If your mental model of AI in marketing is still ChatGPT writes a blog post, you are roughly three years behind the live frontier. Agentic AI is the next layer of the stack, and it has already moved out of slideware and into production at brands you compete with. This guide gives you a clearer mental model, an honest take on jobs, and one concrete action.
Stitch leads the IAB New Zealand AI Working Group on agentic systems, co-chairs the Marketing Association's Digital and AI group, and ships agentic AI for clients including Serko, Skills Group and Bayleys. The frames below are the ones we use with our own clients.
Three categories that matter
There are three useful buckets. Generative AI produces output when prompted — ChatGPT's web interface is the canonical example. An AI assistant or copilot sits next to a human and suggests next moves — Claude CoWork, GitHub Copilot. An AI agent perceives, plans, acts, observes and loops on its own.
Autonomy is the unlock. Once an agent can take real actions in real systems without a human in every loop, the cost curve of marketing operations bends.
Case in point — Coca-Cola + Soma
Coca-Cola handed an entire Saudi Arabia campaign to an AI agent called Soma, built by Toronto's Cluep. The agent identified audiences by scanning public posts on TikTok, Meta, Pinterest and Reddit and matching them to mobile IDs via ad exchanges. Generative AI built the creative on the fly — burger fans saw Coke + burger creative, spicy-food fans saw "refreshing relief" creative. The agent then ran 24/7 for two months through Coke's existing Burger King and AMC partnerships.
The result: ~8 million autonomous actions, 828,000 personalised coupon ads, 20,000+ click-throughs — and zero marketing team members in the loop. The agent ran the campaign while Coke's marketers slept.
The infrastructure — MCP is the USB-C moment
In November 2024 Anthropic released Model Context Protocol — a universal standard that lets any agent connect to any tool. By March 2026 the ecosystem hit ~97M monthly SDK downloads, 5,800+ community MCP servers and 16,000+ public servers.
Salesforce, HubSpot, Adobe, Snowflake, Webflow, Meta and Google all ship MCP servers. An agent inside ChatGPT can now spin up an Amazon ad campaign in a single prompt. The plumbing is standard — your stack does not need to be bespoke.
In May 2026 Robinhood shipped Agentic Trading. Paste one MCP URL into your agent config, fund a ringfenced "agentic account", and your AI agent can place real stock trades on a real brokerage account, 24/7, for $0. The MCP era is not a B2B abstraction anymore. It is in retail finance, today.
Reality check — adoption is not value
79% of enterprises claim AI agent adoption. Only 11% are actually running agents in live environments. 34% of marketing teams report running at least one autonomous agent — more than double Q4 2024.
We are roughly where e-commerce was in 1999: the winners are obvious in retrospect, but most companies are still arguing about whether to ship. Nearly 8 in 10 organisations report no significant bottom-line gains from AI generally (McKinsey, 2026). Adoption is not the same as value. Operating model, data foundation and measurement decide who actually captures the upside.
Content — velocity is no longer the moat
One Forrester-cited content programme lost 70% of its organic traffic in 9 months (from 24M to 6-7M visits) while growing revenue 22% YoY. Traffic and outcomes have decoupled. Content velocity is a commodity. What scales now is original data, taste and citation share inside the LLMs.
End-to-end content agents — Writer.com (Vanguard, L'Oreal, Mars; Forrester-validated 333% three-year ROI), Jasper (100+ marketing agents, used by Spark NZ's marketing team), Adobe's Content Production Agent in beta — are the new baseline, not the cutting edge.
SEO becomes GEO
Google AI Overviews now appear on 48% of all queries (February 2026), reaching 2 billion monthly users. An ISB + Carnegie Mellon randomised field experiment found AI Overviews cut organic clicks by 38%. 72% of searches now end with no click.
Only ~6.82% of ChatGPT-cited results overlap with Google's top 10. What ranks on Google is largely not what gets cited by ChatGPT, Perplexity or Gemini. Generative Engine Optimisation (GEO / AEO) is a new discipline, not a tag on the old one. If 72% of searches end with no click, your brand-building playbook for 2030 has to be built for AI answer surfaces — not the blue links.
Social — AI eats every layer
TikTok Symphony (URL-to-video, digital avatars, image-to-video) is delivering documented results: Meoky reported 1.8x purchases and +13% ROAS; YOOX reported -15% cost-per-add-to-cart. 76% of TikTok advertisers plan to increase use of AI creative tools.
AI influencers earn real money. Aitana Lopez, a fully synthetic Barcelona-based persona, has ~370K Instagram followers and earns $6,000-$20,000/month. Brands signed her before realising she was not real — and proceeded anyway.
The CMO is being unbundled
49% of Fortune 500 marketing leaders still hold the "CMO" title — down from 55% in 2024. One in three F500 companies now has no enterprise marketing leader at all. Just 15% of CEOs believe their CMO is AI-savvy. Gartner projects "lack of AI literacy" will rank in the top three reasons CMOs are replaced by 2027.
