Article
How I Use AI to Produce Content That Isn’t Slop
Luke Rodriguez, October 22, 2026
Last November, over breakfast in Lake Tahoe, my mom Lisa Rodriguez told me she wanted to start her own company again, and call it Slugline. She’s spent her career in PR / comms, and a few years before that she’d led her firm, The Seismic Collaborative, through an acquisition and gone to work at the company that bought it. I look up to her a lot, and I wasn’t going to pass up the chance to work with her. The next day I bought the domain and started brainstorming.
I was also in a bit of a transition phase… I’d become burned out on the work I was doing in New York. I felt like I was turning into a portrait photographer who occasionally produced a cool shoot and that I was leaving a lot of my skills on the table. (In hindsight that was pretty harsh. But a little self-criticism is healthy.)
I mostly got sucked into generative photo and video… keeping up with every new model, figuring out what actually gets consistent results, etc. Image gen is one of the coolest and scariest corners of AI. But I don’t think the most interesting use case is the generative model at all. It’s scripting within legacy imaging software itself… After Effects, Maya, Metashape, etc. You can probably connect the dots, and I’ll dive into that another time. Anyways… Slugline was a part-time thing for a while. Then about eight months ago we went all in, and now it’s out in the world.
Our first internal shoot
We wanted Slugline to feel real from day one, so we ran an internal shoot… headshots, talking to camera about where we think the industry is going, examples of the kind of work we do. We came away with hours of great footage. But (as startups do) we’d already started moving away from some of what we talked about, which left a lot of the edited videos unusable.
So I took it on myself to find a way to use what we had. Video editing is actually where I started, back before I knew what to call my career. Six-ish hours of footage, and even with a decade of editing experience I was dreading the assembly edit. The way my brain works, I figured I’d find a way to speed it up (which usually means spending more time figuring out the faster way than doing it normally).
I pulled the transcript from the whole shoot and told an AI model to cut anything that wasn’t an actual take. What I didn’t realize… I was deleting the most valuable thing an agent could have for an assembly edit. AI still can’t judge a delivery the way a good editor or journalist can. An agent reads a transcript, and on paper a forced line can look exactly like a natural one. Newer voice models can pick up some emotion and tone, but they’re nowhere near reliable enough to replace a real journalistic call yet.
Think of it like this: Writing is thinking. If you outsource your writing, you outsource your thinking. Every “you sound like you’re reading a script”, “slow down”, and “that was great” on set was exactly the info an agent needed to perform our actual decisions. The agent isn’t thinking for us. We already did, out loud. Once that hit me, I spent way more time building a system and planning the next shoot than it would have taken to just edit the thing in the first place. But now the assembly edit was pretty much automated.
I’d kind of been practicing for this
My high-volume photography experience helped here. With 6 to 10 headshot sessions a week and 200 to 300 e-commerce SKUs a month, my studio was a storm of moving equipment, shifting schedules and high-stakes projects. The fix was writing down exact color/exposure values for every background and texture, plus exact specs for every lighting setup… the only way a client’s imagery stayed consistent no matter what day something got shot. If it’s not spelled out, you can’t reliably replicate it. That approach became the blueprint for the Slugline production pipeline.
Then we booked our first client
Our first client shoot is where this had to work for real, and where the rest of Slugline comes in. Before we even book a shooting location, Melissa Laverty and Lisa do the story work. Walter (the agent Melissa built) interviews a company’s leaders and pulls out the patterns, and the two of them turn that into messaging: what the company should say, and what it can actually back up. That becomes the shot plan. We walk in knowing what we need to hear. It fixes something I’d noticed for years. In most video work, the people who know the story and the people who cut it are different people… so it turns into a game of telephone. On our shoots, the people who built the messaging are in the room giving direction.
This time we kept the cameras rolling all day, on purpose. Every ‘that was great’ and ‘try that one a little slower’ went straight into the transcript, right next to the take it was about.
It worked. Edits automated in post, decided by humans beforehand… honestly the best of both worlds. I picked up more along the way, pushed our post SOP further, and brought new ideas into the next shoot (in post right now, and going exactly to plan).
Under the hood
Getting here took some work. Most of our cutting happens in my terminal, before footage ever reaches editing software (separate timelines for each short-form clip, etc.). XML can’t cleanly apply those cuts to clips that are already multicammed, since they’re sequences, not video tracks, and stacking every angle instead gets cluttered and a little lossy once you start moving things around. So I spent a few days teaching my agent to properly interpret timecode, waveforms and metadata so it could generate XML instructions that properly adapted proxies and master files to work in Premiere and DaVinci.
Of course, about a month and a half later, Blackmagic shipped an official MCP server for DaVinci Resolve (and the open-source community has built solid ones for Premiere). Now the agent works through Resolve’s own tools instead of XML it wrote itself. It’s not a magic button… some things still need a custom script or a human. But the part that broke most often now mostly just works, and because every piece of our pipeline is separate, swapping it in didn’t break anything else. Same goes for the models. As they get faster, cheaper and smarter, so does our story work.
Today, footage gets transcribed, pulled into DaVinci, trimmed, and checked against the messaging and shot plan. Lisa and Melissa get organized clips with matching transcripts, pick the best takes, and it goes back into DaVinci for color, audio and finishing. The tedious parts of editing are handled, which leaves us a lot more time for the human parts.
What’s next
Lately I’ve been testing whether speech/cadence models can pick up what the agents miss… how well someone’s delivery matches their company’s voice, or which frame to cut on. Super early, and very much a research stage. I’ve also started experimenting with human editors and AI agents working in the same DaVinci Resolve cloud project… It’s really cool, and lets the agents script DaVinci without pausing local access. More on that another time.
Mostly, I’m proud of what the three of us have built. Slugline is officially out in the world, and if your company has a story worth telling, we’d love to help tell it.