Built This Week/AI by Industry

AI by Industry

AI in Marketing

AI already does real marketing work for us and our guests: video ads, SEO posts, outreach and ad concepts. Juma's founder says to spend the saved time on better work, not just more of it.

By Jordan Metzner & Sam Nadler · From our episodes · Updated October 2026

Watch: Episode 35, AI Agents Are Replacing Entire Marketing Teams from 08:16

On Built This Week, AI in marketing is something we use every week, not a forecast. We built an AI video studio, an autoblogger that publishes SEO posts, a podcast outreach CRM and n8n outreach workflows, and lately we draft ad concepts in Claude Design. Our main marketing guest, Iliya Valchanov of Juma, runs a marketing agent and has shrunk his own team with it. His warning is that saving time only matters if you put it into better work.

So we had seven people in sales and marketing. Now we have one. And we're doing more work than before.

— Iliya Valchanov, Episode 35

How companies we talked to use AI in marketing

Juma: a super agent for marketing tasks

Juma plans and executes marketing work: content, performance marketing, ads, images and carousels. It has about 100K users, mostly marketing agencies, then in-house marketing teams, then solopreneurs who don't know how to market. On the show, Iliya prompted it for an April social media calendar for HipTrain. It researched the site, ran a competitive scan, asked a clarifying question, then produced content pillars and day-by-day posts, and a follow-up prompt turned the calendar into an Instagram carousel. Under the hood are hundreds of human-written marketing skills and connectors to Google Analytics, Google Ads, Meta Ads, HubSpot and Search Console.

Iliya measures the payoff differently by customer. Agencies count billable hours and save 10x to 20x the time on a task. Small teams use AI to chase every idea they have. For ads, where budgets are large, he said the savings can reach millions.

So the first thing is it's not about doing stuff faster. It's about doing stuff better.

— Iliya Valchanov, Episode 35

He added two more lessons at 13:00: after coders, marketers are the most eager adopters of AI, and every company needs one person who owns the AI transformation. Buying tools alone fails.

What we built

  • Ryz Labs Video Studio: video was our social team's biggest bottleneck, so Jordan built one internal app that calls Veo 3, Seedance and other models pay-per-use, with an "enhance with AI" button for prompts. One of our video editors generated 20 to 30 capybara clips in about an hour and had a finished commercial within a day. The catch: Veo 3 cost about $6 per eight-second clip.
  • Ryz Labs Autoblogger: writes illustrated blog posts with Claude and OpenAI models and publishes them to each brand's Webflow site, about 150 posts per brand. Jordan's SEO lesson: write for what people search, like "off-site planning checklist," not for your brand name. NTRVSTA got leads from as far as Malaysia, Holland and the UK.
  • Podcast CRM: Jordan pulled the top 1,061 business podcasts and the contact emails in their RSS feeds, and AI writes a personalized pitch that opens in Gmail. He kept the emails humble and didn't ask for anything.
  • Claude-generated n8n outreach workflow: Jordan asked Claude for an n8n workflow to invite podcast hosts on the show and imported the JSON in seconds. Claude gets you about 90% of an automation; the credentials and connections are the last 10%.
  • HipTrain Meta ad concepts in Claude Design: Sam asked for three cheeky campaigns aimed at women aged 50 to 70, with an image and Veo prompt for each, ran one image through Nano Banana and shared the deck with the marketing team for comments.

We also tried other people's tools. In Episode 20, Google's Pomelli pulled a brand's "business DNA" (colors, fonts, images, taglines) from a website and generated social campaigns in minutes. Sam found it useful for quick inspiration; Jordan said some images looked "very AI."

What's still hard

  • Brand risk in automated content. Jordan only recommends auto-publishing once you trust the quality:
But I think that there's a little bit of risk of, you know, ChatGPT hallucinating or just not understanding exactly what the brand is and pushing out content that's not on brand.

— Jordan Metzner, Episode 2

  • Text in images. When we tested Envato's AI image tool for an Instagram ad in Episode 9, the text was misspelled and none of the results were good. Running the same prompt across several models in our own tool did better.
  • Video costs add up. Premium video models are priced by generation time, and generations aren't perfect, so a startup can burn money on retries.
  • Automation's last mile. AI drafts the n8n workflow, but someone still has to wire up credentials and test it.
  • Too many agents. Iliya said he can't run more than three agents at once because he can't process the output of more; Sam said the same.

Episodes on AI in marketing

FAQ

How is AI used in marketing?

On our show it writes and publishes SEO blog posts, generates video ads and images, drafts personalized outreach emails, builds n8n automations and proposes ad concepts. Juma, a marketing agent with about 100K users, turns one prompt into a month-long social media calendar and an Instagram carousel.

Will AI replace marketing teams?

Juma's own go-to-market team went from seven people to one doing more work than before. But CEO Iliya Valchanov says the point is quality, not speed, and that companies fail at AI when nobody owns the transformation.

Should you fully automate AI-written SEO content?

Jordan's rule from running our autoblogger: only automate publishing once you trust the quality, because hallucinated or off-brand posts can hurt you. Target what people search for, not your brand name.

Are AI image generators good enough for ads?

It depends on the model. When we tested Envato's AI image tool for an Instagram ad, the text came out misspelled, while some models in our own multi-model image tool rendered text correctly. Sam later used Claude Design and Nano Banana to produce Meta ad concepts he found high quality.