Matthias Pichler

I got to give a talk at the AI Nerd Meetup about the cool things we do at @joinsauna with agent filesystems

Some of your favorite coworkers 𝐚𝐫𝐞 𝐠𝐞𝐭𝐭𝐢𝐧𝐠 𝐟𝐢𝐫𝐞𝐝 𝐟𝐨𝐫 𝐧𝐨𝐭 𝐛𝐞𝐢𝐧𝐠 𝐀𝐈-𝐟𝐢𝐫𝐬𝐭. Here's how to keep them. 🍻 It's rarely a talent problem. They just never poked at the thing. They don't know what an MCP is, they've never touched an API, "skill" means something from a job description. Whatever AI they have is just ChatGPT with extra steps. Meanwhile everyone who put in the setup work compounds. My fix used to be walking over to their desk and doing it for them. The MCP won't connect. A skill they need doesn't exist. Something's always broken, and twenty minutes in they're looking at me like "how long have you been tinkering with your own setup?" Fair. Years, actually. So we built group chat and spaces into Sauna. When you start a space, you're not setting anyone up. You're handing over part of how you think: your skills, your folders, your authed connections, the context you've built for months. They open a chat that already knows what you know. It's a little like Slack, except for doing the work instead of talking about it. No desk visit. No excuse. And you keep your Friday beer crew intact. Try for free go.sauna.ai/fx

Recap from last night’s amazing @FireworksAI_HQ × @Workato event with @HackerSquadSF 🚀 Huge thanks to @yuan_sihan, @KovacVeljko, @mat_pichler, and everyone who joined us for a great night of technical conversations. Follow @HackerSquadSF for more builder events and technical talks.

Tonight at @HackerSquadSF: three technical deep dives with @FireworksAI_HQ × @Workato @yuan_sihan shared a cost-effective advisor framework where an open-source worker model can delegate harder tasks to a stronger advisor model, improving quality without paying frontier-level costs at every step. @KovacVeljko explored graph-based agent memory, ontologies, and global context indexing — showing how agents can manage knowledge, resolve conflicting information, and support collaborative learning across multi-agent systems. @mat_pichler explained why Git falls short for real-time, multi-user AI agent collaboration, then introduced an API-first filesystem built for atomic transactions, binary support, and persistent sharing in knowledge work. Follow @HackerSquadSF for more technical talks and builder events.

Co-hosting a hack night with @joinsauna and @ElevenLabs on July 16 in SF. Demos from both teams, then a 90-min build sprint. Spaces are limited. DM me for an invite link.

2 hours into the launch and @joinsauna has built us a nice dashboard to track signups

Instead of spending the budget on a celebrity launch we're giving it to the first 2,000 users instead. $80 of credits, every week, until the $1M is gone. If it's good, you'll tell people. If it isn't, no amount of Margot Robbie was going to save us.

I've been running my entire engineering interview pipeline on it, taking away all the busy work and letting me focus on the enjoyable part of collaborative problem solving during the actual interview.

Every company is spinning up hundreds of agents that drift faster than humans do, feeding each other their drift as context. Compounding rot. Sauna is our answer. Shared memory, shared skills, one source of truth across a team. Live today. Code LAUNCH at app.sauna.ai

Context + Intelligence = Work Done. Labs racing on intelligence. We went all-in on context. Walk through of why we think Sauna is the most powerful horizontal AI agent: multilayered context that compounds, 3,000 integrations, browser use with persistent logins, proactive scheduled tasks, multiplayer. Unlimited Opus 4.6 included. Then it acts on all of it. Cloud native, parallel agents, real integrations. Whole separate video on that this week. app.sauna.ai to try the beta. Promo code at the end of the video ($200 off, first 50). Onboarding prompts in the replies.

@sama Is the graph wrong?

