This week the agenda was set not by model launches but by the places where artificial intelligence touches sales and operations: a bank in Türkiye moved its product information straight into ChatGPT, OpenAI brought agents down to API level, and on the payments side the first infrastructure for agentic commerce appeared. As the DijitalPi team we have commented on this week's five critical headlines with their equivalent on the campaign side, and collected the remaining developments in a radar section.
Garanti BBVA Moved Its Banking Information Into ChatGPT
Garanti BBVA has made up-to-date information on its loan and banking services accessible through ChatGPT. Customers can now ask about product terms inside the assistant, without visiting the bank's website.
DijitalPi's Take
The assistant is no longer a source, it is a channel — and the first major move has come from Türkiye. Ranking first in search results and being cited correctly in an assistant's answer are not the same thing; the second one requires a brand to publish its own data in a machine-readable form. In the accounts we manage, the answer to brand questions asked of assistants often comes from out-of-date third-party pages; when pricing, terms and campaign details go stale, it is the customer service team that pays the bill.
Our recommendation: Write down the first 20 questions people ask about your brand (price, delivery, returns, appointments, campaign terms) and put each one to four assistants individually. For every wrong answer, connect the correct one to a page on your site that carries a clear heading, structured data and a visible update date.
OpenAI Introduced the Agents API for Long-Running Cloud Agents
OpenAI has announced the Agents API for building cloud agents that manage long-running sessions and tool use.
DijitalPi's Take
A chatbot answers a single question; an agent follows a job from beginning to end. The piece that makes the difference is context surviving between sessions. In marketing the first payoff is not creative work but repeated operations: scanning a product feed for missing fields, pulling negative keywords out of a search term report, compiling the monthly report. On our side we run this kind of work through the workflows we call engines, and what we see is this: the most expensive mistake is putting an agent in charge of a creative decision, while the fastest gain comes from tasks whose output can be verified.
Our recommendation: List the three jobs your team repeats every week, pick the one whose output has a single correct answer, and set your first agent up there. Run it double-checked for at least two weeks before removing the human approval step.
ChatGPT Work Gains a Data Agent That Builds Dashboards From Plain Language
OpenAI has added a Data agent to ChatGPT Work that analyses company data and builds interactive dashboards from natural language requests.
DijitalPi's Take
What changes is who builds the dashboard: the report request no longer waits in a queue, the person asking produces the answer themselves. The risk here is not technical but definitional. "Conversion" means add-to-cart on one team and payment on another; if it is unclear whether the attribution window is 7 or 30 days, the wrong answer is now produced at scale. Most of the reporting disputes we see in the field come not from a data error but from two different definitions of the same metric.
Our recommendation: Before you open the dashboard, write a one-page metric dictionary — definitions for conversion, ROAS (return on ad spend), channel and attribution window — and grant data access by role. Without that dictionary, the first dashboard will produce arguments too.
Alipay Is Planning a Wallet Agent to Handle AI Payments
Alipay is working on a Vibe Pay-powered wallet agent designed to manage trusted artificial intelligence payments.
DijitalPi's Take
Once the agentic commerce debate reaches the payment layer, it stops being an experiment. If an agent makes the purchase decision on the user's behalf, a product page has to serve two readers at once: a human and a machine. Price, stock, delivery time and return terms may be written on the page, but if they are missing from the structured data the agent finishes its comparison without ever putting you on the list. This will reach Türkiye through marketplace and payment integrations in the near term; the preparation, however, is done with today's product data.
Our recommendation: Fill in the schema.org Product/Offer fields on your product pages (price, currency, availability, returns) and check that the information in your feed matches the information on the page. Where the two sources disagree, the agent decides for you which one is right.
Universal Music and ElevenLabs Are Building a Licensed Remix Platform
Universal Music Group and ElevenLabs are developing a platform that enables AI-assisted remixes built on licensed songs.
DijitalPi's Take
Licensing infrastructure is replacing the copyright fight, and that means the risk finally becomes priceable for every brand that uses music. On the social video side the most common problem is not detection but scope: a track from a platform library is cleared for an organic post but not when the same content runs as an ad. A licensed AI remix could simplify that split — but only where the licence text states that advertising use is covered as well.
Our recommendation: List the music sources used in the videos you published over the last three months and answer one question for each: can this track run as an ad? Move anything with an unclear answer to a licensed source, and evaluate new platforms with the same question.
Also On Our Radar
- According to Reuters, Nvidia is considering investing up to 10 billion dollars in Anthropic's IPO.
- Oracle grew its cloud revenue by 120%; in the same days came reports of the cash squeeze caused by AI data centres and a growing redundancy budget.
- Microsoft plans to add 26 gigawatts of data centre capacity; the US administration is also loosening environmental rules to speed up data centre construction.
- Google released a desktop app for Windows 10 and above that gives access to Gemini from the active window.
- DeepSeek announced V4.1-Flash with 552 billion parameters and a 1 million token context; Moonshot AI is targeting 2 billion dollars in annual revenue.
- In the legal vertical Harvey raised 550 million dollars at a 15.5 billion dollar valuation — appetite for sector-specific AI continues.
- Cursor opened up Projects, which holds context for months and distributes tasks across thousands of sub-agents.
- On the safety front: Anthropic intervened in five cases where Claude was used in biological weapons research; CEO Dario Amodei proposed a three-stage slowdown plan, and Sam Altman said slowing advanced development is being considered.
- Google DeepMind built an atlas mapping the effect of 9 billion possible DNA variants; Maven Robotics came out of stealth with a 100 million dollar Series A.
The AI Marketing Newsletter is the column where the DijitalPi team comments each week on the standout developments from the world of artificial intelligence and marketing for brands in Türkiye.
