Recent posts
It has to be @Microsoft right
Scoop: HuggingFace has M&A interest and they could get bought for $13B…or higher! https://www.businessinsider.com/hugging-face-could-be-acquired-13-billion-2026-8
The main thing that frustrates me about remote sandbox development is the additional terminal latency. I still type a lot (amid Wispr'ing), and it's obviously noticeable even on servers near SF. I'm working on extending fx (from @vercel_dev) with support for attaching to a remote session over tailscale RPC. It feels *really* good. I've always been sensitive to typing latency. During COVID in my neovim maintainer days, I spent an entire week profiling every bit of the LSP client and lifting the performance sensitive components (mostly diff tracking and json parsing) into C.
I contributed ~1/2 of this! Used to be more impressive pre-AI but glad to be able to use fx for both tokenmaxxing on @OpenAI and switching to @vercel AI gateway for overflow.
http://fx.sh now supports your @Grok & Codex subs Test it out on a sandbox. Instant to install:
Zig has one *major advantage* over Rust in the AI coding agent era, it's extremely fast to compile. I was having 4-5 fx agent iterate overnight extending it with < 1s compilation time which is huge for testing. Already have a working prototype of a plugin system via WASM that I'm super bullish on, I *may* abandon Pi.
Introducing fx, a tiny, open, native coding agent from Vercel Labs. Originally an internal tool, fx is a harness and CLI written in Zig, optimized for research and embedding in larger systems. Today, we're open sourcing it. fx is built on three principles: 1. Fast. A single native binary, no runtime to install. It cold starts in 10µs and does no unnecessary work or I/O before accepting input. fx is the answer to "how fast can a coding agent be?" 2. Light. The 6.3MiB binary uses single-digit megabytes of memory at baseline, made for instant installation and embedding in resource-constrained environments and agent sandboxes. 3. Open. Apache-2.0, model and provider agnostic, suitable for local and cloud inference. Its small core extends through skills, plugins, and MCP. Minimalism is an obsession throughout the entire harness: system prompt, tools, features, binary. The goal was to keep context usage and time to first token low, and make fx optimal for model benchmarking, sandboxing, evals, and gyms. You can use fx directly or embed it as infrastructure. The CLI feels more like a Unix shell than an IDE in the terminal: it preserves scroll history, produces minimal output, and uses complex TUI rendering very, very sparingly. Programmatically, 𝚏𝚡 𝚊𝚜𝚔 --𝚓𝚜𝚘𝚗 gives structured output, 𝚏𝚡 𝚊𝚌𝚙 connects to editors and other clients, and WebAssembly can even run the whole thing inside the browser (see: https://t.co/wf2Trg47sC). Privacy is a design constraint: no product telemetry, sessions and usage stay local, and no source code or prompts are shared with any endpoint other than inference. With local inference and auto-updates off, fx is fully hermetic. fx is experimental. Use at your own risk and expect frequent changes. Chat with us on X (https://t.co/A2AB2YythC) or file issues (https://t.co/GEjTHSoa1J). 𝚌𝚞𝚛𝚕 -𝚏𝚜𝚂𝙻 𝚏𝚡.𝚜𝚑/𝚜𝚎𝚝𝚞𝚙.𝚜𝚑 | 𝚋𝚊𝚜𝚑 https://t.co/g2uEuXhGnt
What a great notification to get while vibe coding on a Friday night :)
@hedra_labs Theses are the *BEST* Tools on the market. I had Ai Toolkit|artist and let's just say" Hedra is my permanent home" thank you! @mjlbach
エージェント（特に画像処理や並列処理）で最大の課題は、良好なマルチプレイヤーオーケストレーションの構築と非線形ワークフローへの対応だった。 数ヶ月前に社内でこれを試してからは、標準的なプロンプトベースのツールをほぼ使わなくなった。Hedraの統計情報を見るのも含めて、ほとんどのリサーチワークフローを新しいworkspacesで実行している。 FigJamとClaude Codeと強力なビジョン言語モデルを合わせたイメージだね。
原文を表示 (en)
The biggest challenge on working with agents (especially for visual or parallel work) has been building good multiplayer orchestration and accommodating non-linear workflows. I've basically stopped using standard prompt based tools since we had this internally a few months ago. I do most of my research workflows (including looking at Hedra stats) in our new workspaces. Think FigJam meets Claude Code meets powerful vision language model.
