If you only check one highlight this week:
B2B CMO? Check out the 1st two demos and find a systems thinker on your team who can help you build that type of stuff
Everyone else: check out Jacob's demo to learn how you can now quickly and easily build an agent (in under six minutes). Or if you want to be inspired by what one can build with AI, watch Mark Hull's demo of his Chief of Staff. It's also a precursor of what's coming soon. Most likely from Anthropic.
-François
Five impressive agents showcased at Workflow Demo Fest II
My friend James Raybould and I hosted our second AI Workflow Demo Fest on Thursday.
You will find the video and AI-generated summary for each talk in the following sections.
Our 120 attendees were blown away: only five-star reviews.
It was easily the most inspiring and motivating AI event of 2026 so far.
The demos were much more impressive than they were 3 months ago for our first edition.
It's not that our presenters were bad three months ago. Far from it.
It's mostly the result of:
How much AI tools have improved in ONLY a few months
We had (a little) more time to learn and use them better
Builders who are great system thinkers AND very good at their craft:
for Mark, running a company and leading Product
for Kieran, Marketing. Especially creating and publishing great content
🏅 Mark, our winner, spent 200 to 300 hours and about $10,000, using AI tools (mostly Claude Code), to build a super-impressive strategic Chief of Staff. When you see what it does, you will hopefully, as I do, think that investment is reasonable.
🥈 Kieran architected (really worth a read) his personal content creation team impressively. It writes like him and adapts to every platform he posts on. It also self-improves as he gives feedback, automatically updating all his skills.
🥉 Deedi's demo will be highly applicable to any marketing or product marketing leader seeking to track their competitors' launches and messaging shifts. She built a tool that identifies significant changes in competitors' offerings or positioning that Bubble should pay attention to. And it proposes reactive messaging.
Jacob demonstrated, again, that you can build an ABM agent and proof of concept in six minutes or less. I can only imagine how much better he could have made it with an extra hour or two. There is no excuse to send generic email sequences anymore, when you can run such a workflow.
Finally, Peggy showed us that you can replace even the most successful Martech tool of the year (Clay) with a good architecture and a series of skills markdown files. She showed how you can do that in a terminal, but you can go pretty far by using a web or desktop AI coding app as well.
The five agents that will hopefully impress you too:
🏅 An impressive strategic chief of staff agent built with Claude Code
In 6 minutes, Mark Hull demonstrates his semi-autonomous chief-of-staff agent built with Claude Code that manages his 15-person startup's operations across 25 integrated tools.
Connects to 25 different tools, including Slack, Apollo, Instantly, Notion, Linear, and Figma
Agent analyzes meeting notes and email context to suggest 3-4 response options, then executes with approval
End-to-end product development: Takes product notes and automatically creates PRDs in Notion, breaks down tasks in Linear, generates prototypes in Figma, and builds with Claude Code
Proactive roadmap suggestions: Analyzes competitors and customer feedback to recommend what features to build next
Context-aware prioritization: Dashboard shows urgent items based on Hull's work patterns and current projects, not just due dates
Marketing automation: Generates support materials and LinkedIn posts automatically
"YOLO mode" for full automation: Completes entire workflows without intervention when Hull wants to test maximum capability
Sales process automation: Identifies prospects, ranks against ICP, drafts outbound messages, enriches with LinkedIn data, and adds to CRM
🥈 Kieran’s strategic and self-improving content AI team, built with Claude Code
Kieran Flanagan demos his AI content team built in Claude Code. It automatically researches trends, creates modular content blocks, and improves itself (i.e. its skills) through feedback loops.
Now, it's not this tool that made Kieran reach 105k followers on LinkedIn, but it surely helps.
Auto-generates audience profiles from domains instead of ICPs. Flanagan says ICPs are outdated for content marketing.
Creates writing styles by analyzing any creator's existing content. Just input a name and platform where they publish.
Builds platform-specific styles for Substack, LinkedIn, YouTube. Each platform requires different content approaches.
Maintains a content queue that populates automatically. Research agents add trending topics and lookalike content weekly.
Uses lookalike research based on top-performing posts. System analyzes your analytics to find high-performing content patterns.
Labels content by type: data nugget, educational, spicy take. Each type serves different audience engagement goals.
