Danilo Joksimovic

And we're off! Our first annual conference SCULPT just kicked off with three of our biggest feature launches yet: 𝗦𝗰𝘂𝗹𝗽𝘁𝗼𝗿: Your AI GTM co-pilot. Build live workflows or table analyses in natural language conversation to save hours of manual work. (Try "Build me a list of CEOs at mid-stage healthcare companies in SF along with their work email" or "Analyze my list of closed-lost deals and tell me why accounts didn't buy.") 𝗔𝘂𝗱𝗶𝗲𝗻𝗰𝗲𝘀: Removes the 50K row limit. Sync millions of records, combining signals, enrichment, and CRM data in a single table for precise targeting. Create highly specific segments by layering multiple data types without stitching together separate tables. 𝗦𝗲𝗾𝘂𝗲𝗻𝗰𝗲𝗿: Go from signal to campaign in minutes. Send native email campaigns without leaving Clay, with data and bulletproof deliverability baked in. These launches are some of our biggest this year, and we hope they will help you come up with, experiment and execute your best growth ideas faster than ever. Read the full feature details in our post linked below.

476

40 Comments

🚀 Big news for AI agent builders: Agent Reinforcement Trainer (ART) now integrates seamlessly with 🦜🔗 LangGraph! If you’ve ever built multi-step, ReAct-style agents, you know the pain: endless prompt tweaking, fragile reasoning, and lots of manual trial-and-error. With this new integration, you can now train your LangGraph agents with reinforcement learning — making them smarter, more adaptive, and more efficient over time. ✨ What this means in practice: * Agents learn to reason through multi-step problems without hand-holding * They improve tool usage automatically (no more prompt spaghetti) * They make smarter decisions in complex workflows * Training is scalable and robust, powered by ART’s reinforcement engine And the best part? It’s a drop-in upgrade: * Swap in ART’s init_chat_model and wrap_rollout * All interactions, tool calls, and reasoning chains get logged automatically * Plug straight into ART’s RULER reward function for training — no custom rewards needed 💡 Imagine your LangGraph prototypes evolving into production-ready agents that actually get better with use. That’s the power of combining LangGraph’s agent framework with ART’s reinforcement learning. 👉 Full docs & examples here: https://lnkd.in/gwWj-xxx Can’t wait to see what the community builds with this. If you’re experimenting with multi-step AI agents, give your agents the ability to learn from experience!

1,889

80 Comments

Excited to announce Structured AI is backed by Y Combinator! We’ve now raised $1m to bring our vision to life: what if AI could give design engineers in construction a full day back every week, just by automating the repetitive tasks they hate? Y Combinator is the birthplace of category-defining companies like Airbnb, DoorDash, Dropbox, and #ConTech pioneers like PlanGrid, an Autodesk company and EquipmentShare. But what excites me most is working with Tom Blomfield, who founded two unicorns (Monzo Bank and GoCardless) and has been an incredible group partner. Our mission stays true: make something people want. For us, that means talking to users every day, shipping fast, and measuring real impact on engineering quality and time. We've flown to 12 cities, sat in engineering offices, and seen firsthand how talented engineers lose hours to manual workflows that could be automated. Every conversation shapes our product. We believe the future of engineering should be about problem solving and creativity, not paperwork and endless emails. If you're an engineering leader in #AEC looking to redefine your workflows with cutting-edge AI, we should absolutely talk. Issy, Brandon

887

251 Comments

There’s been a lot of recent news about AI’s scheming. Last week Anthropic Claude Opus 4 blackmailing developers, this week ChatGPT o3 refusing to switch off. These should be a wake-up call for everyone building AI. When AI starts tampering with code and bluffing its way through tests, it’s no longer just a technical achievement. It’s a moral risk.   Technology for good means, if it isn’t yet proven good, it shouldn’t be released as the engineering isn’t finished. Just like the FDA wouldn’t approve a drug until its target efficacy is high and its side effects are thoroughly tested, neither should powerful AI models be released until they’re understood and proven safe.   This is exactly why at Gloo we don’t just ask, "Can we build it?" We ask, "Should we?" and "Will it serve human flourishing?" Tech advancement, especially AI, must always first be measured by its impact on our collective flourishing. Our models are not just evaluated for performance. They are aligned with values like truth, human dignity, and faith. External audits? Absolutely. Guardrails? Required. Ethical alignment? Non-negotiable.   This technology is moving fast. If we don’t ground our work in values now, we may not like what takes root later. https://lnkd.in/gjbdCHUk

