Blake Commagere

Fortnite generated billions in cosmetics revenue because the game worked first. Players stayed because they enjoyed the experience. Monetization followed engagement. Too many publishers get this backward. They design monetization systems and tack the game on top. Players leave. Revenue collapses. The pattern is clear across failed live service launches. Studios that put player engagement first build durable revenue streams. Studios that design around monetization first watch retention curves collapse within weeks. This isn't a soft issue for equity analysts. It shows up in MAU/DAU trends, ARPDAU, and payer conversion rates. The metrics tell you which publishers understand the model and which are chasing short-term extraction. We covered this with Lewis Sterriker in our recent LinkedIn Live on gaming. Full article breaks down what metrics matter, how to spot monetization risk early, and where regulatory pressure is building. Found this useful? Share it with your network. Link to the full article in comments.

7

2 Comments

Exciting to see the incredible progress from Peptone, a company we’ve backed since the early days at Walden Catalyst Ventures. Peptone just announced Peptron-O, developed in collaboration with NVIDIA — a groundbreaking AI engine that brings ensemble-first structure prediction to the disordered proteome. This is a major scientific and technological milestone, targeting one of the most complex and previously inaccessible areas of biology: intrinsically disordered proteins (IDPs). By shifting the paradigm from static models to dynamic ensembles, Peptron-O opens new doors for drug discovery — especially in areas like oncology and neurodegeneration, where IDPs play critical roles. We’re proud to have supported Peptone on their mission to build the infrastructure for next-gen molecular physics. Their work is a great example of how deep tech can transform life sciences. Read more about the announcement: 👉 https://lnkd.in/eygnKQ67 #AI #Biotech #DrugDiscovery #DeepTech #LifeSciences #WaldenCatalyst #Peptone #NVIDIA #PeptronO with: Kamil Tamiola, Ph.D. Francis Ho Shankar Chandran Nicolas Autret Andy Kau Roni Hefetz Victoria Slivkoff

56

1 Comment

Excited to host our next Gradient Descending x Munich roundtable with OpenAI on *parallel agent swarms* April 22nd🔥 We'll be joined by Benedict Kerres to dig into Codex, their coding agent that's quietly changing how dev teams ship software. The default model for AI in code is still pretty sequential: one prompt, one task, one bottleneck. Codex opens something different: multiple agents running in parallel, writing features, fixing bugs, proposing PRs, while you stay focused on the work that actually needs you. Spots are limited: if you're building with or on top of AI dev tools, come join the conversation in our Earlybird Venture Capital office 👇 https://luma.com/vzebgk6q cc Eva Spannagl, Laura Modiano, Laurin Class, Gabriel Schnabel

41

4 Comments

GPT, Grok, Claude and Gemini put to the test with video games. AI Village’s “Claude Plays Whatever it Wants” experiment gave models like GPT-5, Gemini, and Claude a week to complete as many video games as possible. The results? No agent fully “won” a game, but there were sparks of progress: - Claude solved word puzzles and reached 3,000+ in 2048 - Gemini explored nearly 20 games before stalling on execution - GPT-5 showed creative detours, even building tools mid-challenge The outcome? A reminder that while today’s AI agents shine in coding and structured tasks, open-ended environments still expose gaps in reasoning and focus. The encouraging part: each attempt highlights where agents need scaffolding, and points to how fast they’re improving. Do you think AI agents are closer than we realize to mastering “any task”?

46

51 Comments

Pendle Print #99 • sPENDLE is now live - a new liquid staking token with 14-day withdrawal period replacing vePENDLE • Protocol revenue will be used for $PENDLE buyback, and distributed to sPENDLE holders • New pool highlights: cUSD & stcUSD 23JUL2026, mUSD & msY 9APR2026 • Caps raised for PT-sUSDe markets (Ethereum & Plasma) on Aave • $XRP markets are now live on @boros_fi Substack link: https://lnkd.in/gbUGC-hy

36

2 Comments

Sensor Tower acquires Playliner Another data acquisition to bolster their mobile games insights offering. Sensor tower continues to make acquisitions to strengthen their data moats around their core competencies.