Enrichment feature breaks content into modular blocks. Includes story hooks, case studies, quotes, counterarguments, visual elements.
Self-improves through feedback loops. When you review content queue, all skills update automatically.
Can run fully automated with manager agent. Creates weekly content strategy and executes without human input.
Integrates multiple APIs: X, OpenAI, Perplexity. Pulls trend data from various sources simultaneously.
🥉 A competitive messaging intelligence agent built with AirOps
Deedi Brown explains why and how she built an automated system using AirOps to track Bubble’s competitors’ messaging changes in the crowded AI app development space.
Weekly scrapes competitor homepages, navigation pages, and recent blog posts automatically
Compares current messaging against previous snapshots stored in knowledge base
Assigns messaging shift scores from 1-10 to indicate the severity of changes
Sends Slack alerts when shifts score above the threshold or product launches are detected
Tracks specific competitors like Replit, which scored 5/10 for positioning shift plus major Agent 4 launch
Analyzes how competitor changes intersect with Bubble's own positioning
Generates LinkedIn post drafts for CEO responses when significant shifts occur
Runs on Mondays but considering daily cadence depending on token costs
Still testing whether competitors’ website scraping captures real shifts better than social media posts
Working to optimize Slack alert format so the team actually reads and acts on notifications
💡An ABM workflow based on LinkedIn comments with Relay
Jacob Bank demonstrates building an AI workflow in Relay.app that automatically finds HubSpot LinkedIn commenters and generates personalized ABM campaigns for each prospect.
The workflow runs weekly and pulls the most recent HubSpot LinkedIn posts automatically
AI analyzes 20 comments per post to identify one decision maker at a medium-sized business as the best prospect
The system generates exactly 3 personalized ABM ideas: custom landing page, physical gift, plus one surprise option
Natural language prompting builds the entire workflow without manual flow chart construction
The AI selected "Syed" as the best prospect and created a "CRM to revenue" landing page concept specifically for him
Physical gift suggestion included a customized notebook referencing the prospect's job title with specific budget parameters
The surprise third option was a 90-second personalized documentary
LinkedIn data collection covers three categories: official API actions, publicly available information, and automated browser actions
Comments proved sparse compared to reactions on brand accounts (workflow needs adjustment to pull reactions instead)
The system includes feedback loops to refine prospect targeting and output format
Relay.app positions this as "meta-prompting" where multiple AI prompts run within one workflow structure
📁 Replacing Clay with a custom agent based on skills markdown files
Peggy Rayzis shows how she replaced Clay with a custom agent called Kite that runs entirely on Markdown files for client prospecting and outreach.
Skills as instructions: Agent loads "skills" (packages of Markdown files containing instructions, examples, and test scripts) for tasks like ICP definition, outreach, enrichment, and lead scoring
Live prospecting demo: Fed her agent the attendee list from Luma, scraped 65% of LinkedIn profiles and social posts, then searched for "AI-pilled marketers" in real-time
Pain point identification: Agent searches social posts to find prospects frustrated with specific problems (like GTM tooling issues), then auto-generates personalized outreach emails
LinkedIn automation: Automatically sends connection requests to identified prospects during the demo
Tech stack: Built on Pi (the agent framework behind OpenClaw), uses SQLite for small datasets, scales to Turbo Puffer for hybrid search with larger databases
Data collection: Webhooks track signals like founder social engagement, signups, job postings mentioning relevant categories
Weekly reporting: Generates reports connecting hot leads to different touchpoints, pushed to Slack
Skills customization: Generic skill repositories produce generic marketing. Best results come from customizing skills with company-specific examples and tests
Starting point: Create one marketing skill combining your positioning, messaging, and ICP documentation that teams can use across Claude, Cursor, or other tools
Final Words
I think the companies that give everyone on their team a team of agents are going to kick the shit out of the companies that replace their teams with a team of agents.
Thanks for sharing these highlights with busy marketing execs around you.🙏
Someone forwarded you this email? You can subscribe here.
François | LinkedIn
I'm a CMO, advisor, and "CMO Wingman". Yes, that's a thing :-). Ask my clients: in this AI era, CMOs need a strategic proactive advisor more than ever. I’m former CMO at Twilio, Augment Code, Apollo GraphQL, Decibel, Udacity and Head of Marketing for LinkedIn Talent Solutions.
