4273207
A hub for startup news, trends, and insights, covering the global startup ecosystem for founders, investors, and innovators. Community: @startupdis Buy Ads: @strategy (this is our only account).
📣 Sam Altman’s 2 rules for surviving as a founder
In a recent reflection, OpenAI CEO Sam Altman shared two lessons he wishes he’d learned earlier in his career — both are brutally simple, and extremely useful for any founder.
🔸 Ask for what you want — too many founders self-censor. They don’t pitch bold hires, ask for partnerships, or go after the deal. Fear of rejection kills opportunity before it’s born.
🔸 Crisis gets easier with time — the first few feel like the end of the world. By the 19th, you’ve built survival muscle. Every disaster teaches you that it’s possible to recover — and eventually, you stop panicking.
💡 The takeaway: confidence doesn’t come from being fearless. It comes from failing and surviving — again and again.
💊 Ex-Twitter CEO jumps into weight loss tech
Linda Yaccarino, former CEO of X (Twitter), is now leading health-tech startup eMed Population Health — a company building an AI-powered platform for patients using GLP-1 drugs like Ozempic.
🔸 eMed previously ran a COVID home-testing platform with step-by-step remote guidance
🔸 The company is now pivoting to the booming GLP-1 space: weight loss and Type 2 diabetes treatment
🔸 Yaccarino has no healthcare experience, but brings big-league media and partnership skills
🔸 Her hire signals eMed’s ambition to grab both market and investor attention
In the $100B+ GLP-1 race, eMed just added a headline-making operator to its arsenal.
🔍 Perplexity caught bypassing website indexing rules
Cloudflare has flagged Perplexity for evading site-level restrictions by using disguised crawlers. When blocked via robots.txt, it reportedly switches to a browser-masked bot to continue scraping content.
🔸 Perplexity now excluded from Cloudflare’s trusted bots list
🔸 Website owners can block it directly via Cloudflare settings
🔸 Cloudflare shared technical details of Perplexity’s cloaked crawler
The move raises fresh questions about AI data ethics — and how long these blocks will even work.
🚀 ByteDance unveils Seed Diffusion — a new diffusion-based language model
Instead of generating text token-by-token, Seed Diffusion creates the entire output at once — similar to how image models like Midjourney work.
🔸 Outperforms Google and Inception Labs in most benchmarks
🔸 Key edge: speed — over 2,000 tokens/sec, 5.4× faster than standard models
🔸 Open for free testing: Seed Studio
A major leap in generation architecture — and a serious play from ByteDance in the AI race.
💳 The hidden history behind modern card payments
Before Apple Pay, Stripe, or embedded checkout, there were paper vouchers, magnetic stripes, and punched cards. This short timeline shows how payments evolved — from forgotten wallets to biometric tokens.
🗓 From 1914 to 2024:
🔸 1914 — Western Union issues the first metal “buy now, pay later” cards
🔸 1949 — A forgotten wallet inspires the birth of Diners Club
🔸 1958 — AmEx launches a paper card, replaced by plastic in 1959
🔸 1967 — Barclays rolls out the first ATM in England
🔸 1974 — The smart card is patented in France
🔸 1979 — Visa introduces the magnetic stripe reader
🔸 1983–84 — Holograms become standard to prevent fraud
🔸 1990s — Chip & PIN and global EMV specs transform card security
🔸 2007 — Contactless payments go live
🔸 2023 — Mastercard announces a shift to biometrics and tokenization by 2030
From punched cards to payment tokens, this is what 110 years of financial UX looks like.
📊 Excel killer? Meet NanoCell — a blazing-fast, privacy-first spreadsheet tool
A new tool just dropped that might be the cleanest Excel alternative we’ve seen — NanoCell, built by a veteran analyst for speed, simplicity, and control.
🔸 Handles huge datasets, financial models, formulas, and visualizations — all lightning-fast
🔸No complex macros or formulas needed
🔸 Built with only the essentials — no bloat, no noise
🔸 Keeps all data intact — no auto-loss like some competitors
🔸 Runs fully on a static server — safe for sensitive docs, no data leaks
👉 Try it here 👈
Open-source on GitHub
One-click tables, no Excel pain.
🧠 OpenAI drops its first open AI model since GPT‑2
OpenAI has released GPT‑OSS, its first open-weight model family in over six years. It includes a 120B and 20B version — both available under Apache 2.0, optimized for local use, and surprisingly powerful.
