Marketing with AI Weekly Roundup: Salesforce Retires Journey Builder for Agents, HubSpot's Breeze Agent Moves Into Marketing, and Google Ships a Free Brand-Building Agent

Edition #37 - Sunday 7th June 2026

Jun 07, 2026

Welcome to issue #37 of the Marketing with AI Weekly Roundup. Published every Sunday, it shares the top stories, playbooks, data and quick hits from the world of Marketing with AI over the last 7 days.

Subscribe to receive the week's most important AI marketing stories direct to your inbox.

Claude for Marketers Masterclass is now live.

If you’ve been using Claude as a chat window and wondering why it doesn’t quite deliver - this is why.

The course teaches you how to run Claude as a complete marketing system, built from 30+ live workshops attended by 300+ marketers.

10+ hours of pause-and-do content. 9 bonus assets. One new module every month through 2026 — so the course grows as Claude does.

Trusted by marketers at Adobe, JPMorgan Chase and the Chartered Institute of Marketing. Now an Inc Magazine columnist.

Early-bird pricing available until Monday 15th June at 8am.

View the course →

Back to this weeks AI Marketing Roundup…

This was the week the agent stopped advising marketers and started doing the work.

In the space of three days, the three platforms most B2B marketing teams actually run on — Salesforce, HubSpot and Google — all shipped agents that build emails, optimise live journeys, hold customer conversations and stand up entire brands.

Salesforce went furthest: it’s retiring Journey Builder, the backbone of Marketing Cloud automation for over a decade, and replacing it with agent-driven Flow orchestration.

The throughline for you: the question is shifting from “which AI tool should I use?” to “which of my workflows should run on autopilot — and where do I keep a human in the loop?”

Below: the three platform moves you need to act on, two steal-this Claude playbooks (with the exact prompts and steps), the new tools worth your time, and the AI-search data that should change how you think about visibility.

Let’s get into it.

🔥 Top Stories

1. Salesforce rebuilds Marketing Cloud around agents — and Journey Builder is on the way out

At Salesforce Connections (3–4 June, Chicago), Salesforce launched Agentforce for Marketing: AI agents that don’t just recommend, they act — building email variants, optimising journeys and adapting live campaigns to real-time performance signals. The bigger structural news sits underneath it: Flow-based orchestration is replacing Journey Builder, the tool that has run Salesforce Marketing Cloud (SFMC) automation for more than ten years. Marketing Cloud has also been rebuilt natively on Salesforce’s core platform, so the CRM and the marketing engine finally share one data model. Einstein’s email AI now adds predictive subject lines, send-time optimisation and CRM-driven dynamic personalisation.

What to do this week: If you’re on SFMC, this is the biggest change to your stack in a decade — start scoping the Journey Builder → Flow migration now rather than when it’s forced on you. Just as important: decide, in writing, where you’ll let agents act autonomously (subject-line tests, send-time tuning) versus where they stay in an approval loop (offers, pricing, anything brand-sensitive). The teams who define those guardrails early will trust the agents faster than those who wait.

2. HubSpot turns its Breeze agent loose on marketing — not just support

From 2 June, HubSpot’s Breeze Customer Agent is available to all Pro and Enterprise customers across the Marketing, Sales, Service, Content and Operations Hubs — not just Service, where it started. It autonomously resolves more than half of customer conversations, pulls live CRM details (renewal dates, order status), takes real actions like password resets and human handoffs, and learns from your knowledge base, site content, PDFs and blog posts. It sets up in minutes and works across email, chat, WhatsApp and Messenger.

What to do this week: The moment that agent touches top-of-funnel and nurture conversations, marketing owns part of its brain — its tone, its facts, its answers. Audit what content the agent will draw on (knowledge base, blog, PDFs) and fix anything outdated or off-brand before it starts speaking for you. Then decide which marketing conversations it should actually handle versus route to a human. Treat the knowledge base like ad copy: if it’s wrong, it’s wrong at scale.

3. Google’s Pomelli goes agentic — a free agent that builds your brand identity, brand book and website

Google Labs expanded Pomelli, its free AI marketing tool, with three big additions: a Pomelli Agent that builds your “Business DNA” (brand identity) from uploaded docs, photos or a chat; Brand Books that auto-generate a usable guide with your images, fonts and colours; and Websites that stand up a complete site in a few clicks. It’s still free in beta and available in 170+ countries. Paul J. Lipsky’s walkthrough cleared 50K views in five days — a clear sign marketers are paying attention.