499

43 Comments

Design partners might be the most overlooked secret weapon in building something people actually want. Before writing a single line of code at Slintel, we sold Excel spreadsheets for $500/month to founders who needed tech stack data, as a service. No product. No platform. Just manually compiled spreadsheets. Those first few customers weren't just "early adopters" – they were design partners who shaped everything. They weren't buying our vision; they were helping us build it. Fast forward to today. I'm building again in the Agentic AI space, and the first thing I did? Found design partners. Yes, doors opened faster this time (a successful exit helps) and I've made some friends along the way. But the approach hasn't changed. Here’s the framework we used to select our Design Partners: 🔹 Conducted 40+ customer interviews with target personas to deeply understand their problems. 🔹 Presented our solution to gauge which customers would be immediate movers. 🔹 Narrowed the list further by evaluating responses to five key questions: 1️⃣ Is this a top 5 problem for the persona? (The most critical factor. This helps determine budget allocation. In challenging markets, only the top 3-5 problems get funding, while everything else is seen as "nice to have.") 2️⃣ Will they build in-house or buy? (Assesses their mindset. While not as crucial as the previous question, most companies will prefer to buy if they can't build everything internally.) 3️⃣ Will they commit to bi-weekly calls for the next few months of build? (Evaluates the importance of the solution. Senior executives won’t commit time unless it’s a pressing pain point.) 4️⃣ Will they deploy for their end users? (Assesses their willingness to roll out the solution.) 5️⃣ What’s their timeline for deployment? (Evaluates urgency and readiness.) And of course, these design partners have stuck with us throughout the build phase and some of them have become our first real customers. Their relationship is deeper than a transaction – they're invested in seeing you succeed because they've been part of the journey. And you’ve made their voice and their needs key to shaping your product. In a world where most startups fail because they build something nobody wants, design partners are your insurance policy against irrelevance.

117

6 Comments

Unpopular opinion: The RL hype is a red flag for technical depth. When I hear founders or VCs casually dropping "reinforcement learning" in pitches, it's usually code for "we don't actually understand our problem space." Here's the reality check: RL only works when you have: - Perfect simulation environments - Extremely fast iteration cycles - Clear reward functions - Tolerance for massive compute costs That's why it succeeds in games (perfect simulation) and search optimization (Massive compute capacity at Deepmind AlphaGo/Anthropic/OpenAI) but fails spectacularly in the messy, slow-feedback real world. The graveyard of "RL-powered" startups tells the story. They burned through $$$$ trying to apply algorithms to problems that needed simple heuristics or supervised learning. Most of the time, a well-tuned evolutionary algorithm will outperform your fancy RL setup while using <1/10th the compute and delivering results you can actually explain to stakeholders. The next time someone pitches you an RL solution, ask them: "Have you tried solving this with a basic optimization approach first?" Watch how quickly the conversation changes. I am surprised when people are surprised that Direct Preference Optimization (DPO) can exceed PPO-based RLHF [1] or Reflective Prompt Evolution Can Outperform Reinforcement Learning [2]. [1] https://lnkd.in/gCY3bcTQ [2] https://lnkd.in/gQRwyQyM

45

2 Comments

While we haven't officially launched, can't help but be excited to see this great artcle in the @WSJ featuring LGND AI, Inc.(http://lgnd.io) and the opportunity for AI applied to earth observation data for all sorts of important uses. This is a huge and important data source that is vastly under represented in the current AI platform shift - and we aim to change that. Big stuff ahead! https://lnkd.in/ghd76mBB

89

9 Comments

We're officially stepping into the world of visual intelligence. Until now, Clueso (YC W23)’s intelligence layer worked entirely on audio – transcribing voiceovers and using LLMs to infer context, structure, and pacing. But what about silent videos? With our latest release, Clueso can now automatically generate a voiceover script for videos that don’t have a talk track. This is our first step into analyzing visuals and understanding what a video is trying to say – even when no one’s speaking. And we’re just getting started. Soon, Clueso will: 🎯 Automatically redact sensitive info on screen ✨ Add visual effects that focus on the right sections automatically and a lot more! Super excited for what’s ahead. Visual intelligence unlocks a whole new layer of possibility.

