Field Notes from OMR 2026

Canva opened with a slide showing screenshots of comments under AI-generated posts. Hollow. Repetitive. Soulless. The room laughed and winced at the same time, because we've all watched feeds get flooded with the same beige content that technically does the job and emotionally does nothing.

Claire Darley's argument was that the cure isn't slower AI use, it's more deliberate AI use. Canva's internal teams have started running a pre-publish gut check that has nothing to do with whether AI was involved: would a real person stop scrolling for this? If not, the speed didn't help.

The data they shared is a useful reference point. 46% of marketers said empathy and emotional intelligence are the capabilities AI can't match. 40% pointed to human imperfection, the rough edges that make things feel like a person made them. 39% each for brand intuition and relationship building. None of that lives in a vibe, it's the practical work of editing, choosing what to cut, holding a point of view.

Most of the marketing press has been calling 2026 the year of 'anti-AI marketing,' which is shorthand for the realisation that velocity has been completely democratized. Everyone produces faster, so what becomes scarce is taste, restraint, and cultural read. The Drum, Adweek, and Marketing Dive have all been arguing the same point through Q2: when AI tools become universally accessible, what separates great creative from average is the idea itself.

Martin Kihn brought the kind of data that actually reframes how you plan the next two years. He walked through a share-of-search projection where traditional SEO declines and GEO, Generative Engine Optimization, rises steadily from 2024 through 2030, with the crossover landing around the end of the decade. Gartner is forecasting a 25% drop in traditional search traffic by 2026. By 2028, AI-generated answers handle about 40% of what search used to own.

Translated into something we can act on: if a brand isn't surfacing in AI-generated answers, it's becoming invisible in a growing share of the discovery funnel. The newer learning that's been showing up across the marketing press since the event is that GEO isn't only about your own content. Generative engines lean heavily on third-party citations, so coverage in independent publications, analyst content, and well-structured PR matters more, not less, in an AI-first search world. Searchengineland and HubSpot have both been writing about this shift through April and May.

He also laid out how the marketing organisation reshapes around all of this. Strategy stays human: brand, identity, guardrails. Operations stays human too: orchestration, audience intelligence, compliance. Execution moves to agents: campaigns, content, personalisation, analytics. The whole structure flips from channel-first to audience-first. Agents pick the touchpoint, we define what the brand sounds like.

The deployment data was the part that felt least like a pitch. Salesforce reported a 119% surge in agent deployment across their customer base, 94% consumer satisfaction with AI agents, and a 300% conversion lift from agent-referred traffic. Whether or not the absolute numbers hold, the curve is real, and Gartner's projection that 40% of enterprise applications will include task-specific agents by year-end matches what we're already hearing in client conversations.

SAP's session went deep on enterprise architecture, and is the one I'd recommend listening back to. They opened with a distinction that sounds simple but quietly rewires how you think about every tool we use: Classical SaaS gives you an interface, you click and configure within the boundaries someone designed. SaaS plus AI gives you an intent layer, you describe what you want and the system figures out how to get there.

That shift changes who the user is. You no longer need someone who knows the system inside out. You need someone who knows the outcome they're after. SAP's framing is that we're already past the prompting era and into agentic territory, where everyone in an organisation becomes an orchestrator, directing several agents across tools rather than being a power user of one. This is exactly why Stan's Agent Marketing Setup for Nytro matters so much right now. It gives us a hands-on playground to figure out what kind of orchestrators we actually want to be, before we start advising clients on the same shift.

One slide stayed with me for a different reason. 'If an AI agent can rebuild the core workflow of your product in a weekend, what exactly are your customers paying a monthly subscription for?' SAP's answer was context. Process knowledge, governance, memory that persists, data loops that actually learn. The business context is the defensible bit, not the model.

This conversation has only intensified since the event. The press has taken to calling 2026 the year of the 'SaaSpocalypse,' with around two trillion dollars wiped off software stocks since early 2026 as enterprise buyers started questioning the per-seat license. SaaS isn't dying, the overall spend keeps rising, but the way it's bought and used is being redrawn. IDC is predicting that by 2028, 70% of software vendors will have moved off pure seat-based pricing toward consumption, outcomes, or capability-based models. Underneath all of that is SAP's simpler point: the value is moving from the application into the orchestration layer above it.

OpenAI came in hot. Nick Turley walked through how Codex has expanded way past 'tool that writes code' into something closer to a connected workspace. The April 2026 'Codex for almost everything' update layered in computer use on macOS, an in-app browser, gpt-image-1.5, persistent memory, scheduled automations, and 90+ plugins across Jira, the Microsoft 365 suite, Notion, Slack, Drive, and Gmail. It sits across your entire stack, not just inside a chat window.

The fireside chat with Narbeh Mirzaei (VP Engineering, Choco) and Niklas Harzheim was where the practical line got drawn. A chatbot answers your question. An agent triages your PRs, drafts your follow-ups, runs your reporting, flags when something breaks. It owns the task end to end, not the conversation.

Two concepts worth saving from Turley's keynote. Skills are reusable expertise packages, basically a bundle of instructions, resources, and optional scripts that lets an agent follow a workflow reliably across runs. Skills shipped for Codex in late 2025 and have been quietly catching on inside teams since. The other was Pets, agents with personality and persistence designed to feel collaborative rather than transactional. His broader argument was that we're heading toward AI prompting us, noticing something needs doing and proposing the action before we think to ask.

The Codex settings showed three permission tiers: default, auto-review, and full access. That governance dial is the question we'll all be answering inside our own organisations soon. How much autonomy do we hand an agent, and what review layers do we keep around it.

This session laid out the clearest architecture diagram of the whole event. Concentric rings expanding outward: LLMs at the centre, then AI Agents, then Agentic Systems, then Agentic Ecosystems. Each ring adds complexity. By the time you reach the outer layer you're dealing with orchestration frameworks, governance, multi-agent RAG, observability, security, and human-in-the-loop controls.

Insider One brought it down to ground level with their AI Decision Engine. Six agent types: Predictive, Conversational, Generative, Shopping, Support, Insights, all pulling from unified data and pushing to every channel: web, app, email, SMS, WhatsApp, push, search, retargeting. The orchestration layer picks the channel, the message variant, the timing, all based on behavioural signals and engagement likelihood.

OTTO showed what that looks like when it's running for real. Continuous learning loops feeding back into personalisation. Decision engines working at the moment of contact rather than relying on segments somebody built three weeks ago. The whole approach moves from campaign launches to lifecycle-level engagement.

The category has moved noticeably even in the days since the event. eGain just launched Agentic Studio with multi-agent orchestration aimed at end-to-end customer request resolution. Adobe unveiled CX Enterprise Coworker and Adobe Engagement Intelligence, a decisioning engine focused on customer lifetime value. PwC's number that 79% of organisations are already adopting AI agents, alongside Gartner's projection that 40% of enterprise applications will include task-specific agents by year-end, is the backdrop for all of this.

The session closed with something less technical. Leadership and human differentiation in the AI era. After all that architecture, the speakers came back to the same place everyone else did that week. AI scales execution. We hold direction.