🔸 GPT‑OSS 120B matches o4-mini in reasoning benchmarks
🔸 GPT‑OSS 20B runs on a laptop and rivals o3-mini
🔸 Supports chain-of-thought reasoning and tool use
🔸 Full commercial use allowed under a permissive license
🔸 Designed for transparency, privacy, and autonomy
🔸 First OpenAI release meant to compete with LLaMA, DeepSeek, Qwen
👉 Try it here 👈
This is OpenAI’s biggest step back toward open AI — and it’s fast, free, and local.
🧪 PlayerZero is building an immune system for your code
Most tools try to speed up developers. But bottlenecks often sit elsewhere — like in QA. That’s where PlayerZero steps in: AI agents that detect and fix bugs before production, trained on your repo’s history.
🔸 Learns from past incidents, patches, and test cases
🔸 Suggests auto-fixes directly in pull requests
🔸 Designed for complex monorepos and legacy systems
🔸 Hooks into CI/CD in minutes
🔸 Used by teams like Zuora to monitor billing infra
🔸 Just raised $15M from the likes of Databricks' founder and Vercel’s CEO
Fun fact: the founder demoed it to the creator of Next.js — who didn’t believe it at first, but then invested.
🎬 The ultimate AI toolkit for video creators — in one link
We found a goldmine for content creators: a browser-based platform with every top AI video tool in one place.
🔸 Text-to-video, subtitles, noise removal, face cloning — it’s all here
🔸 Supports trending models like Veo 3, Minimax, Kling, and more
👉 Save your dream video editor 👈
Perfect for creators who want speed, quality — and less post-production pain.
💰Why an AI engineer turned down $1.5B from Zuckerberg
In his bid to build the ultimate AI dream team, Mark Zuckerberg reportedly offered ex-Meta engineer Andrew Tulloch a staggering $1.5B deal to return — a 4-year package of salary, equity, and performance bonuses.
But Tulloch said no❕
🔸 He currently owns 3.75% of his AI startup, Thinking Machines
🔸 Based on current market buzz, that stake could be worth $3B in the next round
🔸 Turning down $1.5B wasn’t irrational — it was mathematical
🔸 Now it’s a race: will Thinking Machines hit IPO before the AI bubble deflates?
Zuck offered a fortune. Tulloch bet on more.
🌐 Build landing pages in seconds — meet Pagy
Pagy just launched: a blazing-fast browser-based tool for creating websites and personal pages without writing a single line of code.
🔸 No installs — works entirely in your browser
🔸 Pick a template, edit text/images/links right on the page
🔸 No designer, no developer, no code needed
🔸 Built-in hosting and analytics
🔸 100s of ready-made designs from devs and the community
👉 Try it here 👈
A perfect tool for fast MVPs, link-in-bio sites, or spinning up ideas on the fly.
📊 Powered by Crypto Insider
🖌 Photoshop adds new AI tool for seamless image blending
Adobe has launched a new feature in Photoshop called Harmonize — it automatically matches lighting, shadows, and colors to blend one image into another in seconds.
Perfect for creators, meme lords, and anyone tired of manual tweaking.
Drop it in, harmonize, done.
🧊 Frozen for 30 Years: The Oldest Embryo Baby Born in the US
A boy named Thaddeus Daniel Peirce was born in Ohio from an embryo frozen 30.5 years ago — the longest-known frozen embryo to result in a live birth.
🔸 The embryo was created in 1994 and stored by Linda Archerd, who gave birth to one child that year and kept the remaining embryos frozen for decades, paying $1,000 annually.
🔸 With age and menopause, Archerd decided to donate the embryos, but only to a white, Christian, married US couple.
🔸 The Snowflakes program from Nightlight Christian Adoptions matched the embryos to Lindsay and Tim Peirce, a couple who had tried to conceive for 7 years.
🔸 Rejoice Fertility Clinic in Tennessee performed the transfer — one embryo didn’t survive thawing, but two implanted, and one led to birth.
🔸 Most clinics refuse to work with outdated freezing methods. Rejoice’s founder believes every embryo “deserves a chance at life.”
A surreal mix of faith, biotech — and long-term moral investment.
📣 Why Sam Altman Says You Shouldn’t Track User Growth Early On
In the early stages of a startup, it’s not about how many users you have — it’s about how much they love your product.
Altman explains why obsession with absolute growth is a mistake, and what you should focus on instead:
🔸 A small group of obsessed users is better than a wide pool of casual ones
🔸 Deep engagement signals product-market fit more than raw numbers
🔸 Retention and frequency are your best early metrics
🔸 Word-of-mouth is the ultimate validation
Almost all great companies start with a product that a few people love.
