This is the eighth post in my AI Agent Series, where I walk through each agent I run operationally in my business and show you exactly how it works. This one handles ad creative generation for TikTok and Meta.
The previous posts in this series covered morning email triage, VOC ad copy analysis, Klaviyo flow building, blog publishing, client onboarding, the iMessage command interface, and the DTC growth dashboard. Each agent removes one specific bottleneck. This one removes the one that quietly kills most paid social accounts: running out of creative before running out of budget.
The Before: Creative Production as the Actual Bottleneck
The conventional explanation for why a paid social account stalls is targeting, or pixel quality, or budget allocation. Most of the time, it is none of those things. It is creative. TikTok's own data shows that top-performing ads last seven to ten days before fatigue sets in. Which means that if you are spending meaningfully on TikTok, you need new creative every single week. At $250,000 per month in ad spend, industry benchmarks put the minimum viable creative volume at 60 to 120 new concepts per month just to have enough variation for the algorithm to find winners.
Traditional production cannot keep up with that cadence. An agency charges between three and fifteen thousand pounds per video and takes three to four weeks to deliver. An influencer costs upwards of a thousand per asset and is not available on demand. An in-house team at that volume requires a full creative department with editors, videographers, and copywriters. Most DTC brands have none of those things at the scale required. So they keep running the same four creatives until ROAS craters, then scramble to produce something new. That cycle is the problem this agent solves.
What the AI Ad Creative Agent Actually Does
The agent starts with a brief. That brief can be as simple as a product name, a target audience, and a goal. From there, the workflow runs in four stages without any further input required.
VOC research
The agent mines customer reviews from Google, Amazon, Trustpilot, and wherever else the brand appears. It is looking for three things: the exact language customers use to describe the problem the product solves, the most common objections that delayed purchase, and the specific outcomes customers report after using the product. This is the raw material for creative that actually converts.
Brief and angle generation
From the VOC data, the agent identifies five distinct creative angles. Each angle is built around a different psychological trigger: desire, fear of missing out, social proof, effort contrast, or identity. Each angle gets a primary headline, a supporting hook, and a 30-second script framework. The language comes from real customer reviews, not from a generic AI prompt.
Visual prompt engineering
For each angle, the agent writes a detailed image and video generation prompt tuned for the platform. TikTok prompts are built for vertical video with a strong visual hook in the first two seconds. Meta image prompts are structured for 3:4 and 1:1 formats with copy-overlay placement in mind. The prompts are sent to Higgsfield, which packages 15 or more video models including Sora 2, Veo 3.1, and Kling 3.0 under one generation layer.
Creative package delivery
The agent returns a structured package: the brief, all five copy angles with full headline and body text, and the generated media files organised by angle and format. Everything is saved to the brand's folder on Google Drive and flagged for review. The operator reviews the package, selects the strongest angles, and uploads directly to Ads Manager.
The full workflow, from brief submission to delivery of a complete creative package, runs in under 30 minutes. The same output from a creative agency would take three to four weeks and cost between five and twenty thousand pounds. From an influencer pipeline, assuming availability and the right fit, the timeline is similar and the cost comparable per asset. The agent does it in half an hour, any time of day, for any product in the portfolio.
The Brand Memory and Context Layer
What separates this from a generic AI image generator is what the agent knows before the brief arrives. Every creative run is loaded with the brand's context: the product range, the target audience profile, the tone of voice, the visual identity guidelines, and the history of what has performed well in previous campaigns. It does not start from a blank prompt. It starts from a trained understanding of the brand.
This context layer is what makes the creative feel on-brand rather than generic. When the agent generates a TikTok hook for a wellness supplement brand, it knows that the audience responds to specificity over aspiration, that certain health claims require careful framing, and that the brand's previous top-performing creative opened with a relatable problem rather than a product claim. That institutional knowledge, stored and applied automatically, is the difference between creative that looks like it came from the brand and creative that looks like it came from a template.
What the Output Actually Looks Like
A standard creative batch for a skincare brand running on Meta and TikTok would look like this. Five angles based on the brand's review data. Angle one: a before and after framing built around a specific skin concern pulled verbatim from a five-star review. Angle two: a social proof stack opening with a quoted customer result. Angle three: an effort contrast showing what the customer was using before and what changed after switching. Angle four: an ingredient authority angle targeting the informed buyer who already knows what to look for. Angle five: a lifestyle identity angle for the aspirational buyer who is not yet in pain but wants the outcome.
The agent does not produce mood boards or wireframes. It produces platform-ready files: video in 9:16 for TikTok, images in 3:4 and 1:1 for Meta, with copy overlays positioned for safe zones. The operator receives a folder, not a brief.