187

18 Comments

I’m interested to see if AI in building market-viable products (MVPs) will become a game-changer in 2025. Here's what I’m noticing: 1) AI tools are now being integrated into the earliest stages of product design, from prototyping to customer validation. 2) There’s a growing emphasis on predictive analytics to forecast market demand even before the MVP hits the market. 3) AI-powered customer feedback analysis is drastically reducing the time needed to refine MVPs post-launch. According to McKinsey’s AI adoption report: 1) Over 70% of companies using AI in product development report a significant reduction in time-to-market. 2) AI-driven market research tools are increasing accuracy in identifying customer pain points by up to 85%. 3) Companies leveraging AI to test MVPs report a 2x improvement in conversion rates after iterative changes. A big part of creating and scaling market-viable products with AI will come from: 1) Automating research and prototyping: Leverage AI tools like ChatGPT or DALL·E to brainstorm concepts, simulate designs, and conduct early-stage testing. 2) Using data-driven insights: Apply predictive models to identify market trends and assess risks before launching an MVP. 3) Improving iteration cycles: Integrate AI-powered tools like sentiment analysis or A/B testing platforms to refine products rapidly based on real-time feedback. AI isn't just a tool—it’s becoming a co-creator in the product development journey. What are your thoughts on the role AI will play in shaping MVP success? #MVP #Technology #AI

183

33 Comments

AI coding agents remind me of the early tech teams at Next Games. At Next Games, our tech team built everything from scratch: backend, live ops, integrations. They were creative, opinionated, and proud of their work. You could sit down with them and discuss why certain solutions made sense. Working with AI coding agents feels similar, just without the room full of people. You need structure to keep them consistent. AI agents can build massive things in days, but they still need a human in the loop. They don’t remember context, past decisions, or how systems fit together. It’s like a CTO who goes on holiday and forgets what happened before. They’re brilliant at focused tasks. Give them perfect context and they fly. Miss a detail and they stall. That’s the game now, learning how to manage creative machines.

36

9 Comments

the last few weeks feels like a mini breakthrough for AI apps hiding in plain sight - partial autonomy, high agency models, and context eng vs prompt eng. are all incredibly important concepts a lot of these principles were apparent but were just recently articulated (mostly by Karpathy - legend!!). one is the concept of partial autonomy apps - autonomy sliders, app specific UI, and “keeping the AI on a leash” two is the idea of small, high agency models that know how to solve problems vs knowing answers via tool-use / reasoning / multimodality (“maximally sacrifices encyclopedic knowledge for capability”) three is context engineering vs prompt engineering - putting the right stuff into working memory, building around the “jagged intelligence” of models etc these should probably be the starting point for founders thinking through AI native apps Refs: Partial autonomy apps: [Karpathy @YC startup school] High agency mini models: https://lnkd.in/gthCXtMe Context Eng: https://lnkd.in/gha2i7BX

101

10 Comments

For those building in the Agentic AI space, one thing becomes clear quickly: you can't improve what you can't see. That's why I'm excited to share that RagaAI Catalyst now integrates directly with CrewAI. If you're building multi-agent workflows with CrewAI, you can now: ✅ Trace agent behavior in real time ✅ Test your agents before production ✅ Monitor performance and failures after deployment The integration is built for observability. Every action your #CrewAI agents take can be logged and analyzed through Catalyst. This gives you a clean audit trail and deep visibility across the lifecycle of your AI systems. This has already numerous enterprises ship faster and catch issues before they reach users. Now it’s available for anyone building with CrewAI. Let’s build more reliable agentic systems! To Learn more: https://lnkd.in/gDFSemnH #AIOps #Observability #AgenticAI #GenAI

36

2 Comments

Just spoke to a YC founder friend. He used to spend hours coding. Now he talks to AI all day. His new process for building his product: 1/ Use WisprFlow with Cursor to speak in granular detail about what he wants the AI to build 2/ Watch Cursor write the code 3/ Make edits when needed He's able to build multiple features simultaneously without typing anything. The speed at which engineers can ship code has never been higher. Companies are building products in days, not weeks or months. If you're not using AI too, you'll be left behind.

97

31 Comments

Just 2 hours after Anthropic released Claude 3.7, Justin Torre integrated full support for it within Helicone (YC W23). In case you missed it - Claude 3.7: • Outperformed other models on SWE-bench Verified and TAU-bench for software engineering tasks • Implements reasoning with configurable "thinking budgets" extending to 128K tokens • Works with Claude Code terminal-based assistant for local file modifications • Maintains the same pricing as 3.5 Sonnet ($3/M input, $15/M output tokens) If you're interested in the technical details, check out our analysis of the available information w/ real-world examples 👇 (link in comments)

35

3 Comments

Everyone’s hyped about reindustrialization but here’s the truth. Factories and grind aren’t enough. The real move is making hardware faster and cheaper to develop. That’s what changed the game for software. Build speed went up, iteration exploded, and ROI started compounding. Hardware is going the opposite way. Costs are climbing. Timelines dragging. ROI shrinking. If we don’t flip that we’re stuck. You want a hardware boom? Change how we build. Speed isn’t just a perk. It’s the unlock. That requires tech change! More gusto and policies won't hurt but aren't the real strategy.

84

11 Comments