7

��� https://lnkd.in/gUfR7u8n When it comes to C++ frameworks, there’s no universal “best.” A real-time trading engine and a cross-platform photo editor need very different foundations. That’s why the smartest teams start with project goals, then evaluate frameworks on: ⚡ Performance overhead 🌍 Cross-platform maturity 🤝 Community & ecosystem 📜 Licensing fit 🎯 Learning curve Framework choice isn’t just technical—it’s strategic. Pick wrong, and you’ll pay for it in deadlines, maintenance, and morale. 👉 Which C++ framework has been the best fit for your projects—and why? #cpp #programming #softwareengineering #devcommunity

1

The next AI wave is not about who builds the best workflow. It is about who controls the channel into the buyer. We've reached the pivotal point to where AI hype has finally reached deal flow saturation. Similar to the 2020/21 Web3/DeFi boom, venture is going to start looking into allocating funds to AI startups that have immediate distribution flows/channels to sell their products to. What is a distribution channel? A distribution channel is the repeatable path a startup uses to consistently reach, acquire, and retain its target customers at an efficient cost. This comes in the form of: strategic partnerships, proprietary sales pipeline/funnels, brand following/audiences (commonly in D2C), etc. What this means: HOW you apply AI will be extremely scrutinized, building wrappers, simple AI agent workflow automations, etc. will not be enough to move the needle. The only way you will succeed is if you can demonstrate a proprietary channel pipeline with enough scale to jusitify your AI product An ex. startup: AI reimbursement recovery copilot for rural critical-access hospitals (CAHs) and their cost-report driven Medicare revenue Distribution channel (highly proprietary): Partnership with a niche CAH cost-reporting + reimbursement consultancy (the firm that already prepares and defends the hospital’s Medicare cost report) Firm footprint: 180 CAHs under contract Product placement: bundled as an “ongoing recovery + audit defense” add-on inside the firm’s retainer How the channel works (repeatable flow): Trigger event: every cost report submission + audit response cycle creates a guaranteed workflow entry Warm intro: consultant is already trusted and already inside the finance ops stack Embedded data access: you get standardized reports, denial logs, and audit correspondence as part of the engagement Quick numbers (VC-friendly): Reach day 1: 180 hospitals Conversion assumption: 15% = 27 hospitals Pricing: $4,000/month platform fee + 10% of recovered reimbursements Conservative recovery: $250k/year per hospital ARR math (base fee only): 27 × $4,000 × 12 = $1.296M ARR Variable upside: 27 × $250k × 10% = $675k/year Total year-1 run-rate: ~$2.0M Why this is truly niche and defensible: 1. CAH reimbursement is a specialized sub-market with unique rules, timelines, and incentives 2. The channel is a small set of gatekeepers (cost-report firms), not broad “healthcare AI” marketing 3. Distribution is anchored to a mandatory annual process with recurring audit events, not discretionary spend 4. Competitors cannot easily wedge in without replacing an entrenched advisor relationship Curious to see how this will play out moving forward, it seems founders that save time building product will have to spend their excess time building these channels/GTM playbooks.

4

1 Comment

All I want for christmas is a ~60B5A MoE model in fp4 with vision encoder + MTP, 262k token context in 5GB and a CPU/GPU/NPU with 64GB ram at 200GB/s bandwidth to drive a comfortable 50 tokens per second local inference with good agentic and coding quality on a 20 watt power budget. It is quite simple how local inference is going to have its breakthrough in less than a year from now.

4

Windsurf’s Claude Cutoff Sparks AI Platform Drama Lon and Alex discuss why Anthropic abruptly pulled access to Claude for Windsurf, just as the startup awaits a rumored $3B OpenAI acquisition. Is this bold move by Anthropic just business, or a warning sign for devs betting on closed models? Key Points: • Windsurf (fka Codem) got less than 1 week's notice before being cut off from Claude • Raises platform risk concerns for startups using closed foundation models • Ironically, the move could accelerate the OpenAI acquisition #AI #Anthropic #OpenAI #Windsurf #Claude #Startups #PlatformRisk #VC #ThisWeekInStartups

16

1 Comment

AppLovin Q1 2026 earnings: 59% ad revenue growth, the hybrid monetization opportunity for games "In prepared remarks, AppLovin’s CEO highlighted a trend of mobile games adopting a hybrid monetization model (ads and in-app purchases), given the opportunity to monetize through advertising via AppLovin without promoting competitors." https://lnkd.in/d8af4-W5

76

2 Comments

South Korea’s SQUARES raised $1.6M in Series A funding to enhance its AI-powered no-code web builder Q Shop and accelerate global expansion and personalized feature development. https://lnkd.in/gcvjMvuW