What to do this week: If you’re a solo marketer or a small B2B team with no design budget, this is a genuinely free shot at a brand kit, brand book and website — run your own brand through it and compare the output to what you’d pay Canva Pro or a freelancer for. The deeper play: Pomelli’s “Business DNA” is a machine-readable brand identity, exactly the kind of structured asset AI agents will increasingly reference. Building one now is future-proofing, not just a freebie.

📋 The Playbook

1. Steal Brock Mesarich’s “Body → Brain → Hands” Claude Cowork system

Claude Cowork (the desktop app’s workspace mode) gets messy fast unless you structure it on day one. Brock Mesarich’s framework — built live in his 2 June video — gives you a clean, repeatable setup in three layers. Here’s the whole thing.

The Body (your workspace folder). Cowork works inside a real folder on your computer that Claude can read, write and edit. In the Claude desktop app, click “work inside a project,” choose “choose a different folder,” create a new one (call it cowork or similar), and select “always allow” so it stops asking permission for every change. Then structure it: one folder per area — clients, content, personal, finances — so it never turns into a junk drawer.

The Brain (CLAUDE.md + memory). Two files run the intelligence. The CLAUDE.md file loads every single time you chat — it tells Claude who you are, what you’re working on and how to talk to you. Brock’s rule: be specific or it’s useless. “Be helpful and professional” changes nothing; “short, direct answers, no filler, never use em-dashes, lead with benefits not features” actually steers the output. Each subfolder gets its own CLAUDE.md so Claude focuses only on the project you’re in. The memory.md file grows as you work — say “remember this” and Claude stores the fact for future chats. Trim it periodically so stale context doesn’t pile up.

His setup prompt (paste into Cowork): “Set up a clean Cowork workspace structure for me inside this folder. First, ask me two quick questions: (1) What are the 1–3 projects/ventures I want to manage here? (2) One line on who I am and what I do, so you can write my context file. Then create the structure with a root CLAUDE.md, a root memory.md, and a subfolder per project, each with its own CLAUDE.md, memory.md, inputs and outputs.” Tip: switch Claude from “ask” to “act without asking” so it doesn’t stop at every step.

The Hands (connectors, skills, scheduled tasks). This is where it does real work:

• Connectors: Customize → Connectors → browse and add the apps you use (Gmail, Google Calendar, Drive, Slack). For anything not natively supported, add the Zapier MCP server to reach thousands more apps. Set frequently-used ones to “always allow.”
• Skills: A skill is a job you teach Claude once so it runs the same workflow every time. Powerful example — teach it your email voice: “Learn how I write emails. Study my voice by reading my last 20 sent emails in Gmail, then build a reusable style guide and save it as a skill called ‘email response.’” Now every reply it drafts sounds like you.
• Scheduled tasks: Run a skill on a cadence. Brock’s daily 7am morning briefing scrapes his inbox and calendar, flags urgent items, pulls AI news for his channel, and posts a summary to Slack with a link to a dashboard. He also runs a weekly skill audit to prune skills he no longer uses.

Two cost/ops notes: scheduled tasks burn Claude usage, so don’t run things daily that only need to run weekly — and toggle “keep awake” on, because tasks won’t run if your computer is asleep. (Worth re-reading alongside this week’s Quick Hit on Anthropic’s 15 June billing change.)

Watch here

2. The three Claude MCPs every small marketing team should wire up first

If you’ve heard “MCP” (Model Context Protocol — the open standard that lets Claude connect to outside apps) and tuned out, this is the practical starting stack from Zubair Trabzada’s AI Workshop. No code required. In the Claude desktop app: + button → Connectors → browse connectors to add each one.

1. Canva MCP — on-brand graphics from a prompt. Connect Canva (free account works), then ask in plain English: “Using the Canva MCP, make me a promo graphic for a limited-time weekend sale — use a template you think fits.” Claude generates several variations inside Canva. The pro move: screenshot the one you like, upload it back, and say “apply my brand kit — logo, colours, fonts.” If you’ve set up a Canva brand kit, it pulls your real brand styling automatically. This turns “I need a quick promo image” from a 20-minute Canva session into a 60-second request.

