Tomas Halgas

JUST IN: Peter Carlsson and Siddharth Khullar emerge from stealh with a $10.7M pre-seed to launch Aris Machina the agentic OS for modern manufacturing. 🤖 The Stockholm-based startup is developing an agentic OS that unifies factory data and enables real-time reasoning, adaptive workflows, and self-improving production. Here are the details: 🔥 $10.7M pre-seed led by Earlybird Venture Capital, with Village Global, AENU, Planet A Ventures, SNÖ Ventures and top angels. 🔥 Agentic OS that connects machines, systems, and data streams into one unified platform. 🔥 Two core workspaces: Gemba (real-time factory intelligence) and Protos (R&D, design & physics-informed modelling). 🔥 Founders: Siddharth and Peter previously built and scaled systems at Tesla and Northvolt. Congrats to Siddharth, Peter, and the team! 👏 Let’s go 🇪🇺 --- 📌 Update: Just interviewed Siddharth Khullar on New Wave, episode is now live. Listen here: https://lnkd.in/eNfN38Td --- Want to keep up to date on European climate-tech news? Follow me Hugo Rauch or subscribe to New Wave: https://lnkd.in/e5HmZAZA

756

46 Comments

Greylock is leading the seed in SuperMe, the AI-native professional network. Most expertise is not yet legible to LLMs. Professionals demonstrate expertise in private contexts: product docs, meeting notes, and most often - in their own heads. SuperMe is building a way to extract and structure these insights into AI profiles, grounded in an expert's real words and personality. And it is network-first; the more conversations one's AI profile has, the more opportunities for experts to reinforce or create new insights -- expertise compounds. This is one of the strongest examples of founder-market fit we've seen at Greylock. Casey Winters is the most in-demand product & growth advisor that marketplace operators (self included) wish they could hire. And Ludo Antonov built & led engineering teams at some of the most iconic network businesses over the past decade: Whatnot, Lyft, and Pinterest (where he worked closely with Casey). The two started this company out of our offices, and have until now been quietly building a highly curated network with profiles like Elena, Lenny, Brian, etc. We see a unique opportunity to create the professional network graph in a world where likes & follower counts have lost meaning. More on our investment below:

286

13 Comments

We won Adaptyv's protein design competition in December with an AI-generated antibody variant against EGFR. Only 1 was tested, we had a bit of a "one more thing" moment last week: After in-vitro validation of binding and expression we learned all 12 of our designs would have taken position 1-12 in the competition. Three even achieved sub-nanomolar binding, outperforming our original competition winner. Here's what makes this remarkable: 1) Our (at the time) intern did this in a few clicks on Cradle, beating 130 teams of highly trained computational biologists using state of the art open source models 2) In-silico metrics completely missed these designs - they ranked between 322nd-623rd in computational predictions and were deemed not worthy to be tested. Be careful trusting (structural) computational metrics, you may be selecting out a lot of goodness. 3) The designs were highly diverse, with significant edit distances between variants, showing we thoroughly explored the sequence space. And for those worried about immunogenicity of ML-generated antibodies: These designs maintained or improved humanness scores with no increase in immunogenicity risk compared to the template. This is what Cradle was built for: making biologics engineering a few scientists can do in a garage without any knowledge of GenAI. There's a lot more in our full article (link in comments)!

333

31 Comments

was at OpenAI hq yesterday and got to talk with their head of product about the new agentkit 🤯 a lot of people think this kills zapier (i even thought so at first). it doesn’t. agentkit is for devs, not no-coders. it’s about putting ai inside products, not automating workflows. this makes it easier for teams to build agents that live inside their own apps, not connect tools together. so yeah, this isn’t about replacing zapier. it’s about replacing ui. #ai #openai #chatgpt #agents #startups #developers