2

Excited to announce Emergence Capital lead investment in Vesence's $9M seed round!   Legal AI should be a slam dunk. It’s a text-rich domain with tons of training data; well-suited for LLMs. Thus, it's no surprise that AI demand amongst lawyers has been extraordinary.   Unfortunately, in many scenarios, the reality hasn't yet delivered on the promise. Many AI tools fail for the same reason: they prioritize generation over precision.   Hallucinations and inconsistencies turn time-saving innovation into a liability and lawyers have to re-review every clause and citation, which adds hours of work instead of saving them.   Before AI, firms solved this with armies of reviewers scrolling through Word docs line by line.   Vesence is tailor made to bring precision to law at scale.   Instead of generating new text that might be wrong, Vesence uses AI to ensure what’s already written is right.   It enforces firm-wide standards, validates references, flags inconsistencies, and polishes language to perfection — all inside Word and Outlook.   Even more impressive than the product are the founders. Henrik and Ludvig embody the precision and persistence we look for at Emergence.   Henrik built the first version solo, including full Microsoft integrations. And Ludvig, a former pro athlete, brings the discipline of an elite competitor. Indeed, they literally slept at a top Swedish law firm to study the workflows firsthand as they built out the product. Intensity driving precision.   We're thrilled to welcome Henrik and Ludvig to the Emergence family.

328

16 Comments

🚀 Why ERC-8169 is a Game-Changer for Real-World Assets (RWA) ERC-721 NFTs are great for digital collectibles—but they lose ownership history on transfer. For real-world assets like real estate, art, or IP, provenance and auditability are non-negotiable. Enter ERC-8169 / ERC-721H — the three-layer NFT model designed for permanent, verifiable ownership history: 1️⃣ Immutable Origin – Who minted the asset first 2️⃣ Historical Trail – Every owner, chronologically 3️⃣ Current Authority – Who owns it now ✅ Why it matters: • Full on-chain proof of ownership history • Regulatory & legal compliance ready • Founder/early-adopter perks survive transfers • Historical auditability even after burn • Supports governance tied to original ownership 👍 Reference compliance framework: We’ve added modular, ERC-3643-inspired compliance modules: KYC, country restrictions, holder caps (Reg D), and lock-ups (Reg S/D). Pluggable, upgradable, and fully hook-composable — no redeploy required. Use cases: Tokenized real estate, limited-edition art, IP rights, legal/estate documentation. ERC-8169 bridges the gap between NFTs and real-world compliance — truly a next-level NFT standard. https://lnkd.in/eUX7xBrJ

4

Jensen on how SEGA saved NVIDIA by converting the final $5M payment from a failed contract to develop a graphics chip for the Sega Dreamcast console in 1996. Sega 3x'd that investment when Nvidia IPO'd three years later (in the often annoying 'hindsight is 20/20' category, that investment would be worth about $1 trillion today. Sega made the right call at the time though to sell). Great story, and shows the importance of being honest and humble in business.

411

41 Comments

NeurIPS is this week, and on my way to San Diego I’ve been replaying a bunch of last year’s talks in my head. The loudest takeaway then was that speech recognition, transcription, and voice AI were basically “solved.” I didn’t buy it then, and I buy it even less now. If voice AI is solved, why couldn’t I dictate a text or talk to my car? The most clarifying feedback I’ve heard didn’t come from a benchmark. It came from a roofing franchise manager sitting next to me on a flight. He told me he wouldn’t let voice agents handle scheduling: “It’s not good enough. My customers know it’s an AI, and it feels disrespectful.” That’s the real bar. Not whether a model tops a leader board, but whether it earns trust in a human moment. Yes, speech models are improving. Word error rates keep dropping, synthesis is more natural, and latency is trending down. But “improving” is not “solved.” Real-world voice still needs: * robustness in messy environments * low end to end latency that feels conversational * prosody and timing that match intent * reliable handling of overlap, disfluencies, interruptions, and emotion Those are not just engineering cleanups. They’re open research problems. Most voice agents today are “cascades”: ASR (audio to text) → LLM → TTS (text back to audio). It works, but it comes with structural limits: Latency compounds across stages. Even small delays feel awkward in conversation.Errors propagate. A missed word or bad punctuation upstream gets amplified downstream. The speech signal gets flattened into text, so we throw away prosody, emphasis, timing, pitch, and the cues that tell you if someone is frustrated, joking, or on the verge of hanging up. That last one is why so many voice agents feel robotic. A frantic rant and a calm complaint can map to similar text, but they absolutely should not get the same response. Gradium is tackling this with a two-pronged approach: First, they’re closing the gap on cascades by pushing accuracy, cost, and latency to a place where real-time deployment finally makes sense. That matters because cost and responsiveness are still the blockers for adoption. Second, they’re training end-to-end speech models that operate directly on audio and preserve the full signal instead of funneling everything through text. That’s the architectural unlock. If a caller sounds irritated, the model can condition on that emotional texture and respond in a tone that’s actually sympathetic, not just polite words read in a chipper voice. This is hard. It takes fundamental research, serious engineering, and careful data work. But until models can handle the “umms,” overlap, interruptions, and emotional variance that define human speech, voice will stay a novelty instead of becoming a default interface. We’re excited to back Gradium at seed because we think they’re building toward that future.

46

1 Comment