2. Zapier MCP — one connector, ~8,000 apps. Instead of adding Gmail, Calendar, Slack, HubSpot and the rest one by one, the Zapier MCP gives Claude a single doorway to thousands of apps (free Zapier account needed). The standout feature is granular permission control the native connectors don’t offer — e.g. let Claude find and draft emails but never delete them. Wire up your daily apps, then ask things like “Using the Zapier MCP, check my Gmail for unread emails today and flag anything that needs attention,” or “...find my meetings tomorrow.” This is the closest thing to a real assistant that sees across your whole stack.

3. Apple Notes MCP (the sleeper) — instant recall of everything you’ve ever jotted down. If you’re on Mac, the read-and-write Apple Notes connector (built by Anthropic) lets Claude search years of notes that sync from your phone. Less of a marketing tool and more of a personal-ops unlock — but for anyone who dumps ideas, numbers and snippets into Notes on the go, “find that campaign idea I saved last month” suddenly works.

What to do this week: Start with Canva and Zapier — they cover the two biggest small-team time sinks (making assets and hopping between apps). Set sensible permissions before you set anything to “always allow.”

Watch here

🛠️ New Tools & Features

1. OpenAI Codex “Sites” — turn a plan into a shareable microsite by URL

Codex now lives inside the ChatGPT app, and Sites lets you turn an analysis, plan or brief into an interactive web page or mini-app you share via a link. Marketer use: instant internal campaign hubs, interactive one-pagers, or quick landing-page prototypes — no developer required. Business/Enterprise first.

2. OpenAI Codex “Annotations” — surgical edits instead of full regenerations

Point Codex at the exact part of a doc, slide or spreadsheet you want changed and tell it what to fix — it edits that, not the whole thing. This is the editing pattern marketers actually want: tighten one section of a deck or brief without nuking the parts that already work.

3. Google Gemini Omni — generate brand video from text, images or audio

Announced at Google I/O (19 May) and now rolling out, Gemini Omni is Google’s any-to-any multimodal model that outputs video with synchronised audio from any combination of inputs, plus conversational editing (add/remove objects, change scenes) and AI avatars. Marketer use: product demos, explainer clips and personalised ad variations generated from your brand assets. Currently short clips (~10 seconds at launch via Omni Flash).

4. Claude effort control comes to the interface

Anthropic added a low/medium/high “how hard to think” dial next to the model selector inside claude.ai and Cowork (previously API-only). Marketer use: crank it to high for strategy, briefs and positioning; drop it to low for bulk drafting to move faster and spend less.

5. Claude Code “Agent View” — supervise several agents at once

A new view to run and oversee multiple Claude Code sessions in parallel: start agents, send them to the background, and jump back in only when one needs your input. Marketer use: run content, research and reporting automations side by side with light-touch supervision.

6. CoSchedule Smart Editor — drafting plus brand-style enforcement in one screen

An AI writing workspace aimed at corporate marketing teams: draft copy, check it against your style guide and edit in a single view. It’s a crowded category — the differentiator here is the embedded brand-style checking, useful if keeping a distributed team on-voice is your pain point.

📊 Research and Data

1. Reddit is now ~40% of all AI citations — and each engine sees a different web

5WPR’s AI Platform Citation Source Index 2026 synthesised 680M+ citations across ChatGPT, Google AI Overviews, Perplexity, Gemini and Claude. The headlines: Reddit is the #1 cited source across every major AI engine at roughly 40% frequency; Wikipedia dominates ChatGPT (26–48% of its top-10 citations); and the top 15 domains capture 68% of all AI citation share — concentration far more extreme than Google’s PageRank ever produced.

Why it matters: This is your Answer Engine Optimization (AEO — getting cited inside AI answers) playbook in one dataset. You can’t optimise once for “AI” — each engine pulls from a near-different web. And community presence (Reddit especially) is now a genuine B2B visibility lever, not a nice-to-have. If your brand and category aren’t represented in the conversations and sources these engines trust, you’re invisible in the answers buyers now start with.

2. 96% of B2B marketers now use AI — but most are self-taught

Demand Gen Report’s 2026 B2B Trends research (300+ B2B marketers) found 96% now use AI in their roles and 47% rank it as the trend they’re most excited about. The gap underneath the adoption: across the wider 2026 research, a majority say AI training is critical to their success, yet most report their employer provides none — and 18% cite incomplete data as their single biggest barrier to confident decisions.