124

8 Comments

Dear Founders, AI hype? A bit too much out there. I use AI. I love AI. I back AI. I invest in AI. I think the future is AI. BUT… for non-technical founders? AI hasn’t moved the needle. It didn’t level the playing field. Not even close. You might see further. You might move faster. But you still can’t build the bridge. What AI does for non-technical founders: - Helps you mock up an MVP - Writes decent code snippets - Builds landing pages, workflows, pitch decks - Accelerates ideation and insight All useful. None game-changing. What AI doesn’t do (yet): - Build scalable, secure systems - Replace a real CTO or technical cofounder - Architect a product for 10,000+ users - Debug in production - Convince a VC you can actually ship So yes, AI is a booster. But it’s NOT a builder. Not yet. Wondering why VCs aren’t biting? It’s probably not your pitch. It’s that you’re still missing the right HUSTLER + HACKER combo. Until you find it… You might be a startup. But you’re not VC-attractive or VC-backable. Not yet. Yours truly, David PS Like this? Join DavidBeatsGoliath.co—weekly newsletter for underdog founders. Join Founder Chapel (https://lnkd.in/gH-anZvp AMA + discussions among authentic founders.

90

45 Comments

Very excited for our new investment in NobodyWho, a Copenhagen-based startup pioneering the future of Small Language Models (SLMs). While the world races to build massive, energy-hungry cloud LLMs, NobodyWho is taking a different path: bringing AI directly to your device. Their open-source engine allows developers to run powerful SLMs locally on laptops and phones—no servers required. 🚀 Why we are backing them: * Sustainability: Drastically reduces the carbon footprint of AI (up to 100x lower footprint!). * Privacy: Data stays on the device, ensuring true sovereignty and security. * Efficiency: Enables organizations to break free from cloud lock-in and high compute costs. I remember the first time I spoke to Cecilie Waagner Falkenstrøm many years ago creating art using AI way before the AI-hype, so ahead of her time and visionary. Now teaming up with Asbjørn Olling, achieving fast adoption from 5,000+ developers. We are proud to join PSV and The Footprint Firm in this €2M round. Welcome to the Norrsken Evolve family! 🤗

122

5 Comments

I’m a co-founder of an AI infra startup, yet I’m not an engineer. Here’s how I make it work in a tech-heavy niche: 01 Own vision-to-market fit My core role as a non-technical co-founder is to work with vision and ensure it resonates with investors and the market, but also aligns with what’s technically achievable. After each round of vision adjustments or investor feedback, it’s my job to recalibrate with the tech team, figuring out what’s realistically achievable (especially given the size of the team relative to our capacity and ambitions). 02 Founding team chemistry We’re lucky to have exceptional engineers among the founding team, who regularly sharpen our vision (both technically and strategically) helping us stay ahead. And by “us,” I mean me and my co-founder, who also helps me navigate deep-tech discussions and explains ML concepts in plain English. But we never drew a hard line between “tech/product” and “biz/strategy/fundraising.” We constantly intersect and build together. So my advice would be: besides hiring the best people you can find, trust them fully and share ownership of strategic decisions. 03 Do reality checks constantly AI is evolving bloody fast, you know it, and it’s challenging to keep up with all updates. There’s also a growing gap between deep expertise and surface-level understanding in the market, sometimes even among people you expect to dive more deeply into the technical specifications. To navigate this, I constantly stress-test our technical ideas with entrepreneurs and devs, the people for whom we’re building the solution in the first place, not just with investors. We’re building something investors excited about, but customers understand more clear and use. The latter is more important. 04 Dedicate the time to learn I dig into research pages, follow relevant people and Slack channels, and leverage internal bots to stay updated about all news around foundational and infra levels of AI. Karpathy and researchers from frontier labs are my daily reading. Also, yes, it means brushing up on math and physics fundamentals. If you're in this field and not an engineer, staying updated is your *second job*. It’s non-negotiable. Same goes for understanding the fundamentals. Because if you don’t speak the language of your engineers, don’t expect to be taken seriously. Even with a background in building a GenAI company before the LLM hype, if I didn’t keep training myself, I’d lose touch with both our team and product reality. 05 Mutual respect and trust Understanding the technical landscape at this point is simply = building credibility. I’d probably lose my team’s, partners’, and investors’ trust if I started BSing rather than listening and learning fast about tech specifications. Bottom line: In deep-tech, having a stellar team is essential. The value of talent here is easily 3–5× higher than in consumer tech. No surprises, great engineers today can have Ronaldo-level sign-up bonuses.