Hedra Agent just got a major upgrade! Introducing Spaces and Connections, making it easier than ever to launch your next high-impact business strategies and social media campaigns. Spaces keeps all your assets in one place, from initial research to your final assets. Plus, sharing with your team is now seamless thanks to direct integrations with Slack, Gmail, and your CRMs. Comment “HEDRA” and share this post to get our exclusive prompt guide and 20% off your first month’s Creator Plan (limited to the first 200 people)!
The new gpt image 2 destroys anything that is not nano banana and legitimately has a sense of humor. Joint VLM/generation models are insanely powerful.
The next era of Hedra is here, we still have a lot launching soon. The last month was a whirlwind: * Omnia -- the first audio to video world model with class leading consistency * Hedra Avatar -- our commodity model for 10 minute long audio to video with full multilingual support and speaker selection * Hedra Teams -- expanding our enterprise offering with workspaces, sharing, and more * Agent -- our new creative-visual intelligence, this has been by far our most popular product in testing. We're excited for the next chapter as we push beyond visual creative tools. The agent unlocks new opportunities for Hedra and the team and I are stoked for what's coming next.
Introducing Hedra Agent, the unified intelligence for visual understanding and creation. Go from idea to content with extraordinary efficiency. Describe the outcome you want, and the agent handles the formatting, platform adaptation, and style. Whether it is a social post, a product ad, or a launch video, Hedra Agent orchestrates it start to finish. Drop in your product page, get back a full campaign with hero shots, social formats, and copy. Share a trending format you like. Agent recreates it in your voice and visual style. Describe the video you've been picturing, Agent builds it from scratch. A self-improving partner for your visual content. The era of vibe creation is here.
If you've ever wrestled with getting a dozen AI models to actually work together... you'd understand how much value Sakib has added to the team. When I first met Sakib, we were in the process of overhauling Hedra's first-time user experience and adding inspiration natively throughout the UI. We didn't have the engineering bandwidth to do it at the time, so we started by hiring a head of engineering to start shipping fast. Pieces were moving, but we still needed someone who could orchestrate all these different models - language, video, image - into one cohesive workflow. Sakib was that guy. He figured it out on his own, made the most out of every model we work with, and got them to play nicely together. It's funny - I've never met Sakib in person, but he's become essential to how we're growing Hedra. (image below is Hedra AI-generated - looks real right?)
Update: just in the last week, we've 2x'd our daily active users. I think it's partly because, for the last couple of months, we've been completely heads-down building product. We got into a good rhythm of just building and shipping, and then started rolling things out again. 2 positive outcomes from this focus: - Growth took off - We're expanding the team (a lot). We're hiring marketing and GTM roles which is new for us because it used to be all engineering. This'll be an interesting next phase. Excited to kick off the year with such a positive push. Turns out the best marketing is just making something good.
Today we're launching Omnia - Hedra's first general purpose world model. A lot of video models today are still feeling like uncanny valley. The movements are exaggerated, the expressions make you a bit uncomfortable, and you can tell something's wrong even if you can't articulate what. We spent a lot of time post-training Omnia specifically to fix all of this. And I'm happy to report that when we ran testing, the results were noticeably different from what you can get with other models right now. Try it now at https://t.co/xRSZTfIeQ6
lol I'm taking credit for the name, the feature is really fun on my iPad
Introducing He-Draw Draw anything (stickman, dragon, race car) → we instantly turn it into hyper-realistic, oil painting, cyberpunk, whatever you want. Zero talent required. Your worst doodle is now better than 99% of prompts. We’re giving 1,000 credits to the first 500 followers who want to try it. Reply “HEDRAW” and I’ll DM you a code.
Probably a hot take from someone building an AI video company: I don’t think A24, TikTok, or any social network will be predominantly AI content in 10 years (even if the tech exists). People come to social networks for genuine connection and creativity. They want information, sure – but delivered in ways that feel human. That need isn't going away. I *do* think there will be new distribution channels that are geared toward types of content that are perfectly suited for AI: educational videos, fun skits, microdramas, legal case reenactments. But again, these are specific use cases that can’t replace the human element of social networks. Not a fan of the amount of alarmism in the industry around things we care about going away. I think instead of stressing about something that won’t happen, we should be optimistic about new ways to create content that wasn't feasible before.
