Startups & Ventures

🧠 What it’s like to work at OpenAI — insights from a Codex engineer Calvin French-Owen, ex-startup founder and one of the engineers behind Codex, spent a year at OpenAI before leaving. He recently published a long reflection — here are the standout takeaways: 🔸 No one uses email internally — all comms go through Slack 🔸 Headcount grew from 1,000 to 3,000 in a year — processes constantly break 🔸 No strict product roadmap — priorities shift fast, sometimes no quarterly plan at all 🔸 Lots of bottom-up initiatives — small, semi-independent teams start research on their own 🔸 Flexible team structure — people can be moved between projects instantly if needed 🔸 High secrecy — Slack channels with access levels, can’t tell others what you work on 🔸 Many ex-Meta folks joined — not necessarily poached, just moved 🔸 Twitter is monitored constantly — post an idea and it might get picked up 🔸 No swag stockpile — merch drops are rare; first one crashed the Shopify store 🔸 Codex sprint was brutal — slept 5–6 hours, compared it to Y Combinator intensity 🔸 From first line of code to Codex launch: 7 weeks — says he’s never seen anything like it

OpenAI moves fast, breaks things, and keeps secrets. Not for everyone — but a dream for builders who like chaos.

🔘 Ultimate browser extension for prompt hoarders — SuperPrompt SuperPrompt is the perfect tool for anyone who works with AI prompts regularly. It lets you save, organize, and instantly access all your favorite prompts — right from a sidebar.

No more digging through notes or docs. Just drop your prompt into the database and it’s there when you need it.

🛠 Try it here

❔ A riddle for startup founders Two companies. Same industry, similar product. Very different paths. Company A: Solo founder 🔸$10K MRR 🔸$500 on servers 🔸Max $2K on marketing 🔸~$7.5K monthly profit Company B: 🗄2 founders + 1 investor 🗄$100K MRR 🗄$80K in monthly costs (infra + marketing + salaries) 🗄~$20K monthly profit Now the question:

Who’s doing better? Or more importantly — which setup would you choose?

Explain your thinking in the comments 👇 (Hint: it’s not as obvious as it looks.)

What are people asking ChatGPT in 2025?

In April 2024, 44% of queries were about software development.

Now, people are asking more about themselves.

📉 Small businesses risk falling behind in the AI revolution New research shows that SMEs are falling far behind in AI adoption — and it could cost them productivity, talent, and growth. 🔸54% of small businesses use AI less than once a month 🔸38% say they’ve never used it 🔸 Only 24% feel their company invests meaningfully in AI 🔸Millennials are adopting 2–3x faster than Gen X or Boomers 🔸 Those who do use AI report better output, less stress, and even shorter workweeks

Big companies are already building automation pipelines. SMEs that don’t catch up soon could lose their edge entirely.

🌐 Japan breaks 1 petabit/second fiber speed record Japanese engineers just pushed the limits of data transmission: over 1 petabit per second across 1,800 km of fiber — using cables compatible with today’s internet infrastructure. 🔸That’s 10¹⁵ bits/sec — a quadrillion 🔸Enough to stream the entire Netflix library in one second (allegedly) 🔸Achieved using a custom multi-core optical fiber 🔸Compatible with existing transmission systems 🔸Reinforces that internet infra can handle surging global demand

You learned "kilo" from kilometers. Now meet "peta" — the new speed benchmark for the digital era.

⚡️ Google killer? Comet browser by Perplexity and NVIDIA goes live Google’s in trouble: NVIDIA and Perplexity launched Comet, the first AI-native browser — and early testers say it’s a game-changer. 🔺Comet Assistant works on any site: summaries, search help, problem-solving 🔺Perplexity is baked in — smarter, faster, and cleaner than Google 🔺Built-in AI agents book hotels, write emails, pay bills 🔺Manages tabs, bookmarks, and removes clutter by itself 🔺Comes preloaded with ad blockers and pop-up killers Access is now open

📞China is developing cyborg bees for surveillance Chinese researchers are working on remote-controlled bees — real insects fitted with microchips that allow human operators to direct their movement. The chip weighs just 74 milligrams and is light enough for a bee to carry in flight.

