After 4.5 years at Polygon, it would have taken something pretty special to get me to leave. Turns out, it was AI. Over the past year, I became slightly obsessed with figuring out how far we could push it inside marketing. What started as an experiment turned into an internal AI marketing system that changed how our team worked and how much we could accomplish. And for me personally, it changed what I wanted to do next. I wanted to go all in. Today, I’m excited to share that I’ve joined @Antseed to lead GTM and build the marketing function from the ground up. What BitTorrent did for files, Antseed is doing for AI inference. We’re building an open market where anyone can buy and sell access to AI models. People choose which models they use, who serves them, and what they pay. I really believe AI inference shouldn’t be controlled by just a handful of companies. A growing share of AI will run on hardware owned by individuals and independent providers, and those people will need a way to reach buyers, build a reputation, and get paid. That’s the opportunity I see in Antseed, and I think it's gonna be huge. Excited to join @AmosMeiri , @kotevcode, and the rest of the team as we announce our seed round today. LFG 🐜

This one is exciting to ship! I’ve tested it with our shared marketing agent, giving it access to 700+ models for text, images, and video at lower costs. I can also give teammates their own keys and budgets, and keep track of who’s using what and how much they’re spending. Looking forward to see how others will use it in their apps, agents, or teams!

hate to say this but video editors are cooked we paid an agency a lot of money to create our launch video, they worked on it for almost 2 weeks, and the cut still wasn’t what we wanted 2 days before the launch. so even though i have zero video editing experience, I decided I'd do it with AI. I gave Claude Opus 5.5 our raw footage and used the free version of DaVinci Resolve. I prompted the cuts, animations and music, everything agentic without touching the clip manually once. one crazy part is that Claude can't listen to the audio, but it was able analyze spikes in the sound data and reduced specific moments. eventually I finished it right before the launch. if anyone curious ping me in the comments and I'll share the agency version in private so you can see the differences.

This is going to be fun

For those using AI video tools, they know how expensive they can be. An open market will push prices and already most of whats offered on day 1 is cheaper than leading AI video companies. Go try it and tell us your how it goes!

Today we are launching the Antseed API gateway. If you’re running multiple agents or sharing AI access with a team, you can now give each one its own API key and spending limit through a gateway you control. Send the guide below to set yours up.

Join us in building a global, open market for AI inference

🤣

Smart - circle wants to grow the usage of the network by letting people speculate on their token, without actually committing to a token launch

1.6B tokens is wild I spent 29.9M tokens for $2.73 on GPT-6 Astra through @AntSeed to build a food delivery game in Tokyo

all these years we've been trying to run away from the word bot, and now X just laughs in our face and takes us full circle: bot -> chatbot -> virtual assistant -> ai agent -> ai teammate -> bot

and paid influencers to boost it.. nothing wrong with that, but the production wasn’t why it reached 2m views

kills me that claude knows almost everything in the world except it's own capabilities

opus 5 is out, here's how marketers should pick a claude model: • opus 5 is your strategist and builder It's now the strongest model at half the price of fable. use this model for claude design, coding internal marketing tools, work that needs real marketing judgment like GTM planning, positioning and messaging, and building the skills your team reuses. • sonnet 5 is your creative executor ideal for content writing, landing pages, blogs, emails, social posts, editing, everyday research, and general marketing work. it's much cheaper than opus, and excellent for daily output that requires less strategic thinking. • haiku is your volume processor great at classifying data, tagging content, monitoring brand mentions across a ton of sources, processing large datasets, etc. when speed and cost matter most, haiku is the clear choice. • fable 5 isn't for you way more expensive than the others, and only ahead in a few narrow areas like cybersecurity. i can't think of a marketing task where it's worth paying double for it. putting this set up to test it in my workflows this week. will share more on it soon.

you don’t need to train your own model to know you need a better striker

most people misunderstand how ai agents work. skills are a better approach than general instructions for an agent. agent instructions are what gets read every time vs skills that only get read when “invoked” this way you keep a lean context, get faster responses, and simultaneously you can teach your agent infinite skills