Carlos Granier

@dsqjaffa Monday

10/ try the API - build your own monday drop. matt barely read the docs, he just pointed OpenClaw at them and the agent figured out the endpoints on its own. reply "MONDAY" and i'll send you the full setup. (must be following)

11,000,000 views in under a month. here's how: @TheMattBerman built a viral content agent for his clients on OpenClaw (steal the EXACT setup - thread below)

@peterpawlak @make_hq @claudeai @TallyForms Probably... But I didn't know this at the time. Now I'm all out of credits for the week 😂

@cgranier @make_hq @claudeai @TallyForms I don't think Claude knows exactly how to create good scenarios in Make. Right now it's much simpler to just create simple app on Vercel with forms exactly as you want and data manipulation in backend. This Claude understands and can create without much problems.

Some days @claudeai surprises me with how smart and thorough it can be. And then there are days when it will eat all my credits attempting to reinvent the wheel. This is a @make_hq flow to read a @TallyForms submission, grade it in Claude and email a report.

Sora's numbers are shocking. Is consumer AI video dead or just using better tools?

Full roundup here: rctv.com/posts/ai-video…

Just stumbled on this AI Video Weekly Roundup that breaks down the brutal reality behind the hype. OpenAI just killed Sora after burning $15M/day at peak… for only $2.1M total revenue ever. Downloads dropped 66%, retention collapsed, and the $1B Disney deal died with it. Meanwhile ByteDance’s Seedance 2.0 is crushing benchmarks at 1,269 Elo (but still blocked in the US), and Luma’s new Ray 3.14 is 4x faster + 3x cheaper. Wild week in AI video. Worth a read 👇

@peterpawlak @make_hq @claudeai @TallyForms Simple flow: collect data from form, evaluate answers in Claude, return a score plus formatted form data. And yes, the solution was to force JSON though the "advanced settings" in the Make node. Claude was instead just trying to fix the broken JSON with more nodes.

@cgranier @make_hq @claudeai @TallyForms Depends on what you want to do. I’d probably use Gemini for simple data manipulation or/and try different prompt. Depends on the chosen model as well. But usually you can force json by selecting structured output and providing template or example how it it should look like

@peterpawlak @make_hq @claudeai @TallyForms The real problem here was that the Anthropic node wouldn't generate a valid JSON payload, even when instructed to do so. So Claude's solution was to brute-force it into JSON using a bunch of extra manipulations nodes. Turns out it was just a matter of adding a schema to the node.

Yeah and tried to iterate on data from webhook and then aggregate them again. Probably not needed before sending them to Anthropic. But yeah could be done more efficient. But I liked to work with webhooks in Make. Now it’s easier to just use Claude code and actually code automations.

@ryanlightbourn @FCG_Studio 😬

@FCG_Studio No, but there's a respectable supply of stachybotrys.

A guy in his 40s with a MacBook Air made this in his mom’s basement with $500 in credits. This is what microbudget film & TV looks like now. Models used in the description.

@claudeai @make_hq @TallyForms The real problem here was that the Anthropic node wouldn't generate a valid JSON payload, even when instructed to do so. So Claude's solution was to brute-force it into JSON using a bunch of extra manipulations nodes. Turns out it was just a matter of adding a schema to the node.

@peterpawlak @make_hq @claudeai @TallyForms That's an interesting distinction. But apart from the webhook, Claude was trying to wrangle the output of the Anthropic node instead of just having it be JSON. Granted, this was a very obscure setting that took me several web searches to discover...

@cgranier @make_hq @claudeai @TallyForms The difference is webhook is instant on event action and the second scenario runs on interval. If you can wait it’s ok.

@renoiseai MAKENOISE

Launch week special for early supporters — 40% off. Follow @renoiseai + reply "MAKENOISE" and we'll DM you the code.

Don't make videos — program them. Introducing Renoise = Claude Code + Seedance 2.0. Scale your video ad production 100x. One product photo → hundreds of finished creatives.👇 renoise.ai #Renoise #RenoiseAI #CreateWithRenoise #Seedance2

Quick heads-up if you're trying to set-up an MCP server in @claudeai Desktop (Windows): claude_desktop_config.json will not read variables from your environment, so you'll need to hard code them in. If you know a way that works, let me know...

Runway @runwayml demoed sub-100ms AI video at @nvidia #GTC2026. Not faster generation. A new product category. Real-time video means AI isn't in your production pipeline ⚡ it's in your product. Most enterprise AI roadmaps aren't built for that shift. #AIStrategy #AIVideo

Full breakdown of this week's shifts: Sora retirement, Runway Characters, Luma Agents, and the NVIDIA GDC numbers —> in this week's roundup: rctv.com/posts/ai-video…

1/ Everyone's talking about Sora folding into ChatGPT and Luma Agents landing at Publicis. They're missing the actual story this week. Local 4K AI video just went mainstream. And it's going to restructure who can compete in this industry. 🧵

9/If you're advising clients on AI video strategy right now and you're not building local generation into the conversation, you're giving them an incomplete picture. The cloud tools are excellent. But they're not the whole story anymore. 🔁 Repost if this changes how you're thinking about the stack. → Full breakdown on this week's shifts: [rctv.com/posts/ai-video…] #AIVideo #GenerativeAI #AIFilmmaking #CreativeAI #LTX

8/Watch GTC next week (March 16–19). Jensen Huang's keynote is expected to cover next-gen silicon and physical AI. If NVIDIA announces anything that further accelerates local video inference, the "cloud is required" argument takes another hit. The direction is clear. The pace is the variable.