Use cases? State surveillance. Authorities plan to deploy the cyborg bees for monitoring, eavesdropping, and intelligence gathering.

Nature meets control tech.

🔥 Base44: The profitable AI startup built solo and sold for $80M Base44 is an AI-powered app builder that turns plain-text prompts into fully functional web apps. It was created and scaled by a solo founder — and recently acquired by Wix for $80M. 🔍What makes it stand out 🖱️Built and scaled by one person — no devs, no team 🖱️Reached 400K users without VC funding 🖱️Routes prompts across Claude, ChatGPT, and Gemini to maximize output 🖱️Daily product updates, viral loop built-in via social sharing 🖱️Grew via LinkedIn and direct feedback loops with early users 🖱️Generated profit early and kept full control before exit 📈 The exit and after Wix paid $80M, with $25M set aside to retain the team post-acquisition. Base44 is now part of Wix Studio, continuing to operate independently.

Maor Shlomo may not have built the first solo unicorn — but he might’ve shown us exactly how it starts.

📊 25 AI startups on the rise Not every AI breakthrough comes from Big Tech. This Forbes-backed list highlights startups quietly building the infrastructure, creativity, and productivity layers of the AI economy. 💰The biggest funding magnet? Productivity tools — with $700M+ across just 8 companies 🧠 From LangChain to Perplexity, Harvey to Cleanlab — each is solving a different bottleneck in the AI stack ✍️ Midjourney made the list without raising a dollar 🔒 But not everyone will survive — policy, trust, and data security are already major battlefields

The AI race isn’t just about model size — it’s about who trains faster, ships smarter, and earns trust first.

⚡️ Meta takes 49% stake in Scale AI — the dataset factory behind Big Tech models Founded in 2016, Scale AI builds the infrastructure that trains the trainers — cleaning, labeling, and structuring data for top AI systems. It’s now 49% owned by Meta, which paid $14.3B and brought CEO Alexandr Wang into its elite AI group. 🖱️240,000+ contractors handle massive multi-modal data 🖱️Text, audio, 3D, and synthetic data for RLHF at scale 🖱️Quality via majority vote, dynamic calibration, and more 🖱️Upload raw — get back validated datasets via full-cycle API 🖱️Rapid-feedback pipelines using private foundation models But trust issues surfaced fast: leaked docs revealed prompt sets, tasker emails, and internal client data — including Bard and xAI. Meta’s rivals (OpenAI, Google) are now freezing projects. The big question: can Meta own the pipeline and still convince competitors to keep sending blood to its lab?

📺 Top Talks from Y Combinator’s AI Startup School In June, Y Combinator hosted a 2-day AI bootcamp for over 2,500 students and PhDs. They brought in top voices from OpenAI, Tesla, Microsoft, and more. Here’s a handpicked list of the most valuable talks: 🔹Andrew Ng, co-founder of Google Brain Topic: Building Faster with AI 🔹 Andrej Karpathy, former Head of AI at Tesla & OpenAI co-founder Topic: Software Is Changing (Again) 🔹 Elon Musk, founder of Tesla, SpaceX, xAI Topic: Digital Superintelligence, Multiplanetary Life, How to Be Useful 🔹 Sam Altman, CEO of OpenAI Topic: The Future of OpenAI, ChatGPT’s Origins, and Building AI Hardware 🔹 Satya Nadella, CEO of Microsoft Topic: Microsoft’s AI Bets, Hyperscaling, Quantum Computing Breakthroughs 🔹 Fei-Fei Li, “Godmother of AI” Topic: Spatial Intelligence is the Next Frontier in AI 🔹 François Chollet, creator of Keras & ARC Prize Topic: How We Get to AGI

Coming soon: sessions from the founders of Perplexity, Anthropic, and Google DeepMind’s John Jumper.

