flat with mid-market.

Enterprise Brain will capture that living decision logic‌ — ‌creating a queryable record of your marketing organization’s accumulated wisdom.

And it scales safely with governance built in. As your team attempts 10x more campaigns, you need to maintain brand consistency, legal compliance, and strategic alignment. Decision traces enable this by making institutional knowledge queryable and precedent discoverable‌ — ‌so autonomy can expand without chaos.

For marketing leaders, this means finally having infrastructure that captures institutional knowledge. For CIOs, it means agentic systems that are governable, auditable, and actually get smarter over time.

How it will work: Two requirements 

After working with enterprise leaders deploying real AI workflows‌ — ‌not demos, actual production‌ — ‌we’ve landed on a framework for how organizational intelligence actually compounds when context is intentionally captured, governed and reused:

1. Your agents must sit in the execution path

Being in the execution path means work cannot happen without going through your agents. Not adjacent (recommendations), not downstream (analyzing after), not upstream (planning). Where the decision commits‌ — ‌where it becomes action that changes state.

Examples of being IN the execution path:

• Campaign content cannot publish without agent-powered approval workflow
• Legal/brand gates enforced through agent system
• Asset creation happens in agent environment
• Quote generation flows through agents
• Pricing exceptions require agent-mediated review

Being adjacent (viewing dashboards) doesn’t compound‌ — ‌you’re only seeing artifacts after reasoning is lost. To capture reasoning, you need to be there when the decision commits.

This is where WRITER Agent operates today.

2. They must capture decision traces at commit time

Decision traces are the organizational dark matter‌ — ‌the reasoning, context, and precedent that connect actions to the roles, definitions, preferences, and standards that informed them..

Different from campaign reports (what happened) or approval workflows (who signed off), decision traces capture:

• What customer insights informed your positioning
• What competitive pressure shaped your messaging
• What past campaign performance guided your channel mix
• What brand precedent you referenced
• What outcomes validated or contradicted your approach

If you’re not capturing the meaningful decision traces at the moment decisions commit, that knowledge is gone forever.

Every new campaign starts from scratch. New hires take months to understand “how we do things here.” When Legal approves an exception, nobody knows if it creates precedent. Your highest-performing campaigns succeed for reasons that never get documented.

This loss is expensive. But it’s also solvable.

This is what Enterprise Brain will capture automatically‌ — ‌not from inferring patterns over dozens of repetitions, but from the documented reasoning your teams already produce.

The transformation roadmap: What changes when memory compounds

Here’s an example of what we envision happening when agents and memory integrate in marketing:

Months 0-3: Faster decisions

Iterations drop precipitously, for example, from 8 to 2. Approval cycles could compress from 12 to 4 days as agents surface precedent instantly. Early adopters should see this compression happen within the first quarter.

Months 3-6: Better defaults

System discovers which messaging frameworks work for which audiences, which claims legal approves. Agent first drafts improve 3x by starting from learned patterns, not guesses.

Months 6-12: Safe autonomy

Blog posts with a certain number of (say 200+) precedents auto-approve. Launch announcements route to human review with precedent attached. Marketing output increases meaningfully with more sophisticated governance, not less.

Months 12-24+: Organizational intelligence

New CMOs query “Why do we emphasize security in enterprise campaigns but ease-of-use in SMB?” and get data-driven answers from actual outcomes, not tribal knowledge. New hires onboard in weeks by querying the system. Customer success insights automatically update product marketing strategies, which then refresh sales battlecards, leading to the launch of new demand generation campaigns.

Intelligence flows across agent workflows without manual coordination. Switching cost becomes massive because accumulated intelligence is irreplaceable.

Cross-functional intelligence compounds automatically as context graphs extend to other teams. When Sales captures a customer objection and Marketing develops messaging that addresses it, both teams’ agents get smarter. Legal precedent informs product claims. Customer success insights update demand gen campaigns. The context graph doesn’t create silos—it breaks them down.

This is the compounding moat advantage the industry threads were pointing at‌ — ‌and we’re building it with the same enterprise focus that’s driven our success so far.

---

Why we’re uniquely positioned to build this 

We’re not announcing vaporware. We’re building Enterprise Brain because:

We work exclusively with Global 2000 enterprises. Every insight about what context graphs need to capture comes from real marketing leaders struggling with this exact problem. We’re not theorizing—we’re solving the pain points we see every week.

We already sit in the execution path. WRITER Agent is deployed in production at companies like Vanguard, Prudential, and Qualcomm. We see where decisions commit. We know what needs to be captured.

We own the full stack. Our WRITER-built Palmyra LLMs, our Knowledge Graph, our agent orchestration—we control every layer. We’re not stitching together third-party components. We can instrument decision capture at every level.

We’ve been thinking about this for over a year. The orchestration graph architecture we published wasn’t marketing‌ — ‌it was our engineering team working through exactly this problem, informed by deep partnerships with enterprise marketing leaders.

---

I’m deploying the enterprise brain we’re building on my own marketing team first‌ — ‌not as a demo, but as production infrastructure. Over the coming months, I’ll be sharing real results: iteration counts week-over-week, approval cycle compression, how new team members onboard using queryable intelligence.

For marketing leaders thinking through this transformation:

• Request early access to be among the first to build a context graph with our enterprise brain
• Read our engineering deep-dive on the orchestration graph architecture
• Connect with me on LinkedIn — I’d welcome the conversation about moving from AI assistance to AI autonomy

The window to build your context graph is now. 

The industry just figured out the problem. We’re building the solution for marketing — ‌and you can be part of shaping it.

---