Gary Sheng
Building AI-Powered Content Tools for Media Companies
Gary Sheng has been quietly building custom AI tools for media clients since early 2025. Not platforms. Not SaaS products. Specific tools that solve specific problems for specific teams, then iterating based on what those teams actually need next.
The pattern is always the same: sit with the client, understand what their team does every day, identify where they're losing time or quality, and build something that fixes it. The tools aren't theoretical. They're in production, used daily by content teams generating tens of millions of impressions.
Client A: A High-Volume Media Company
The first major engagement was with a media company that operates multiple brands and runs a team of content creators posting across platforms at high volume. Their existing workflow had several pain points that were obvious candidates for automation.
The Meme Generator
The team was creating image-based content (bold text overlaid on images) manually. Each piece required finding or creating an image, formatting text, and exporting. Gary built a custom meme generator that went through multiple iterations with the team before landing on a version they use every day.
"We went through several versions," Gary says. "You can't just ship v1 and walk away. The team has to live with it, tell you what's annoying, what's slow, what doesn't match the brand. Then you iterate until it disappears into their workflow."
The Video Reformatter
The team had a painful process: find a video on X (Twitter), download it (which is never straightforward), add a caption, and repost it as a Reel on Instagram. Every step had friction. Gary built a mobile-friendly app where anyone on the team can paste an X video URL, add a caption, and get a ready-to-post video out the other end.
"That one was pure workflow automation," Gary says. "No AI magic needed. Just removing every unnecessary step so the team can focus on the editorial judgment of what to post, not the mechanics of posting it."
The Image Stylizer
This was the tool that changed what the team thought was possible. Gary built an app that takes any reference image and transforms it into a specific visual style that the team battle-tested together. One of their brands uses a courtroom-sketch aesthetic. Every image from Getty or news sources gets stylized into that look. The result is instantly recognizable content with a distinctive brand identity, produced at volume.
"The stylizer wasn't automating an existing process," Gary says. "It was something they weren't doing before because it would have been impossibly expensive and slow. Once they had it, they realized they could create a visual identity that's theirs. Now if you've seen their content a couple times, you recognize it immediately."
The team uses it constantly. It's become foundational to how they create content across multiple brands.
The Results
The combination of these tools produced the media company dream: better quality, more distinctive brand identity, and higher volume. The team went from spending significant time on production mechanics to spending almost all their time on editorial decisions (what to post, what angle to take, what's worth amplifying).
The content reaches tens of millions of impressions. The tools didn't create the audience. The audience was already there. But the tools made it possible to serve that audience with better, more consistent, more visually distinctive content at a pace that would have required a much larger team.
Client B: A Podcast Content Strategist
The second client is a content strategist working in podcasts. Different industry, different daily workflow, but the same underlying pattern: identify what the person does every day, find the friction, build tools that remove it.
Gary built a suite of tools that make the strategist's work faster and more consistent. The specifics differ from Client A (podcasts have different production needs than social media content), but the approach is identical: start with the existing workflow, automate the tedious parts, then watch as the client's imagination opens up about what else is possible.
"That's the progression every time," Gary says. "First you automate what they already do. Match the quality, save the time. Then naturally they start saying, 'What if it could also do this?' Their imagination opens up because they're no longer buried in production work. They can think about what's next."
How He Works
Gary charges $175/hour and works directly with the client and their team. No handoffs. No requirement documents that get passed to a separate engineering team. He sits in the room (or on the call), understands the problem, and often has a working version within the same session.
"If I was outsourcing this to a software engineer, I'd still have to translate the client's needs into specs, wait for a build, review it, send feedback, wait again," he says. "Instead I'm building it on the spot. The client sees it working in real time. We iterate together. By the end of the session they have something they can use."
This is closer to the forward deployed engineering model that companies like Palantir use: collapse the distance between the person who understands the problem and the person who can build the solution. In Gary's case, they're the same person.
The Stack
Gary builds primarily with:
- Claude Code for rapid development and iteration
- Remotion for video and image generation tooling
- Next.js / React for web-based tools the team can access on any device
- Google Gemini and OpenAI APIs for image generation and stylization
- Custom scripts and automation pipelines tailored to each client's workflow
The tools aren't complex. They're specific. Each one does exactly what the client's team needs, nothing more. That specificity is what makes them actually get used instead of gathering dust.
The Pattern
Across both clients, the progression follows the same arc:
- Automate existing workflows. Start by doing what the team already does, just faster and with less friction. This builds trust and saves immediate time.
- Match or exceed quality. The output has to be at least as good as what the team was producing manually. If it's worse, they won't use it.
- Open the imagination. Once the team isn't buried in production work, they start seeing possibilities they couldn't before. New kinds of content. New visual identities. New formats that would have been too expensive to try manually.
- Iterate and expand. The first tool leads to the second. The second leads to the third. Each one compounds the team's capacity and the client's trust.
"Every company is a media company now," Gary says. "The ones that figure out how to produce better content, faster, with a more distinctive voice, are the ones that win. These tools aren't replacing the creative judgment. They're freeing the team to focus entirely on creative judgment."
Gary Sheng is the founder of Applied AI Society and an applied AI practitioner specializing in media and content automation. He works directly with media companies and creative professionals to build custom AI tools that increase output quality and volume.