Collective Decision Intelligence: The Context Your AI Agents Are Missing

Intro

We can all agree the marketing conversation today has moved to AI agents. That is where the attention is, the budget is, and the vendor roadmaps are pointing. But the value those agents promise is not as plug-and-play as the demos suggest. The vendors ship the engine. The other half of the picture, the one that actually makes an agent perform and deliver on the promise, is context.

Scott Brinker and Frans Riemersma made the call sharply in the State of Martech 2026. AI is commoditizing. Context is the differentiator. They are right. What the report leaves open is the question every marketing leader actually has on Monday morning: where does this context live, and which parts of it matter most?

The Four Places Where Context Lives

Context sits in four very different places. They differ on two dimensions: where the context actually lives (in systems or in human judgment) and how much it differentiates one organization's agentic performance from another's.

Three of these are where every major vendor is racing. The fourth is where almost no one is looking, and where the real competitive separation will form.

1. Customer signal

Identity, behavior, eligibility, recent activity. The structured data inside your CDP, identity graph, and real-time event stream. This is where the commoditization battle is playing out. The gap between large players and niche specialists is narrowing fast, and warehouse-native solutions like Hightouch are pulling off a quiet disruption on the packaged CDP model. Necessary, increasingly available, and no longer differentiating on its own.

2. Marketing's own memory

Your MAP, CDP, and journey orchestration tools sit on years of campaign history, segment behavior, content engagement, and personalization rules including the ones that flopped. Most teams already use this data, but mostly for reporting. The shift to using it for context engineering is just beginning. Reporting answers what happened. Context engineering asks what the agent should know. Same raw material, different job. If you are looking for the fastest unlock available to your marketing team right now, it is sitting inside platforms you already pay for.

3. Brand voice and rules

Teams genuinely work hard on this one. They invest in style guides, voice systems, AI brand tools like Adobe's Brand Concierge and Writer. The intent is real. But making brand truly usable by an agent runs into practical challenges. Most enterprise brands are multi-voiced by design, with variations by audience, channel, market, and product line, which is a much bigger encoding job than uploading a style guide. And the brand AI tools that exist often sit in a separate workflow from the agent that actually needs them, so pulling the right voice variant into a decision in real time is still mostly custom work.

4. Collective decision intelligence

Now we are at the layer where Adobe and Salesforce are also starting to read the room, while almost no other vendors dare to look yet. This is the daily reasoning of your marketing team. The tradeoffs, the rejected alternatives, the editorial calls, the soft signals that shift a decision. The least captured of the four. The greenfield. And the moat. We will unpack why this matters, and what makes it so hard in the next section.

The Moat: Your Collective Decision Intelligence

Think about the decisions your marketing team makes on any given day.

1. The campaign approved over two alternatives that were just as defensible.
2. The cool-down window set at 90 days instead of 60 because someone in the room mentioned a pattern they saw last quarter.
3. The creative pulled at the last minute because the CMO read the market mood and said not today.
4. The AI-generated copy that passed every brand rule check and still got sent back because it did not meet the team's bar of taste.
5. The offer threshold tightened from $500 to $750 because of a hint the head of strategy dropped about competitive activity.

None of these lives in your CDP or your warehouse. It lives in the room where the decision was made. The line the CMO said before anyone wrote anything down. The alternative considered and rejected. The reasoning behind the specific number.

This is your collective decision intelligence. It is what actually makes your marketing organization good at what it does. An agent operating on structured data alone has zero access to it. And nobody owns it - the captured reasoning of the organization has no accountable owner.

This is exactly what could shrink your agent's value realization runway.

Adobe's orchestration push and Salesforce's Agentforce 360 are starting to point here. But neither has cracked the harder problem: how do you capture the reasoning of a specific marketing organization and make it available to an agent? The engine is theirs. The fuel is still yours to refine.

I made a related point after Adobe Summit: orchestration is the battleground. Context is the terrain underneath it. And this collective decision intelligence is the part of the terrain no one has finished mapping yet.

Three first moves

Capturing collective decision intelligence sounds ambitious, but the entry point is smaller than it seems.

Walk backward from a single decision the agent will make

Pick one. Next-best offer, service routing, the timing of an outreach. One decision gives you finite, fundable scope. A use case gives you an open-ended request and a stalled program.

Inventory the context that decision needs

Ask three questions. What is the minimum the agent must know to decide defensibly? What additional context would meaningfully change the answer? What does compliance or brand need to see before approving? Anything that fails all three is noise.

Start capturing collective decision intelligence the simplest possible way

Add a rationale field to your existing approval and decision workflows. When a campaign is approved, when a threshold is changed, when an alternative is rejected, capture the why. Not an essay. A sentence or two. Over time, that field becomes a corpus an agent can retrieve. The richer version is keeping the alternatives that were considered and rejected, with a line on what tipped the call. Taste, calibration, and judgment are learned by comparative example, not by rules.

Conclusion

The agents will get better. The platforms will keep converging. MarTech product vendors will keep shipping engines that look more alike every quarter. What stays yours is how your team actually decides. Capture it deliberately and your agents start to perform like you.

Who in a marketing organization should own this practice is a harder question than it first appears. I will come back to it.