🤖 Neural Agent — your AI-powered desktop assistant
A powerful open-source AI agent that can take full control of your computer: search files, browse the web, fill forms, send emails, and more.
🔸 Automates all routine tasks while you code, design, or think
🔸 Runs on Claude, GPT-4, Azure OpenAI, and Bedrock — top-tier models
🔸 Fully integrated UI — type commands into a small bar and get results instantly
🔸 Simple setup with an easy-to-use API and step-by-step install guide
👉 check it out 👈
📊 Powered by Crypto Insider
🐺 China’s new battlefield beasts: robo-wolves, not robot dogs
China has officially unveiled a new class of robotic fighters — not cute quadrupeds, but 70-kg robo-wolves armed with rifles and grenade launchers.
🔸 Deployed in recent military drills, these robo-wolves fought alongside human troops on the front line
🔸 Controlled via FPV drone data, some were equipped for recon, others for direct combat
🔸 Commanders say the packs complete tasks efficiently and intimidate enemy soldiers
Robotic warfare is shifting fast — and psychological edge might matter just as much as firepower.
🚀 Top 10 accelerators for early-stage startups
If you're building a startup and looking for the best launchpad, here are the most respected accelerators globally — ranked by quality of alumni, access to investors, and overall reputation:
🔸 Y Combinator – The gold standard. $500K early checks, global brand. Alumni: Airbnb, Stripe, Dropbox.
🔸 Techstars – Huge network in the US and Europe. 4,100+ portfolio companies, $30B+ raised.
🔸 500 Global – Focused on emerging markets (LATAM, Asia). 2,600+ startups across 5 continents.
🔸 StartX – For Stanford founders. No equity required. Deep Valley ties and elite network.
🔸 Antler – Global program that starts at idea stage. Strong founder matching across 30+ locations.
🔸 StartupBootcamp – Fast launch tracks with corporate partners in Europe, Asia, Africa.
🔸 Alchemist Accelerator – B2B-only, deep expertise in enterprise and infrastructure.
🔸 Plug and Play Tech Center – Corporate partnerships, global reach, rich industry tracks.
🔸 Berkeley SkyDeck – UC Berkeley-backed, strong in deeptech and university-linked startups.
🔸 NFX Fast Track – Pre-seed/seed program with elite mentorship and fast-track access to NFX Signal.
Before applying, compare terms, graduate traction, and geo fit. The right accelerator can change your whole trajectory.
✍️ 4 lessons from Microsoft Copilot's AI mess
Despite leading in enterprise software, Microsoft’s Copilot rollout shows what not to do in AI product design. Broken context, poor UX, and generic tooling are symptoms of deeper design failures — and they’re avoidable.
Here are four hard-earned lessons:
🔸 Meet users where they are
Copilot assumes one-size-fits-all. Power users get frustrated, novices get lost. Its Teams integration works because it automates boring tasks within familiar workflows. But Word, Excel, and Outlook? Slow, confusing, inconsistent.
🔸 Context is everything
Copilot fails to account for app versions, user settings, or data environments. Great AI tools like Cursor thrive because they start narrow, respect user context, and expand carefully. Copilot ignores that, and it shows.
🔸 Design for failure
AI will mess up — the question is how gracefully. Copilot offers no useful feedback loops. Unlike Replit, which previews agent output early, Copilot delivers errors late and opaque. No trust, no iteration.
🔸 Avoid overreach
Microsoft tried to bolt “AI” onto everything without clarity or purpose. Copilot often tells users how to do things, rather than just doing them. A good AI product solves one real task completely — not 10 vaguely.
The irony? Microsoft owns GitHub Copilot — a well-scoped product loved by devs. But its enterprise AI tools feel more like a press release than a real solution.
🧠 GPT-5 is here — and it’s OpenAI’s biggest leap yet
OpenAI just released GPT-5, its most advanced AI model to date, and for the first time, it's available to all ChatGPT users — including those on the free plan.
The model is faster, more accurate, and more versatile across writing, coding, and even healthcare.
🔸 Outputs over 2,000 tokens/second — 5x faster than before
🔸 Hallucinations reduced, safer replies now use "safe completions"
🔸 Handles logic, reasoning, and long documents far better than GPT-4
🔸 Free users get GPT-5 Mini after hitting usage caps
🔸 GPT-5 now integrated across Microsoft 365 Copilot and Azure
OpenAI also demoed “vibe coding” — generating full apps from prompts in seconds — showing just how far generative UX has come.
