Vicente Moreno

5k credits!!?? I am def trying to break it. Honestly the thing that surprised me first and still find key in every project is locking this context. Super useful for not having to re-explain characters, locations, names, lighting… it’s just essential. Curious to see how Agent Two behaves ;)

My most underrated Claude Cowork schedule: let it sort your desktop so you spend 0 seconds doing it. I take 30 to 50 screenshots a day. Things I paste into Claude, stuff I want to keep for later, stuff I send to people. Opening them, naming them and deciding where they go takes a surprising (and unnecessary) amount of time, so I stopped doing it. My desktop turned into a pile of "Screenshot 2026-07-14 at 11.42.03". Since summer started, Claude Cowork does it for me every morning 🌅 I work all day and pollute the desktop. Next day I open the laptop, the task runs while I go get my first coffee (~9:05 to 9:20), and when I'm back it's sorted. It opens every unnamed file -> reads what's actually inside -> names it descriptively -> moves it into the folder where it belongs: LinkedIn posts, article resources, design inspo, workflow screenshots, AI videos, work stuff. It also watches the videos. Screen recordings and downloaded videos get opened, watched, and named by what actually happens in them. Then placed in the right folder. The win is not only time saved. It's that I can find a screenshot from weeks ago in five seconds, instead of scrolling through 400 files named after timestamps. The whole setup is a schedule and a prompt. I told it where my files live, what my folders mean, and how I want things named. I can always review the summary of each run. Add exclusion rules if you have files you want to keep private and out of Anthropic's hands. How to set it up: 1. Install the Claude desktop app and open Cowork 2. Give it access to your desktop (or whatever folder is a mess) 3. Paste the prompt (add your own folders and naming rules) 4. Tell it to run every morning or multiple times a day. It will create the schedule itself ⭐ 5. Watch the first run, correct anything it got wrong, then let it go. Start automating what you hate doing. Not what sounds impressive. Prompt in the first comment ⬇️

I said I'd make the fish race. Episode 2 is about building Monaco first. Still on Agent One by invideo, still the same weird concept. This round was world-building. Location reference sheets of the Monaco circuit, four panels per generation, different lighting per shot. The yacht harbour at golden hour. Things that stood out: 1. It questioned me. My brief said fictional sponsors, and one generation came back with what was basically the Shell logo. It stopped and asked if I wanted to tweak it or keep it as parody. It caught the inconsistency against its own brief. 2. Referencing previous generations. It built character sheets from the fish I approved, not the ones rejected, and pulls those into every new shot. When one video came out hallucinated, the reason was simple: the right references never got passed. 3. Grouping generations per topic/beat. This is something I have not seen in other tools. Amazing for your mental model and quickly understanding what they are. 4. Advanced prompting generations: It generates good prompt where it @ the relevant references and includes negatives. That's the whole trick honestly. Consistency isn't a setting. It's what you feed the models and how you tell them to use it. Next episode: a dedicated commentator agent so the voice stays the same across every shot, experimenting with the camera angles and the actual race! 🐟 🏎️

I gave Agent One by invideo a stupid brief. Formula 1, but the cars are 7 metre fish. Five of them, racing through Monaco, all needing to stay recognisable across dozens of shots. Keeping characters consistent is usually the biggest challenge. Didn't really feel like it for this one. You hand it the brief and it delegates the difficult work (prompting and referencing) to one or multiple agents. It generates, reviews what came back and kills the bad takes before they reach you. At one point it killed its own tuna. Made it, looked at it, decided it was not good, and tried again. I never asked it to. This is def my favorite thing. The features I touch on in this video, which is Agent One 101: - One shot briefing - Agent-human collab - Auto-rejection of weak generations - Tweaking generations precisely - Parallel generations - Character consistency - Updating the brief, context and world as we go It dropped the driver out of a shot halfway through, I typed "add the driver, what are you doing" like I was messaging an intern, and it caught it and re-ran the shot itself. Honestly the biggest lesson had nothing to do with the tool. I wrote the brief properly before I opened it. That step up front saved me a lot and made the agent way more precise and aligned with the idea. The agents move fast. They still need someone who knows what good looks like. This is episode one. Next one I actually make the fish race. I will share the process, and when it breaks, you'll hear it here. More coming soon ;)

I never announced it. 3 months as an official Airtable creator. Not even my friends knew. I got selected for their first ever cohort. Either you created content and wanted to learn Airtable, or the other way around(me) And I already love the tool and can't shut up about it. I made a deal with myself. Let's try this content thing. How difficult can it be? It was not super easy, but when the first payment arrived, it opened my mind SO much. I learnt a ton of things: - Content creation and planning - How to get tech creative (if this concept exists) - Understand ugc value - Talking to camera - Battle the fear of posting I also built an (Airtable + Claude) system around the whole thing. Because I could not resist the idea of automating my own workflow. Drop the idea and get back 90% the process. Execution went from 10 hours to 30 minutes. The 7 main lessons + that system are in the carousel. And if you have a 9-5 and still think UGC means influencers holding skincare products: slide 2 is for you. Thanks to the Airtable team and Alex McDonnell for the trust. And the unlimited AI credits I definitely abused. #AirtablePartner #AirtableCreatorClub