📌 Save this post — it’s packed with signal.

🤖 RealSense goes independent to power the rise of physical AI RealSense, Intel’s former computer vision unit, has officially spun out and raised $50M to scale its vision systems for robotics, biometrics, and AI-driven automation. The company continues shipping its signature depth cameras — already used in 60% of the world’s humanoid and mobile robots. ✅ Launched the new D555 depth camera with PoE and Vision SoC V5 ✅ Partners include Unitree Robotics, ANYbotics, Fit:Match, and Eyesynth ✅ Backed by Intel Capital, MediaTek, and a leading semi PE firm ✅ Active across robotics, healthcare, access control, and automation • Holds 80+ patents and serves 3,000+ customers worldwide

RealSense says it’s building for a future where robots don’t replace humans — they protect and empower them.

🧠 Physical AI is coming fast. RealSense wants to be the lens it sees through.

🧠 How to convince investors the problem is real A common founder line:

The problem is obvious — we’ve lived it ourselves.

Okay. But as an investor, I might not even know this problem exists. Saying “parents hate carrying car seats” sounds like "trust me, bro". I need proof this is a pain worth paying for. Here are the strongest ways to validate a problem early: 1️⃣ Money Show a screenshot:

In one week, we got X payments from Y customers.

Even if it’s small transfers to a personal card — when someone opens their wallet, they’re voting with actions, not words. 2️⃣LOIs (Letters of Intent) A document from a real company saying:

We’re ready to buy, if the product meets these terms.

Not a contract, but serious signal. Shows you’ve talked to actual buyers, not just “the market.” 3️⃣ Paid pilots / test sales Built a prototype — and someone paid for it? Perfect. Even 10% of final price counts. Especially valuable in DeepTech or hardware: early preorders, even scrappy ones, are a green flag. What doesn’t count as real validation: – “Friends said the idea is cool” – “European fleets are interested, but haven’t paid” – “People gave great feedback at our demo” – “Survey says 87% would buy it” Nice signals — but not validation. Validation is when people pay you to solve a real problem. The best way to prove a problem exists? Try selling the solution. If they buy — great, you hit a nerve. If they dodge — that’s useful too. Better to find out before the pitch. What validation worked best for you? What convinced your investors? Share in the comments.

🧠 Grok 4 is here — and it might be the smartest AI ever built The new model from xAI blows past Claude, GPT-4o, and Gemini 2.5 Pro.

It’s not just faster — it thinks better.

🔥Best-in-class at coding, math, and reasoning 🔥10× smarter than Grok 3 at analytical tasks 🔥Reliably predicts sports outcomes just by scanning Polymarket 🔥Runs business ops 6× more efficiently than humans 🔥Surpasses PhD-level in every domain — even invents new physics 🔥Uses tools (browser, code interpreter) more intelligently 🔥In voice: can whisper, sing, modulate tone — and respond faster than GPT 🔥Scores 44.4% on the “Final Human Exam” — AGI is knocking Grok 4 is rolling out now — see it live.

🧠 Groq Ramps Up Global Push in the AI Chip Race Silicon Valley–based Groq is scaling rapidly with its custom inference chips, designed to deliver ultra-fast performance for AI workloads without the power drain of traditional GPUs. ⚙️ Built for blazing-fast inference • Runs on proprietary Language Processing Units (LPUs) • Delivers up to 10× faster and more energy-efficient performance than GPUs • Already powering over 100,000 LPUs for real-time AI applications 🌍 Expanding global reach • Opened its first European data center in Helsinki • Recently signed a $1.5B deal with Saudi Arabia • In talks to raise $300–500M at a $6B valuation 🤖 Defining a new chip category • Targets inference workloads—not training—unlike Nvidia • Backed by Cisco, Samsung, BlackRock, and other major investors Groq is reshaping how AI infrastructure is built—offering a leaner, faster alternative for inference at scale. 📊 Powered by V3V Ventures

