Someone Else Wrote the Rules Your Stack Runs On. Martech Futurist | August 24, 2026
The August 20 edition on this blog argued that AI capabilities are commoditizing while the conditions underneath them appreciate: data infrastructure, identity resolution, trust architecture. Four developments over the past week move that argument one step forward. Other organizations now set those conditions, and they set them differently by jurisdiction, by vendor, and by protocol body. Your leverage here is contractual. Marketing teams that read the terms early get to shape what their AI can do; teams that read them at renewal inherit the answer.
Here is the practical consequence. The same AI product delivers different capability depending on terms you did not write. Planning for Q4 means reading those terms and pricing the difference.
Read the European ChatGPT ad terms before you brief the creative
OpenAI turns on ChatGPT advertising across 31 European markets today, six months after the US pilot began in February. Germany, France, Spain, Italy, Poland, Sweden, Norway, Denmark, Ireland, and the Netherlands all go live. Ads appear for Free and Go accounts. Paid and organizational plans stay clear.
The targeting instrument differs from the one US advertisers have used since February. In the European Economic Area and Switzerland, OpenAI launches without personalization. Ads key off the current conversation, approximate location, device type, time of day, and language. Chat history, memory, and past ad interactions stay out of the targeting model. OpenAI Ireland Limited acts as data controller, which puts the Irish data protection authority in the oversight seat. Advertisers receive aggregated performance data and never receive conversations.
That is one product with two capability profiles, separated by a regulatory boundary. A brand that briefs a European campaign against US-style intent assumptions will build its forecast on an instrument that does not exist in those markets.
The advertiser tooling lags the consumer rollout. Ads Manager still reads "Coming Soon" across all 31 markets, so launch bookings route through OpenAI's ads solutions team, agency partners, and technology partners, with self-serve access promised later this quarter. Roughly 20% of ChatGPT queries carry direct commercial intent by OpenAI's own account, and cost-per-click now accounts for most ad spending on the platform.
It reminds me of what Brit Starr, Chief Marketing Officer at CreatorIQ, said when I interviewed her on The Agile Brand podcast: "You have to understand the world you're living in today, and it is a world that is driven largely beyond the control and bounds of the brands themselves."
What to do this week: Split your ChatGPT ads forecast by regulatory zone. Build two sets of match-rate assumptions and two sets of creative briefs. Ask your agency which of the 31 markets they have confirmed against OpenAI's advertiser documentation, since OpenAI named only nine of them publicly.
Put agent identity into your vendor contracts before the agent count climbs
Google transferred the Agent2Agent protocol to the Agentic AI Foundation, the Linux Foundation body that already hosts Anthropic's Model Context Protocol, Block's goose, and OpenAI's AGENTS.md. Axios reported the move on August 17 and the transfer completed on August 20. MCP standardizes how an agent reaches tools and data. A2A standardizes how independent agents find each other, delegate work, and exchange results.
The membership arithmetic tells you how fast this consolidated. AAIF launched in December 2025 with fewer than 40 members. It now counts more than 250, including AWS, Anthropic, Bloomberg, Block, Cloudflare, Google, Microsoft, OpenAI, and Shopify.
The detail that matters for marketing operations sits in the A2A 1.0 release: multitenancy, protocol negotiation, and signed agent cards. A signed agent card is an identity credential. It answers the question of which agent is calling, on whose behalf, and with what verified claims attached. In my AI capability framework, that is the Identity and Permissions layer moving from a thing each vendor improvises to a thing an industry body specifies.
Two consequences follow for buyers. Integration costs fall, which weakens platform lock-in and lets you select models on price and performance. Your permission model also stops being yours alone. Once your CDP agent and your ad platform agent exchange signed cards, the trust decision runs on a specification written in a room you do not sit in.
It reminds me of what Andy Baker, CEO and Founder at Sesimi, said when I interviewed him on The Agile Brand podcast: "Just because you have access to those tools, does not necessarily mean that you have therefore the right to be constantly making and changing content."
That distinction between access and authority is exactly what a signed agent card encodes. Our identity resolution entry in The Agile Brand Guide covers the customer-side version of the same discipline: track consent by identifier, propagate opt-outs across edges before sync, and enforce minimum confidence thresholds for sensitive activations. The agent-side version asks the same questions about machine actors.
What to do this week: Add three lines to your next martech renewal. Which agent protocols does the vendor support. What identity claims does the vendor's agent present when it calls your systems. What audit record does your team receive when an agent acts.
Make data retention a priced line item in AI procurement
OpenAI announced on August 19 that it is testing Private Safety Processing with early customers, a technique the company says detects misuse patterns across related interactions while preserving zero data retention. Customer data stays on customer-controlled infrastructure or sits with OpenAI under customer-held encryption keys. OpenAI plans a broader rollout and a technical white paper in September.
Anthropic moved the other way. The company now requires 30-day data retention for business customers using its most capable models, and stated in a risk report that it expects the decision to be unpopular and to carry real business risk, and that it considers the retention necessary to detect sophisticated attacks that span multiple requests.
Aleah Houze, Head of Product Policy at OpenAI, described the underlying problem in a press briefing: risks emerge across multiple interactions over time, not inside a single prompt and response pair. Both companies agree on the diagnosis. They disagree on whether detecting it requires holding customer data.
For marketing leaders, this converts a security posture into a procurement variable with a price attached. Work an example.