𝐭𝐡𝐞 𝐰𝐨𝐫𝐝𝐰𝐚𝐫𝐞 𝐥𝐚𝐮𝐧𝐜𝐡: 𝐛𝐫𝐢𝐧𝐠𝐢𝐧𝐠 𝐀𝐈 𝐰𝐨𝐫𝐤𝐟𝐥𝐨𝐰𝐬 𝐭𝐨 𝐥𝐢𝐟𝐞 🚀 𝐭𝐥;𝐝𝐫 we've spent the last months building triggers, tools & data sources for wordware - starting today, you can create AI workflows that actually run in your life without writing a single line of code. 2000+ integrations, english as the programming language, it looks like a document but works like magic. see what's possible with AI: app.wordware.ai/lp 𝐭𝐡𝐞 𝐫𝐢𝐬𝐤𝐲 𝐛𝐞𝐭 🎲 we started by building an infrastructure product for technical teams making AI agents. the high-ceiling, powerful platform got traction - 60 real companies using our API some paying 15k mrr, 10mm+ people used these agents. then our last launch happened. 400k new users hit the same wall: "love it, but can't use it without coding." then we realized in today's world everyone is a builder. so we took the leap: what if our technical foundation became the perfect launchpad to make wordware deployable for everyone? 𝐭𝐡𝐞 𝐦𝐢𝐬𝐬𝐢𝐧𝐠 𝐩𝐢𝐞𝐜𝐞: 𝐭𝐫𝐢𝐠𝐠𝐞𝐫𝐬, 𝐭𝐨𝐨𝐥𝐬 & 𝐝𝐚𝐭𝐚 𝐬𝐨𝐮𝐫𝐜𝐞𝐬 ⚡ for the last months, our team has been working nights and weekends to transform wordware from an AI platform that requires engineers to integrate, into something anyone can deploy. you know those AI workflows already in your life? the ones where you copy-paste between different AI chats, manually trigger actions, and piece together insights? now you can describe it once and automate forever. 𝐰𝐡𝐚𝐭 𝐰𝐞 𝐛𝐮𝐢𝐥𝐭 🛠️ • 𝟐𝟎𝟎𝟎+ 𝐢𝐧𝐭𝐞𝐠𝐫𝐚𝐭𝐢𝐨𝐧𝐬 - connect to all your favorite tools • 𝐝𝐨𝐜𝐮𝐦𝐞𝐧𝐭-𝐥𝐢𝐤𝐞 𝐢𝐧𝐭𝐞𝐫𝐟𝐚𝐜𝐞 - if you can write it in english, you can build it • 𝐫𝐞𝐚𝐬𝐨𝐧𝐢𝐧𝐠-𝐟𝐢𝐫𝐬𝐭 𝐚𝐮𝐭𝐨𝐦𝐚𝐭𝐢𝐨𝐧 - not just "if this, then that" but "understand this, reason about it, then act" imagine: your typeform lead comes in → wordware analyzes intent, enriches with research, calculates a lead score, and routes to the right sales rep with a personalized draft email. or: your email triggers a workflow → wordware determines importance, archives the newsletter, flags the urgent request in slack, and drafts responses that sound like you. your intent and taste, AI's execution. 𝐰𝐡𝐲 𝐭𝐡𝐢𝐬 𝐦𝐚𝐭𝐭𝐞𝐫𝐬 💡 • the future of AI is systems working for us behind the scenes. • we're making AI the reasoning engine, not just another tool in the chain. • traditional automation moves data. wordware understands what that data means. 𝐣𝐨𝐢𝐧 𝐮𝐬 🚀 we've raised $30M to build the AI Operating System - where workflows get built, shared, deployed and forked. no waitlists - we're giving out credits to help build this ecosystem. get started for free: app.wordware.ai/lp p.s. huge thanks to our team who pulled all-nighters, debugged on weekends, and somehow managed to ship 2000+ integrations while having fun next stop: the beach office with the wind/kite surfing rack 🏄‍♀️

4o image generation, coming soon to a wordApp near you (as soon as the API is out) 😉