🧠 Foundation EGI Unveils AI-Native Platform to Transform Engineering MIT spinout Foundation EGI has secured $23M in Series A funding to transform industrial engineering using an AI-native platform. The round was led by Translink Capital, with backing from RRE Ventures, McRock Capital, and others. ⚙️ AI that engineers with you • Physics-aware LLM turns messy specs into structured workflows • Supports co-design between humans and AI—not just automation • Speeds up prototyping, simulation, and documentation 🏠 Piloting in complex industries • Live with Fortune 500 firms in aerospace and manufacturing • Built for sectors like automotive and advanced appliances • Helps engineers build smarter and faster 🧪 Deep tech from MIT labs • Founded by Mok Oh, Prof. Wojciech Matusik & Michael Foshey • Based on MIT research into LLMs for design and manufacturing • Translink: “This is industrial AI with real-world impact” Foundation EGI is reshaping the future of product design—where AI doesn’t just assist, it thinks alongside engineers. 📊 Powered by V3V Ventures

🧠Knox Accelerates FedRAMP Authorization for Government AI Use San Francisco–based Knox has raised $6.5 million in seed funding to streamline FedRAMP certification and help U.S. government agencies adopt AI-powered cloud services. The round was led by Felicis, with support from Ridgeline and Firsthand Ventures. ⚙️ Cutting red tape for cloud compliance • Reduces FedRAMP approval from years to just 90 days • Uses automation, pre-cleared infrastructure, and digital control libraries • Addresses a backlog of 30,000+ SaaS providers waiting for authorization 🏠Deployed across federal agencies • Supports DHS, Treasury, and the U.S. Marines • Works with AWS, Azure, and Google Cloud • Offers real-time compliance and security via KnoxAI 💡Built by govtech veterans • Founded by the team behind Adobe’s federal cloud systems • Audited by leading assessor Coalfire • Felicis: “Knox is solving one of the most urgent problems in government tech” Knox is modernizing how U.S. agencies adopt cloud and AI—making compliance faster, smarter, and secure by design. 📊 Powered by V3V Ventures

💼 iCapital Expands Access to Alternative Investments with New Backing New York-based iCapital has raised $820 million at a $7.5 billion valuation. The round was co-led by T. Rowe Price and SurgoCap Partners, with participation from UBS, State Street, Temasek, BNY, and existing investors like BlackRock and Apollo. ⚙️ Simplifying private market investing • Platform offers access to private equity, credit, real estate, and annuities • Serves 114,000+ advisors and 3,000+ firms • Manages over $945 billion in assets 📈 Profitable growth with global reach • Completed 23 acquisitions to date • Profitable since 2019 • Expanding tech and entering new markets 💡Strategic backing for future expansion • Funding supports platform upgrades and product innovation • Aligns with rising demand for alternative investments • Eyes potential IPO in coming years iCapital is making alternatives more accessible—helping investors tap into private markets with institutional-grade tools. 📊 Powered by V3V Ventures

🚀UK Backs Eutelsat in Europe’s Push to Rival Starlink The UK government is investing $191 million to maintain its stake in French satellite firm Eutelsat, as part of a €1.35 billion funding round led by France. The move deepens UK–France cooperation in space amid growing geopolitical focus on satellite networks. ⚙️ Strengthening satellite independence • UK keeps golden-share rights in OneWeb • Complements France’s €717M investment, now owning ~30% • Supports Europe’s goal of securing space-based infrastructure ⛓A strategic rival to Starlink • Eutelsat operates 600+ satellites including OneWeb’s LEO fleet • Aims to counter Starlink’s 6,000-satellite scale with secure, sovereign networks • Targets enterprise and government customers across Europe 💡Fueling next-gen satellite ambitions • Backs the EU’s IRIS² project—290 satellites by 2030 • Tied to €1B French defense contract • Eutelsat shares rose 8–10% post-announcement Eutelsat is becoming Europe’s answer to Starlink—built on sovereignty, security, and strategic investment. 📊 Powered by V3V Ventures