A financial services marketing team has published a zero-retention commitment to its customers. The team runs four model vendors. One vendor moves its top-tier model to 30-day retention, and that model currently carries 35% of the team's personalization inference volume. Three paths open. Re-paper the customer commitment and the data processing agreement, which costs legal cycles and consumes trust the brand spent years building. Route that 35% to a lower-tier model from the same vendor, which costs accuracy on the highest-value segment. Route it to a competing vendor, which costs migration effort and integration testing. Every path carries a number. None of the three is a technical decision.
Price all three before your renewal date, and price them again when the September white paper lands.
What to do this week: Pull the retention clause from every AI vendor agreement your marketing organization holds. Map each clause against the promises on your own privacy page. Where the two disagree, escalate now.
Budget for the contract surface, because it scales with the agent count
Salesforce's Agentic Enterprise Index, drawn from production activity across 400 businesses plus a survey of nearly 5,000 people, reports that the average number of AI agents deployed per organization climbed from five in early 2025 to 13 by April 2026. Agent creation time fell 53%. Employee sessions tripled. Among organizations in the dataset, seven in ten customer service sessions now resolve autonomously, with escalation rates holding steady. Retail, travel, financial services, and the public sector all posted sharp increases.
Spending concentrated at the same time. Ramp data shows the top 1% of US businesses spent a median $7,400 per employee on AI in July. The median company spent $11.95.
Read those two datasets together and the operational picture sharpens. Agent counts roughly tripled in fifteen months. Every one of those agents touches customer data under some vendor's terms, presents some identity to some downstream system, and creates some retention obligation. The contract surface grew at the same rate as the agent count, and almost no marketing organization staffed for that.
This is where Operations capability earns its place as a horizontal. Agent governance runs across marketing, legal, procurement, and security simultaneously. No single function owns it, and treating it as a marketing technology problem guarantees that the terms get accepted by whoever signs fastest.
Consent architecture belongs in the same conversation, which is why the consent management platform sits at the center of most of these decisions. A CMP configured for a browser and a click will not produce a valid signal when an agent handles the session.
What to do this week: Count your production agents. Then count the vendor agreements governing them. If the second number is smaller than the first, you have agents operating under terms nobody read.
Featured Insights
Digiday | OpenAI takes ChatGPT ads into Europe | August 19, 2026
OpenAI switches on ChatGPT advertising in 31 European markets today, reaching an advertising footprint in every major global region inside a year. GDPR forced a different targeting model in the EEA and Switzerland, where the platform launches without personalization and keys ads to the current conversation, coarse location, device type, time of day, and language.
Practitioner takeaway: Forecast European ChatGPT ad performance on a separate model from your US numbers. The targeting instrument is materially different, and importing US benchmarks will overstate expected efficiency.
Axios | Google's A2A protocol gets a new home | August 17, 2026
Google moved Agent2Agent under the Agentic AI Foundation, placing agent-to-agent communication alongside MCP's agent-to-tool integration under one Linux Foundation body. The foundation grew from fewer than 40 members at its December 2025 launch to more than 250. A2A 1.0 adds multitenancy, protocol negotiation, and signed agent cards for identity verification.
Practitioner takeaway: Treat protocol support as a procurement criterion at your next renewal. Vendors aligned to AAIF-governed standards give you a credible exit path; vendors running proprietary agent interfaces do not.
Axios | OpenAI previews zero-retention safety system as Anthropic requires data logs | August 19, 2026
OpenAI is testing Private Safety Processing to detect cross-interaction misuse while preserving zero data retention for eligible enterprise and API customers. Anthropic requires 30-day retention on its most capable models and has publicly acknowledged the commercial cost of that choice. A broader OpenAI rollout and technical white paper are scheduled for September.
My takeaway: Retention terms now differentiate vendors as sharply as benchmark scores. Add a retention column to your model evaluation matrix and populate it before the September documentation lands.
MarketingProfs | AI Update, August 21, 2026 | August 21, 2026
The roundup carries the Salesforce Agentic Enterprise Index data on agent deployment growth and the Ramp figures on AI spending concentration, alongside coverage of frontier model price competition and the IAB's second AI disclosure framework. Editor's note discloses that ChatGPT assisted in compiling the issue.
My takeaway: The gap between $7,400 and $11.95 per employee describes two different operating realities. Benchmark your own AI spend against companies with your agent count, not against the market median.
Key Takeaways
Capability now varies by jurisdiction inside a single product. ChatGPT advertising delivers one targeting instrument in the US and a materially narrower one in the EEA. Plan and forecast them separately.
Agent identity became an industry specification this week. Signed agent cards under A2A 1.0 move the Identity and Permissions layer out of vendor improvisation. Get protocol support and identity claims into your renewal language.
Retention policy is now a competitive differentiator between frontier vendors. OpenAI and Anthropic have taken opposite positions on the same security problem. Price your options before your commitments and your contracts diverge.
Your contract surface grew as fast as your agent count. Agents per organization went from five to 13 in fifteen months. Staff and budget the governance work across marketing, legal, procurement, and security together.
Closing note
I have spent a good part of my career doing martech, marketing operations, and change management work inside Fortune 500 organizations, and one pattern shows up in almost every engagement. The meeting that decides what marketing can actually do is rarely the platform demo. It is the session with procurement and legal three weeks later, where someone reads the data processing terms out loud and a capability the team already budgeted for turns out to be unavailable in half the markets on the plan.
Marketing leaders who sit in that second meeting get to shape the answer. Marketing leaders who skip it inherit whatever the terms happen to say.
Both meetings are on your calendar this quarter. Take the second one.