Why it matters: Adoption is no longer the differentiator — fluency is. If you’re building your own AI skills (which, if you’re reading this, you are), you’re not behind; you’re ahead of the 70% waiting for training that isn’t coming. The opportunity for marketing leaders: the team that systematises how it uses AI will outrun the team that just has access to it.

💡 Quick Hits

ChatGPT’s memory just got a major upgrade.

OpenAI began rolling out “Dreaming V3” (4 June) — memory that synthesises and self-updates automatically instead of waiting for you to tell it what to remember, with 2x capacity for Plus/Pro and an editable memory-summary page. Coming to free users next. Worth auditing what it’s “learned” before it carries assumptions into client work. https://9to5mac.com/2026/06/04/openai-says-chatgpts-memory-feature-is-getting-smarter-and-coming-to-free-users/

Anthropic splits Claude billing on 15 June — act if you run automations.

Programmatic Claude usage (Agent SDK, claude -p, GitHub Actions, third-party agents) moves to a separate, capped monthly credit pool billed at full API rates; interactive use in the terminal and app is unaffected. If you’ve wired up scheduled Claude marketing workflows on a subscription, reconcile before the date. https://the-decoder.com/claude-subscriptions-get-separate-budgets-for-programmatic-use-billed-at-full-api-prices/

Meta launches a “Business Agent.”

Unveiled 3 June, it lets businesses deploy AI agents for customer messaging and commerce across Meta’s surfaces — its latest move to sell AI, not just ads. Early and messaging-led, but one to watch for B2B. https://www.cnbc.com/2026/06/03/meta-business-agent-is-zuckerberg-latest-effort-to-diversify-from-ads.html

AdRoll ships an MCP server.

You can now query AdRoll campaign metrics in natural language directly inside ChatGPT or Claude — part of a clear pattern of ad data becoming chat-queryable. https://martech.org/the-latest-ai-powered-martech-news-and-releases/

ChatGPT now works inside PowerPoint.

OpenAI’s PowerPoint integration is live, bringing in-app drafting and editing to slides. https://openai.com/index/codex-for-every-role-tool-workflow/

Spotify opens an ElevenLabs-powered audiobook self-publishing tool (beta, invite-only).

Authors keep non-exclusive rights to AI-narrated audiobooks — a signal worth noting if audio is on your content roadmap. https://theaiinsider.tech/2026/05/22/spotify-bets-its-future-on-ai-personal-podcasts-licensed-remixes-elevenlabs-audiobooks-and-an-agentic-studio-app/

👀 On Our Radar

Selling to AI agents as a category

Greg Isenberg’s “The Next $100B Market: Selling to AI Agents” (33K views in three days) crystallises a fast-emerging idea: as buyers delegate research and purchasing to agents, marketers will increasingly need to optimise for machine buyers, not just human ones. It’s not actionable yet — but the framing is worth holding onto, because the AI-citation data above is the first edge of exactly this shift. The brands that machines can read, trust and recommend will win the agent-mediated sale.

Watch here

🔮 The Big Picture

Your marketing platform got an agent this week — whichever one you use.

Step back from the individual launches and the pattern is unmistakable. Salesforce, HubSpot and Google all crossed the same line in one week: their agents stopped suggesting work and started doing it. Journey Builder — a tool a generation of marketers built careers on — is being retired in favour of agents that orchestrate journeys themselves. The “should I automate this?” debate is over; the platforms have decided for you.

Which reframes the job. For a decade, marketing ops was about building the workflow — the journey, the nurture, the report. Increasingly it’s about governing it: deciding what the agent is allowed to do alone, what it must check with you, and what knowledge it draws on when it speaks for your brand. That’s a fluency problem, not a tools problem — and this week’s data makes the point twice over: 96% of marketers now use AI, but the ones who’ll pull ahead are the ones who can design and supervise a system, not just prompt a chatbot.

The marketers who thrive from here won’t be the ones with access to the best agent. They’ll be the ones who decided, deliberately, where the human belongs.

Tell me: What’s your biggest takeaway from this weeks news from the world of AI Marketing? Love to hear your thougths in the comments.

Discussion about this post

Ready for more?