The CMO of 2030 is more likely called a Chief Growth Officer — owning P&L, supervising agentic squads, with CFO-grade financial fluency and an explicit AI ethics remit.
The job market is already pricing this in
The PwC 2026 AI Jobs Barometer puts the global AI wage premium at +56%, up from 25% a year earlier. The Anthropic Economic Index (February 2026) found API workflows for business sales and outreach roughly doubled between November 2025 and February 2026.
The simplest split: at risk — junior copywriting, manual reporting, template-driven creative; enhanced — brand strategy, customer research, narrative, taste; emerging — AI Marketing Manager (US median ~$136K), Agent Orchestration Lead, GEO Lead, Marketing Data Engineer.
Agentic AI is already live across Aotearoa
Stitch ships agentic AI in production for New Zealand brands today. Bayleys Real Estate use agentic AI for lead qualification and listing enrichment. Skills Group runs a production chat assistant across four divisions on Zapier with a RAG layer trained on client content (live since April 2026). Stitch is the lead Braze partner on Serko AI, the agentic travel assistant whose US beta launched on 9 May 2026. Spark New Zealand runs modular, purpose-built agents that audit international roaming fees, fetch usage logs, cross-reference travel data and assemble resolutions for service reps in seconds.
The thing that surprised even me
A December 2025 INSEAD study found AI-generated ads won against human-made ads 59.1% of the time — and AI ads tested at a 21-point premium on certain creative quality measures while landing 29.4% cheaper to produce. The point is not that AI is taking marketing jobs. The point is that AI is taking marketing tasks. Your job is to make sure the value you create is on the task list AI is bad at.
Where this goes by 2030
By end-2026, Gartner projects 40% of enterprise applications will include task-specific AI agents (up from under 5% in 2025). By 2028-2030, McKinsey expects agentic AI to power up to two-thirds of marketing activities, accelerate campaign creation 10-15x and drive 10-30% revenue growth for organisations that operationalise it.
A 24-year-old marketer today can do work that took a team of 20 in 2015. The real question is whether you compress 10 years of career growth into 2 — and whether your CV in 2030 reads "I used AI" or "I orchestrated agents".
What to do this quarter
- Build one real agent. Pick one platform — Make, n8n, Viktor, Gumloop, Lindy, Zapier Agents — and ship a working end-to-end agent against a real workflow. This alone puts you in the top decile.
- Pick a vertical and own GEO for it. Track citation share inside ChatGPT, Perplexity and Google AI Overviews, not just blue-link rankings.
- Practise the human-only skills harder. Active listening, reading comprehension, customer empathy, narrative, judgement and taste are more valuable, not less.
- Audit your marketing operating model. Map every workflow against the agentic stack and decide what humans should orchestrate vs. perform.
- Pick the right partner. Most agencies are still doing slideware. Pick one shipping production agents you can reference.
FAQ
What is agentic AI, in plain English?
An AI agent perceives, plans, acts, observes and loops — autonomously. That is the unlock. Generative AI produces output when prompted. A copilot suggests next moves alongside a human. An agent runs the workflow itself across tools and time, with autonomy as the defining trait.
What is MCP and why does it matter for marketers?
Model Context Protocol, released by Anthropic in November 2024, is effectively USB-C for AI agents. It is a universal standard for connecting any agent to any tool. By March 2026 the ecosystem hit ~97M monthly SDK downloads, 5,800+ community servers and 16,000+ public servers. Salesforce, HubSpot, Adobe, Snowflake, Webflow, Meta and Google all ship MCP servers, so an agent in ChatGPT can spin up an Amazon ad campaign in a single prompt. The plumbing is now standard.
Are real brands actually running agentic campaigns?
Yes. Coca-Cola handed an entire Saudi Arabia campaign to an AI agent called Soma. Over two months it took ~8 million autonomous actions, delivered 828,000 personalised coupon ads and 20,000+ click-throughs, with zero marketing team members in the loop. Robinhood shipped Agentic Trading in May 2026, letting an AI agent place real stock trades on a real brokerage account via MCP. This is no longer slideware.
Which marketing roles are most at risk?
Tasks, not titles, are being automated first. Junior copywriting, routine campaign ops, manual reporting and template-driven creative are the most exposed. Roles centred on judgement, taste, narrative, customer empathy and orchestration of agentic squads are becoming more valuable. The CMO title is also being unbundled — 49% of Fortune 500 marketing leaders still hold the CMO title (down from 55% in 2024) and one in three F500 companies has no enterprise marketing leader at all.
What should I do this quarter?
Pick one agent platform — Make, n8n, Viktor, Gumloop, Lindy, Zapier Agents — and build one real working agent end to end. Document it. Then pick a vertical and own GEO (generative engine optimisation) for it: AI Overviews already appear on 48% of Google queries and cut organic clicks by ~38%, and only ~6.82% of ChatGPT-cited results overlap with Google's top 10. Citation share inside the LLMs is the new ranking.