We just rebuilt Hedra's UI and support tickets dropped 75% while usage shot up (especially conversion from signup to first video). Super proud of how we figured out a revamp was needed, and the 3 things we fixed. From the beginning, we’ve obsessively stayed close to customer feedback. Just this week, I reviewed 500 support tickets and answered 170 of them myself. This habit has repeatedly given us direct insight into where users get confused. Based on what we heard, we made three changes: 1. Bugs + engineering quality (I credit Jon for this improvement). On the old platform, videos wouldn't generate, downloads were slow, and people would click something and get no feedback. John noticed this and took us from a seed-stage experience to Figma-level engineering quality. 2. Information hierarchy. Design team reworked it and built a philosophy around guiding users through the experience so it's clear where you go to do things. 3. We added tutorials and automated our support bot to point people to help articles (launching more tutorial videos soon). We're still figuring out how to reduce friction everywhere, but this is a solid start.
I'm obsessed with running a lean team. Like so obsessed that I almost didn’t hire a cracked GTM expert that was perfect for the team. Last month, I interviewed Sandra for a sales rev ops role and I knew immediately she was fantastic. Smart, great energy, someone I wanted to work with. But I wasn’t sure if we were too early to hire for this role. For a while, Sandra and I went back and forth while I figured it out. We talked about growth, community, data, and eventually, I just decided to trust my gut. It told me to always hire good people when the opportunity arises. And I'm so glad I did. She brought a data-oriented approach to both sales and customer analysis which has massively increased retention. Since joining, she’s: - Overhauled our analytics system - Set up Snowflake and DBT - Gotten way more visibility into what our users are creating and why Her current project: working with Paul (Head of Marketing) and Justin (Head of Sales-Led Growth) on paid user acquisition (something we haven't done before). Thank you for joining the team, Sandra. Excited to see what you do next 🚀
OpenAI just sent us one of those 10 billion token plaques. Physical marketing is really having its moment in AI startups right now. Cursor ships branded keyboards; Clay sends stickers to customers. I’m seeing a lot of AI companies go that extra mile with targeted outreach to high-value customers because they want to create advocates who organically market for them on socials. It makes sense. When someone gets something physical, there's an added level of effort that shows you care. The customer will naturally talk about it (like I am now). Although this doesn’t really scale in consumer, it’s a great way to create continual buzz in B2B. We’re trying to figure out what our equivalent would be at Hedra. We're seeing explosive growth on the B2B side as we launch more business-oriented features, so when Paul joined as Head of Marketing, the first thing I asked him was “what’s our version of this?” Working on it. Expect something soon 👀
We went from having no GTM team to a cracked lineup in just a few months: - Justin from HeyGen leading sales - Sandra from HeyGen running rev ops - Paul from DocuSign leading marketing The last piece is growth marketing. Finding someone dedicated to growth is hard because you need a rare combo -- someone who understands internet culture but also has the strategic judgment to know what's worth investing in vs. what's noise. Shoutout to Scott Coleman who's been instrumental in teaching me how to spot excellence in sales, marketing, and growth hires (it all comes down to understanding what questions to ask). I've been interviewing some amazing people and will hopefully have more to share soon. In the meantime, if you know someone who'd crush this role - email us at careers@hedra.com
I ran our engineering standup every day for the first year of Hedra. I was hyper-involved, to the point of giving detailed feedback on every feature before it shipped... turns out I was a huge bottleneck. My team ended up having to wait on me for every decision. I was spread too thin to focus on what actually mattered, and honestly it was exhausting for everyone. Hiring Jon to lead engineering changed things. One of the first things I did was hand him the standup. He has the same product obsession I do, so I knew he could enforce that level of quality (or even higher) without me being a barrier. Same thing happened with sales and marketing. I used to join every sales call, think about paid ad spend, influencer partnerships, launches. Now Paul runs marketing and Justin handles sales. I jump in when I need to but they own the execution. This is what the shift from founder to CEO actually looks like. You find people who care about quality as much as you do, then get out of their way so you can focus on the areas that aren't getting enough attention.
I feel partly responsible for the AI slop epidemic... We've powered a lot of it at Hedra. And I think we need to be honest about what it is: low-effort content that chases novelty instead of substance. Before AI, there was a natural filter. You had to be skilled to create content that got eyeballs. Now anyone can generate Jesus made out of shrimp and watch it go viral. But that’s just what happens with new things. There’s a phenomenon where a new capability drops, everyone floods the internet with the same novel outputs, initial engagement spikes, and then it becomes background noise. I think AI slop is just this expression of novelty as people experiment with new technology. But ultimately, what's exciting is that it’s giving way to new forms of creativity to people who are learning these tools deeply and pushing what art can be. Yes, there is AI slop (honestly, not sorry). But there's also really amazing creatives that are coming out of this era as well.