Each angle comes with a complete copy package: a primary hook, a secondary headline, the full body copy for a static ad, and a 30-second spoken script for video. The operator does not need to write a single line. The review step is exactly that: a review. Does this angle feel right? Is the voice consistent? Is there anything that needs adjusting before this goes into testing? Most batches need minor tweaks. Some go straight to Ads Manager unchanged.
The Creative Velocity Advantage
The most important outcome this agent delivers is not cost savings. It is creative velocity. Brands that refresh creative every seven to ten days report CPMs 22 to 31 percent lower than brands refreshing monthly. TikTok campaigns with five to seven creative assets refreshed weekly achieve 1.5 times higher ROAS compared with campaigns running the same creative for extended periods.
The algorithm rewards freshness. It rewards volume. It rewards variation. It rewards the brands that give it enough material to find what works. Those are not characteristics of a creative agency relationship or a monthly influencer retainer. They are characteristics of a production system that runs continuously, on demand, at the pace the algorithm actually requires.
A client I work with in the supplement space went from producing eight to ten creatives per month via agency to running three agent batches per week, each producing ten or more assets. Within the first 60 days, their cost per acquisition dropped 41 percent on TikTok. Not because the targeting changed. Not because the budget shifted. Because the algorithm finally had enough creative volume to optimise properly.
Inside the system
How we build this for brands
The creative agent is one component in a broader growth system. The VOC engine that feeds the brief, the performance dashboard that tells us which angles are winning, and the Klaviyo flows that catch the customers the ads bring in, all of those run as connected agents rather than isolated tools. The creative agent knows what the winning creative from the last campaign looked like. The email system knows what offer converts best for the audience the ads are reaching. The whole stack compounds because the agents share context.
The creative agent also pulls from the VOC mining agent, which runs continuously across review platforms and surfaces new customer language as it appears. When a new objection emerges in the reviews, the creative pipeline picks it up and can generate a new angle around it within the same week. That kind of responsiveness is not possible with a traditional agency or influencer pipeline. Part of this runs live for portfolio brands today. The full system, configured around a specific brand's data, customer reviews, and creative history, is what we deploy when we take a brand on.
Build Your Agent Team
Get your own AI agent team working on your brand
Caner can help you build your own AI agent teams and unlock growth for your brand. This is not a template or a SaaS tool. It is a custom system designed around your data, your workflows, and your growth targets.
Book A DemoFrequently asked questions
Can I build an AI ad creative agent like this myself?
Technically yes, if you have experience with Claude's API, video generation models like Higgsfield, and agent orchestration. The system combines a brief-parsing layer, a VOC research module, a prompt engineering layer trained on high-performing creative formats, and a video generation API. Building this from scratch and getting it to produce brand-consistent, platform-appropriate creative takes weeks of iteration. The faster route is working with someone who has already built and tuned the system across multiple brands, so you get the output immediately rather than spending months building the infrastructure.
How long does it take to set up this kind of agent?
The core pipeline, from brief input to generated creative, can be configured in two to three days. The meaningful time investment is in training the agent on the brand's context: product details, tone of voice, existing creative that has performed well, and the specific objections the brand faces from its target customer. A well-tuned system that produces on-brand creative without manual prompt engineering takes one to two weeks of setup. After that, the ongoing time cost per creative batch drops to under five minutes.
Does AI-generated creative actually perform on TikTok and Meta?
Yes, with the right approach. The creative still needs to be rooted in genuine customer language, real objections, and platform-native formats. AI does not replace strategy. It executes strategy at a speed and volume that manual production cannot match. Brands refreshing creative every seven to ten days with AI tooling are reporting CPMs 22 to 31 percent lower than brands refreshing monthly. The algorithm rewards creative velocity. AI makes creative velocity achievable for brands that do not have a full in-house production team.
What formats does the agent produce?
The agent produces image ads in 3:4, 9:16, and 1:1 formats for Meta placements, and video ads in 9:16 for TikTok. Each batch starts with a VOC-grounded brief, produces multiple copy angles with headlines and body text, then generates the visual assets for each angle. The output is a structured creative package: brief, copy variants, and media files organised by angle and format, ready to upload to Ads Manager.
How many creative variants does one agent run produce?
A standard run produces five ad angles, each with a primary headline, supporting copy, and at least two format variants. That is a minimum of ten creative assets per run. For brands spending at a level where creative testing velocity matters, the agent can be run multiple times per week, producing 40 to 60 new assets per month at a fraction of what agency or influencer production would cost for the same volume.
About the author
Caner Veli built Liquiproof to global distribution across 3,000+ retailers, then exited. He now runs Purposeful Profits using a combination of operator strategy and AI-powered systems he has built and uses daily, having 10x'd monthly revenue in his own business in the last 90 days.