193

22 Comments

Controlling a character’s motion is super easy now with Kling AI 2.6 💃 Finally got a chance to test this out myself. Check out the results below! A few surprising details: 1. Physics: Look at the hair in the underwater shot - it actually understands the weight and flow based on my head movement. 2. Omg - EYE MOVEMENT!! 👁️ One of the hardest things I’ve found in previous models using a similar technique (video to video) is capturing the eye movement in the original video. You can see in the Avatar test, it’s actually able to capture the blinking and where my eyes move back and forth which is crazyyyy! 🤯 3. Backgrounds: Hard to prompt for background movement. For the Avatar test I prompted for the trees in the background to sway and move but it kept it static. Maybe I just needed to generate a few more and one of them might’ve come out different? 🤷‍♀️ Can't wait to use this on some skits we have planned! This test was inspired by Eric Solorio's awesome videos - y'all gotta check him out! 🎉

148

32 Comments

Cannot be happier to have been invited on board Alexis Richardson, Brian Grant, and Jesper Joergensen's new adventure. Say hello to ConfigHub Inc! A whole generation of startups was built to declare, automate, and scale configuration. But they all followed the same model—managing config using files, variables, and increasingly complex rules. The deceptively simple idea the team posed to us was: what if, instead of files and variables, we used rows and columns? What if we treated config like data? Things get very interesting very quickly. 🤓 💻 We co-led ConfigHub's first $4M alongside Pear VC, Encoded Ventures, and a venerable list of angels. Read more at the links to my post on Crane's blog and Techcrunch in the comments! #vc #startups #configurationmanagement

129

16 Comments

WARNING - most of what you heard about AgentKit is wrong. The new Agent builder by OpenAI is not designed to kill n8n and Zapier. It's designed to do something else entirely. My co-founder Matan Protter (PhD in Machine learning, 9 patents) and I researched the tool capabilities and understood what it's designed for and also why almost everyone got it wrong. Here are the top 5 takeaways: 1. While the tool has a similar interface to n8n it's not designed for workflow automation. How do we know this: No ability to set triggers, limited support in tools, none of the templates has any external tools in it. 2. The tool's main purpose is to build chat experiences (it's text input, text output) How do we know this: You can look at the video open AI published (link in comments) - the example they use is building a travel agent (chat experience). The leading export option is called ChatKit. 3. If it kills anything it probably kills GPTs - the apps that OpenAI allowed user to create were very limited in functionality. This allows to design better chat flows and share them in more ways. How do we know this: OpenAI is attempting to build an eco-system to create moat but they don't need two ecosystems. GPTs didn't make it to the mainstream and usage is limited. 4. AgentKit is designed for creating a chat for your product. It allows you to have OpenAI handle all the conversation while controlling the logic How do we know this: Both ChatKit and Agents SDK are geared towards that. 5. OpenAI wants you to build assistants to handle multi-prompt workflows leveraging states and logic. They realized the limitation of LLMs to handle logic in the prompt and are compensating for that. How do we know this: The templates they shipped are all multi-prompt workflows. It's like combining 5 GPTs into one workflow. So, I think all the AI hype influencers that claimed this is a Zapier/n8n,make replacement are wrong and in fact this is something else entirely. Happy to be proven wrong - let me know your take in the comments.

145

21 Comments

AI slop is everywhere and it's very annoying. Last week we sent several questions to 4 freelancers we pre-selected about a simple project. 3/4 answers were generated - with the same "insights", same wording and all that. We ended up working with the last one, not because they are the best for this job - simply because they "stood out" by replying without AI. It was not an ideal answer, but at least there was a human brain thinking for a second what to answer behind this response and not a copy/paste. If we speak about hiring, cover letters are a thing from the past already - over 80% were AI generated from the recent candidates. Probably even 100%, but at least, 20% remaining were a bit more thoroughly generated to avoid looking like a generic slop. It's damn scary to think how low is the bar here. And what's most important, models are trained on the ever increasing amounts of AI slop. It creates a self-compounding bad feedback loop. I've read in the news today, that Google snippet gave a great review for Mein Kampf as it just digested Amazon AI summary of the same book from Amazon website, that was created on the ironic comments from the review section. Obviously, model didn't understand that and Google referred to the information from Amazon as Authenticity is a premium currency of the internet. I personally choose not to use AI for any of my public comments. Let them be sometimes silly, sometimes without a perfect grammar, but they will be mine.

