Through Ethan Mollick I came across a poem that has not let go of me since. “Waiting for the Barbarians” by C.P. Cavafy, written in 1898. The scene: a city frozen in shock. The senate decides nothing. The emperor sits at the city gate, in full regalia, and waits. Everyone waits. For the barbarians.
But the barbarians do not come.
“What is to become of us now, without barbarians? Those people were a kind of solution.”
The more I think about it, the stronger the feeling gets that Cavafy described exactly how we deal with AI in 2026. A collective sense of expectation swinging somewhere between salvation and apocalypse, in whose shadow real decisions get postponed.
Between forecast and practice
Demis Hassabis, CEO of Google DeepMind, has narrowed his window for AGI from 5-10 years to 5. Ray Kurzweil sees 2029 as the year AI passes the Turing test. The METR benchmark shows that the autonomous working duration of AI systems on coding tasks doubles every 4 months. Anthropic’s Claude Opus 4.6 made a massive jump on that metric.
Soon we will no longer be optimising for eyeballs, but for API calls.
The numbers are probably correct. The conclusion many people draw from them probably is not.
My honest assessment, as someone who works with AI daily and has no academic distance from it: I do not believe in the big bang. I believe the singularity is already happening, at this moment, while you are reading this. It happens as a creeping process, one tool after another, one decision after another, one workflow after another. Until at some point we look back and notice that everything has changed, without there being a single moment you could point to.
That is exactly what makes the waiting so dangerous. Anyone waiting for the bang misses the wave.
What is being built while we debate
While the public discusses AGI timelines, the infrastructure for the next decade of marketing is being laid in the background. Quietly, technically, and with consequences that most decision-makers are not yet clear about.
In January 2026 Google published the Universal Commerce Protocol, an open standard that allows AI agents to handle the complete purchasing process autonomously: from product search through selection to checkout. No website required. The transaction happens headless, in the background, driven by the customer’s agent. In March, Google followed up with Gemini Embedding 2: text, images, video, audio and PDFs in a single index. When an agent no longer just reads the product description but also parses the product video, understands the claim in the ad spot and factors in the PDF data sheet, then every content format becomes a ranking signal.
In parallel, Alibaba showed with Qwen 3.5 what AI adoption looks like in practice: 430 million dollars to get users ordering bubble tea through the AI app. A sevenfold increase in active users. The model was never the product. Seamless integration into everyday life was the product.
For performance marketing that means: soon we will no longer be optimising for eyeballs, but for API calls. Feed quality and structured data will be the new UX design. If your client’s agent cannot read your product cleanly on a technical level, your ranking is worthless, no matter how good your ad texts are.
Three questions that belong on the table now
I say it the way I say it at SlopeLift too: these are not post-AGI questions. These are now questions.
Which processes can an agent handle autonomously? The anti-to-do list is the fastest way in. For us it was data cleaning that used to eat 2 hours a week. Now: 4 minutes. The “nobody wants to do it” category convinces sceptics immediately, because the ROI is visible the next day.
Which data do agents have to be able to read? Since Gemini Embedding 2, text alone is no longer enough. Video, audio, PDFs: everything becomes a signal. Anyone still treating their product feed as a purely SEO topic has slept through the tectonic shift.
Who carries responsibility for AI in the company? AI as a side project of the IT department is the surest route into the 56% without ROI. It takes someone with budget, mandate and operational understanding. Whether the title is CAIO or something else: the function has to exist.
The senate should get started
Cavafy wrote his city as a warning. A community that waits so intently for the great disruption that it forgets to act. The parallel to 2026 is hard to miss.
We debate AGI timelines while the infrastructure for Agentic Commerce is already in place. We wait for the moment when “AI changes everything”, while it has long been doing so: one embedding model after another, one protocol after another, one workflow after another.
In 2008 I debugged Excel formulas and was proud of my VLOOKUP. In 2014 we built rule-based scripts and called it a “workshop”. In 2026 agents run entire campaign management operations, and human value lies in the context, in the strategy, in the 30-second conversation with the client that changes everything. (And honestly: sometimes I miss the VLOOKUP. But only sometimes. And only after the second coffee.)
18 years, the same industry, a completely different job. And the arc is far from finished. We are at the beginning, not the end. The barbarians are not coming. The change was here the whole time.
Perhaps the senate should start legislating.
Back to Part 3 | To the overview
Florian Michael Fischer is Chief AI Officer at SlopeLift, a Vienna performance marketing agency. He has been in the industry since 2008 and works daily on translating AI into real marketing results.