𝐭𝐡𝐞 𝐰𝐨𝐫𝐝𝐰𝐚𝐫𝐞 𝐮𝐬𝐞𝐫 𝐩𝐚𝐫𝐚𝐝𝐨𝐱: 𝐬𝐭𝐨𝐫𝐲 𝐨𝐧 𝐡𝐨𝐰 𝐰𝐞 𝐛𝐮𝐢𝐥𝐭 𝐬𝐨𝐦𝐞𝐭𝐡𝐢𝐧𝐠 𝟒𝟎𝟎𝐤 𝐩𝐞𝐨𝐩𝐥𝐞 𝐬𝐢𝐠𝐧𝐞𝐝 𝐮𝐩 𝐚𝐧𝐝 𝐥𝐨𝐯𝐞𝐝, 𝐛𝐮𝐭 𝐜𝐨𝐮𝐥𝐝𝐧'𝐭 𝐚𝐜𝐭𝐮𝐚𝐥𝐥𝐲 𝐮𝐬𝐞 𝐭𝐥;𝐝𝐫 we built an AI platform that 400k people loved playing with, but couldn't actually use in their real lives without coding skills. on april 8th, we're fixing that with triggers that let anyone deploy reasoning-first AI workflows without writing a single line of code. join the waitlist to be first in line: tally.so/r/n0DlxZ 𝐟𝐮𝐥𝐥 𝐬𝐭𝐨𝐫𝐲: we've been watching something fascinating (and honestly, painful) happen with wordware over the past few months. 𝐭𝐡𝐞 𝐠𝐨𝐨𝐝 𝐧𝐞𝐰𝐬: we've had over 400,000 users come play in our playground. people built incredible AI flows, experimented with our editor, and created genuinely useful solutions. 𝐭𝐡𝐞 𝐡𝐚𝐫𝐝 𝐭𝐫𝐮𝐭𝐡: we lost most of them. and it's not because they didn't like it. 𝐰𝐡𝐚𝐭 𝐚𝐜𝐭𝐮𝐚𝐥𝐥𝐲 𝐡𝐚𝐩𝐩𝐞𝐧𝐞𝐝 (𝐚𝐧𝐝 𝐰𝐡𝐲 𝐰𝐞'𝐫𝐞 𝐟𝐢𝐱𝐢𝐧𝐠 𝐢𝐭) 🔍 the pattern was painfully consistent: - user discovers wordware, gets excited about the potential - they build something genuinely useful in our editor - they hit a wall: "how do I actually get this into my life?" - without a clear deployment path, they drift away frustrated the only users who thrived were our infrastructure customers who could implement the API in their codebase. they're building amazing products and scaling fast. in fact, tens of millions of people are using wordware through these products without even knowing it. but this is absolutely on us. we built a powerful AI engine and IDE, but left a gap between "I built something cool" and "this is now working for me in the background." 𝐭𝐡𝐞 𝐦𝐢𝐬𝐬𝐢𝐧𝐠 𝐩𝐢𝐞𝐜𝐞: 𝐀𝐈-𝐧𝐚𝐭𝐢𝐯𝐞 𝐰𝐨𝐫𝐤𝐟𝐥𝐨𝐰𝐬 ⚡️ here's what we realized: the market has a massive unfulfilled need. automation tools like zapier are great at connecting systems, but the real magic of AI isn't in the connections - it's in the reasoning. it's in having workflows that can actually think about your data with real intelligence. when your email arrives, you don't just want it moved from one place to another. you want an AI that can: - understand what the email is actually about - pull relevant context from notion, your docs, and other sources - orchestrate multiple specialized LLMs and research tools for the perfect response - make smart decisions based on context and history - generate truly personalized responses - extract insights that would take a human hours to find unlike general-purpose AI agents that feel like black boxes, wordware gives you complete explainability. you can see exactly how your flow reaches its conclusions, inspect each step of the reasoning process, and build trust in what it's doing and why. this transparency means you can rely on these systems for critical workflows. 𝐭𝐡𝐞 𝐰𝐢𝐥𝐝𝐞𝐬𝐭 𝐩𝐚𝐫𝐭? we're enabling all this AI-native magic in a document-like editor where the only programming language you need is english. no python, no javascript, no API docs - just write what you want the AI to do, and it happens. this is what makes wordware fundamentally different - we're not just passing data between systems, we're enabling AI-native workflows where the intelligence itself is the value. that's the gap we're filling that no one else is addressing. 𝐞𝐧𝐭𝐞𝐫: 𝐭𝐫𝐢𝐠𝐠𝐞𝐫𝐬 (𝐚𝐩𝐫𝐢𝐥 𝟖𝐭𝐡) 🔥 on april 8th, we're launching triggers directly in the wordware editor. your AI flows can work for you in the background without you needing to know a single line of API code. imagine: your customer support email arrives → wordware automatically analyzes sentiment, pulls relevant docs, drafts a personalized response, and updates your CRM – all with the reasoning power of multiple LLMs working together. your flow can automatically fire when: - an email hits your inbox - a slack message mentions you - a notion page updates - and 2,000+ more integrations 𝐜𝐨𝐦𝐞 𝐛𝐮𝐢𝐥𝐝 𝐰𝐢𝐭𝐡 𝐮𝐬 ✨ the future of AI isn't just in the tools we actively use - it's in the systems that work for us behind the scenes. our system fires within seconds, debugs clearly, and lets you chain flows in our document editor. some of you will take this even further and build actual products on this infrastructure. we'll be right here supporting you. p.s. if you're one of the 400k who tried us and left - we get it, and we're ready for you now. this time, your creation won't just live in our editor. join the waitlist: tally.so/r/n0DlxZ also come join our team: jobs.ashbyhq.com/wordware.ai

Thanks! I really enjoyed the event, thanks to @itsajchan for organizing.

