Context now decides what your AI-enabled stack costs and whether it works. Martech Futurist | July 9, 2026
For two years the marketing conversation treated the model as the prize. Buy access to the best one, wire it into a workflow, wait for the returns. The research surfacing now moves the argument somewhere less comfortable. The model has become the assumed part. What separates spend that pays from spend that stalls is the context wrapped around it: the data that feeds an agent, the real-time signals it acts on, the machine-readable version of your brand that AI systems read when they choose, and the organizational capability to keep all of it current. Get the context wrong and a better model just reaches the wrong answer faster and bills you more to do it.
This is the layer I keep pointing to in the AI capability framework. Models are a mechanism. Memory and context are a layer that every capability depends on, and trust is the layer that decides whether anyone can act on the output. Three pieces from the past two weeks land on that layer from three directions, and one budget survey shows what happens to an organization that funds the model and neglects the context.
Three themes emerge
1. Context debt is the new cloud bill. Forrester's Frederic Giron reframes runaway AI spend as a problem of value, not price. Agentic workflows loop, and every turn re-feeds the context window, so a single agent consumes several times the tokens of a chat call and a multi-agent system consumes far more again. Part of that bill is a runtime tax the company pays because its own knowledge is not machine-readable, forcing agents to reconstruct meaning on every call. Capping the spend leaves the debt in place. Paying it down means engineering the context so each token buys a better outcome. This is compute economics as a force acting on the whole system, and it lands directly on the memory-and-context layer.
2. Stale context is worse than absent context once agents act at machine speed. An agent does not pause to sanity-check. It acts, and any error propagates through every downstream agent before a human notices. Forrester's Mike Gualtieri argues that timeliness is part of data quality, not a separate concern, and that agents need a live picture of the business rather than a snapshot from last quarter. Gartner's spend data shows the organizational version of the same failure: CMOs are pouring budget into the channels AI optimizes most easily and pulling it out of loyalty and retention, which risks tuning the system toward whatever is measurable this week instead of what builds customer value over years. Relationship over transaction is the second thing to erode when context goes stale.
3. Machines now read your brand, and unreadable or wrong context loses the sale before a human sees it. The customer-facing surface has its own context problem. AI systems increasingly sit between a brand and the buyer, assembling recommendations from whatever they can read and trust about you. If your product data is thin, your claims inconsistent, or your public content ambiguous, the model fills the gap with something else. The brand becomes a body of machine-read context, and the accuracy of that context now shapes demand.
Featured insights
Forrester — "Your AI Bill Is A Context Problem" (forrester.com, June 10, 2026)
Frederic Giron and colleagues open with the AI equivalent of cloud bill shock: Uber exhausted its 2026 AI budget in four months and capped engineers at $1,500 a month, and ServiceNow burned through its coding budget early in the year. Giron argues the cap treats a value problem as a price problem, citing Uber's own COO that the link between the spend and anything a customer feels is "not there yet." The real driver is context debt, billed by the token, because agentic workflows loop and re-feed the context window, and because knowledge that is not machine-readable forces agents to rebuild it on every call. He proposes ContextOps as a standing discipline that keeps the operating context current as the business shifts and ties each token to the outcome it produces. Takeaway: Before you negotiate a better rate on tokens, ask which tokens carried real business context and which paid to reconstruct meaning your systems should have supplied. That attribution, not the cap, is where the savings live.
Forrester — "AI Agents Need Real-Time Context: Data Streaming Is How You Are Going To Get It"
(forrester.com, June 18, 2026) Mike Gualtieri makes the case that agents act at digital speed on whatever context they hold, so a poor decision by one agent compounds through every agent downstream before anyone intervenes. Governance matters, and so does timeliness: garbage-in-garbage-out now includes data that is simply late. His answer is a streaming data platform that connects enterprise sources, enriches events on the fly, and detects patterns as they emerge, so a retention agent sees a cart-abandonment event while the customer is still deciding and a fraud agent judges a payment in milliseconds with the customer's history joined in. He frames this as the real-time nervous system that autonomous agents require. Takeaway: Pick one agent you plan to let act without a human in the loop and trace the freshness of every input it reads. If any of it is a nightly batch, the agent is confidently acting on a version of the business that no longer exists.
Harvard Business Review — "How Do You Market to an AI Customer?"
(hbr.org, June 11, 2026) Kartik Hosanagar of Wharton describes moving from a query to a purchase inside a single chat, never opening a retailer's site. His warning to marketers is that AI agents are becoming the gatekeepers of discovery and selection, and brands that ignore this risk becoming back-end fulfillment for systems that own the customer relationship. The strategic questions change accordingly: what does the AI reward, what does it ignore, and what does it trust when it assembles an answer about your category. The brand's machine-readable context, its product data, claims, and public signals, becomes the material the model reasons over, and waiting for consumer behavior to fully shift means answering those questions after the advantage has moved. Takeaway: Run a query in your category through the tools your buyers now use and read the answer as a competitor's AI would read your brand. What the model gets wrong about you is a context gap you can close this quarter.
Gartner — "Gartner Marketing Survey Finds Awareness and Conversion Account for 62.6% of Total Media Spend"
(gartner.com, June 8, 2026) Gartner's 2026 CMO Spend data shows digital media now exceeding two-thirds of total media investment, with awareness and conversion claiming 62.6% of media spend while loyalty and retention fell 29% to under 15%. AI is a stated driver, as CMOs shift budget toward channels they can optimize with it. Two findings complicate the story. Labor's share of the marketing budget rose from 21.9% to 24.5%, which says AI value still depends on people and execution. And the most AI-mature organizations allocate more to loyalty and retention, not less, suggesting that less mature teams are over-indexing on what is easy to tune. The capability to direct AI toward durable customer value, rather than toward whatever the model optimizes fastest, is itself part of the context that determines returns. Takeaway: If your AI is steering budget toward the channels it optimizes best, check whether it is also steering you away from the customers worth keeping. The maturity signal here is spending against retention while everyone else chases acquisition.
Two operators I have interviewed on The Agile Brand make the underlying point more plainly than any framework. It reminds me of what Marcus Fontoura of Microsoft said about where these programs actually break: "There is no AI without data," and the platform teams most often get wrong is the one that defines what a customer or a transaction even is. And on the customer-facing side, it echoes Romain Fouache of Akeneo, who put the stakes in commercial terms: if you do not know what your products are, the model will not invent it for you, and "if you get it wrong once, you will kill the trust of your consumers." Both are describing context, the internal kind that grounds an agent and the external kind that a machine reads about your brand.
Key takeaways
The constraint on AI value has moved down a layer, from acquiring the model to engineering and maintaining the context it runs on. Treat context as something you operate continuously, the way Giron describes ContextOps, rather than something you build once and leave to drift.
Timeliness is now part of accuracy. An agent acting on stale context at machine speed does more damage than one that waits, because the error compounds across every agent it touches before a human can catch it.
Your brand has become machine-readable context. The product data, claims, and public signals that AI systems read when they recommend now shape demand as much as the message a human sees, and errors in that context cost you the shortlist before the buyer arrives.
AI value still depends on people and direction. Rising labor share and the retention discipline of the most mature organizations both point the same way: the capability to aim AI at durable customer value is part of the context that separates spend that pays from spend that stalls.