40

15 Comments

"Who are we building Lyzr AI for?" This is a question that I often get from VCs, customers, and partners. Are we building Lyzr for pro devs? The kind of developers you'll find in a startup or a typical Bay Area company? Not necessarily. (Though the product can be, and is, used by them.) So, who are we really building for? We are building for the largest developer community in the world: the enterprise devs. Over the last ten years, I've had a chance to work very closely with enterprise developers. Unlike their startup counterparts, they work under a lot of constraints: security, guardrails, governance, change management, etc. They spend far more time in testing and shipping features. It was this critical insight that made me think about building an agent framework back in 2023—one that was much simpler and more straightforward than LangChain. But I soon realized it's not just the framework that enterprise developers need. They also need a practical user interface so they can spend less time in an IDE and more time integrating agents into their existing enterprise software. And by the time we completed building Lyzr Agent Studio, we realized we had built a platform that could also be used by business users, too. Today, 33% of all Lyzr users are business users who do not write any code but build agents and multi-agent systems. So, who are we building Lyzr for? The 99% of an enterprise who actually run the business. Enterprise Devs and Business Users.

49

1 Comment

Sprints are dead. We are inventing a new framework for shipping product - Holocratic Development In a world moving at AI-speed, two-week sprint cycles feel like planning with a blindfold on. The pace of change has outpaced the pace of planning. Market dynamics shift daily. Coding tools evolve weekly. What felt important on Monday might be irrelevant by Friday. Yet we still cling to rigid sprint structures, ceremonies, and rituals — hoping they’ll keep us “agile.” But agility isn’t about sticking to a two-week plan. It’s about responding to reality in real time. That’s why we’re embracing a new model: Holocratic Development. A development system built for speed, autonomy, and adaptiveness — where execution happens at the edge. In Holocratic Development: - People closest to the problem make the decisions - Work flows continuously — not in scheduled bursts - Teams self-organize, driven by shared context, not dictated roadmaps - Feedback loops are real-time, not retroactive - This isn’t chaos. It’s high-context, high-trust execution. Sprints, planning poker, backlog grooming — they slow you down when the environment demands responsiveness. Holocratic Development replaces these rituals with constant alignment, empowered teams, and live prioritization. Agility is no longer about process frameworks. It’s about how fast your team can sense, decide, and ship. Lets see how it goes ..

59

10 Comments

The most important skill in AI right now? 𝗣𝗿𝗼𝗺𝗽𝘁 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴. 🤠 But no one wants to say it out loud. It hurts to say this, but most of the more hard-core ML I've done is only useful in very niche cases (and jobs). 😢 Especially when working with Large Language Models. When talking to people at OpenAI or Anthropic, they actively discourage even fine-tuning in most cases. Why? Because in 90% of scenarios, prompting is faster, cheaper, and more reusable (across newer models). 🤐 99% of companies will use LLMs, < 1% will have custom models (or even access to the weights) that you can work with. And even for those that do, prompting is a must-have skill. 🤷‍♂️

131

8 Comments

🚀 Will AI take your job—or supercharge your career? 💡 Are junior developers locked out of the workforce? Not so fast. I was glad to take part in the AI Subcommittee in the Knesset, where we tackled these questions head-on. Our research shows 92% of engineering teams use AI, but only 16% have adapted workflows. Senior engineers hesitate to trust AI-driven code, making it harder for juniors to gain experience. But here’s the thing—juniors are often the fastest adopters of new tech. Once companies build the right AI workflows, GenAI won’t replace juniors—it’ll turn them into seniors faster. The opportunity isn’t disappearing—it’s shifting. Those who learn to work with AI, not against it, will lead the future. 🙏 Thank you Orit Farkash Hacohen for leading this critical discussion! Our research and an article by TheMarker in the first comment

142

10 Comments