want to build culture for the next 10-20 years? 𝐞𝐦𝐛𝐫𝐚𝐜𝐞 𝐭𝐡𝐞 𝐰𝐞𝐢𝐫𝐝. throughout my time in startups, i learned that it's not going to be a quick 2-year sprint, and the people you surround yourself with make all the difference. it's my pleasure to start this series and introduce you to our whole team one-by-one. I feel incredibly grateful to work with these people ❤️  if you know more people like this, read on; you might make 15k 😮💰 first, let me introduce our founding intern, @isaacdyor (yes, that's his actual job title) think about this scenario: you've just been rejected by the startup you applied for. 99% of people would take the L and move on. instead, Isaac decided to run a marathon (???). when Isaac first applied to wordware, we said no. he was a sophomore at the university of Washington with no real work experience. we told him to reach out in three months' time, which he duly did. this time @bertie_ai had a call with him, decided he was still too junior, and rejected him again. Isaac, being a special type of unhinged, decided to prove himself to us by running a marathon with a giant 'W' on his chest. you can read his blog (isaac.dyor.com/posts/getting-…) for his thought process in his own words. had isaac ever run a marathon before? 𝐧𝐨, but that wasn't going to deter him. his phone died and he ran out of water, but he completed it after eight hours. he then sent us the video evidence to try and change our minds. to other people, this may have been a bit of a red flag, but for us, it was the perfect level of crazy. he dropped out of college for the quarter and arrived in san francisco two days later. two weeks ago, he became a full-time permanent employee 🎉🎉🎉 i asked Isaac what attracted him to wordware and he said "just like epicness, you know". Isaac wanted to do something really big (drop out of college), so he needed a really good reason to do it. wordware is that reason. 𝐰𝐡𝐚𝐭 𝐡𝐚𝐬 𝐰𝐨𝐫𝐝𝐰𝐚𝐫𝐞 𝐭𝐚𝐮𝐠𝐡𝐭 𝐈𝐬𝐚𝐚𝐜? that he was way worse at coding than he had thought, working on a real-life complex codebase is a whole new ball game. he said he learnt more during his two-week trial at wordware than he did during his first year at college. 𝐢𝐬𝐚𝐚𝐜'𝐬 𝐠𝐨𝐚𝐥𝐬 𝐟𝐨𝐫 𝐭𝐡𝐞 𝐧𝐞𝐱𝐭 𝐭𝐡𝐫𝐞𝐞 𝐦𝐨𝐧𝐭𝐡𝐬: - create the best AI workflow platform the world has ever seen - get a girlfriend if you know someone who could be interested in being Isaac's girlfriend, refer her, and if she goes on a date with him, he made me promise I will pay you $50. what a legend! However, if you know someone who can help us build the best AI Agent platform the world has ever seen, refer them, and if it works out, 𝐰𝐞 𝐰𝐢𝐥𝐥 𝐩𝐚𝐲 𝐲𝐨𝐮 𝟏𝟓𝐤. LFG more to come! enjoy the video of Isaac below 😊

most common questions i get: so what AI agent should i build? how can AI be useful for me? today, I'm starting a series of videos where I'll build alongside you - covering both personal workflows and work automation. the main syntax of wordware is just plain English. the only other thing is to press slash ( / ) and get started. let's start simple 😊 please open a wordware window and build with me! i enabled every new account to have 5 USD of free credits 💰 many of these will showcase our new feature: triggers & actions. until now, only our infrastructure customers who used wordware with API could fully utilize what they built. the rest of you were stuck in the playground. very soon, you'll be able to set up thousands of triggers and actions inside your Wordware flow. imagine a voice note WhatsApp message triggering a flow that researches with perplexity, uses multiple LLM providers to process the transcript and then saves data to your CRM and shoots off a Slack message when some criteria is met. this isn't just automation - it's the operating system for AI we've been building towards. sign up to join the alpha of triggers & actions here: tally.so/r/n0DlxZ just a recap: what is wordware and what makes it different? wordware is natural language programming for building AI Agents. unlike other no-code tools that box you in, we're building a true programming language with all the power - conditionals, loops, type systems - but in plain english. not just another drag-and-drop toy.

@zeeg Option 2 using sparse fieldsets from the JSON:API standard: #fetching-sparse-fieldsets" target="_blank" rel="nofollow noopener">jsonapi.org/format/#fetchi…

Alright API fanatics. You have an endpoint responsible for returning an objects details. In some cases you need only a few fields, and in others (e.g. edit) you need a lot. Do you: - Use seperate endpoints? - Make it variable (i.e. what gql does) - ???

@thdxr The same way Uber Eats still involves cooking…

imagine being the guy who thinks “serverless actually has servers” is a good point to make

@theburningmonk To me it seemed that the pricing aligns with that of other players( e.g. @osoHQ or @AuthressHQ ). But they are all very expensive IMO. We use @AuthressHQ and have 2 layers of caching in front (in memory in Lambda & fronting their API with CloudFront).

I might have misunderstood the use case for Verified Permissions, but if I was to authorize every user req (as described by the example), it'll be wildly, orders of magnitude more expensive than my application (e.g. API GW + Lambda + DynamoDB)... 🤯