👉 Try GPT-5 now 👈
Altman calls it like having a team of PhDs in your pocket. Whether that’s hype or reality — GPT-5 is a major step toward AI you actually want to use.
⚙️ How to build a voice-enabled second brain in 3 hours
Capture ideas in seconds and turn them into content, strategy, or action — without dashboards or overengineering.
Here’s how the setup works:
🔸 Use Voicenotes.com to record and transcribe voice memos
🔸 Connect Voicenotes to Make.com via Webhook
🔸 A Gemini 2.5 model classifies each note: content, business idea, observation, or reading
🔸 Notes are routed into Notion databases automatically
🔸 Content ideas trigger AI-generated LinkedIn drafts, sent to Typefully
🔸 Business ideas get enriched with SWOT analysis and action plans
Built with just four tools: Voicenotes, Make, Notion, and optional AI APIs. Total setup time: ~3 hours.
🖌 New image model drop: Qwen Image goes public (and free)
China just dropped a beast — Qwen Image, a free, open-source image generator that’s already turning heads.
🔸 Handles any style: realistic, stylized, cartoons, posters, comics
🔸 Understands prompts like ChatGPT — with strong visual-text alignment
🔸 Can edit images with precision (think Flux Kontext quality)
🔸 Preserves original style — no plastic mess, no hallucinated chaos
🔸 Fully open-source and no usage limits
👉 Test the model here 👈
This might be the strongest free image model yet.
🔍 Motive lands $150M to expand its AI fleet platform to the UK
Motive, the AI-powered platform transforming physical operations for fleets and equipment, has raised $150M in new funding led by Kleiner Perkins. The company is now expanding into the UK to help solve challenges in logistics, safety, and driver shortages.
🔸 Motive serves 100,000 customers and 1.3M drivers, with nearly $500M ARR
🔸 AI detects risky driving with 4x better accuracy than competitors
🔸 Clients report 80% fewer accidents and up to 2,000% ROI
🔸 Platform covers safety, compliance, maintenance, and fraud prevention in one dashboard
🔸 The funding will accelerate global rollout, R&D in India, and UK go-to-market efforts
Motive’s edge lies in precision at scale — using AI trained on billions of miles to drive real-world impact across the global physical economy.
🔍 ChatGPT agent controls live cam, finds boat on request
A ChatGPT-powered agent was given access to a public webcam overlooking a marina — and asked to locate a specific boat.
It moved the camera, zoomed in, scanned the docks… and found it.
The line between digital agents and physical tasks is getting thinner by the day.
📣 Bezos on the power shift from marketing to product
In a 2012 interview, Jeff Bezos predicted a major change in how companies should think:
Word-of-mouth is more powerful than it has ever been before.
If I build a great product or service, my customers will tell each other.
🫂Your own AI team, no code needed
Eigent is a new app that lets you build a squad of AI agents to handle work tasks for you — like NPCs, but productive. Think of it as a plug-and-play stack of paid neural networks, working for you in the background.
🔸 Set up custom AI workflows without writing code
🔸 Agents take on tasks like cleaning files, reading PDFs, and crunching data
🔸 Handles everything from accounting to market research
🔸 Use cases include: sorting downloads, processing lab results, parsing CSVs for finance
🔸 Feels like Zapier meets AutoGPT — but actually usable
This isn’t just another AI toy — it’s a quiet power-up for anyone drowning in digital busywork.
🧠 Qwen rolls out new reasoning-focused open model
Qwen just released Qwen3-30B-A3B-Thinking-2507 — a new medium-sized open model optimized for logical reasoning across math, science, and code.
🔸 Strong performance on logic-heavy tasks
🔸 Comparable output quality to top competitors
🔸 Supports 256K tokens of native context — expandable to 1M
🔸 Available now via chat.qwen.ai (select the “Thinking” mode)
Qwen continues to quietly raise the bar for open-source LLMs — with serious context and reasoning firepower.
🧠 Y Combinator shares fresh ideas for AI-first startups
Ahead of its fall batch, Y Combinator revealed new directions it’s excited to fund — all rooted in the belief that modern startups should treat AI not as a feature, but as the foundation.
Here’s what YC wants to see 👇
🔸 AI-powered training for skilled trades
Think plumbers and electricians trained via VR, with AI tutors giving live feedback and adapting lessons in real time.
🔸 Generative video as a building block
Not final product, but part of the stack — from no-code game creation to virtual clothing try-ons and AI “calls” with digital versions of deceased relatives.
🔸 Infrastructure for multi-agent systems
Startups that can help developers manage huge webs of AI agents — prompt routing, context reliability, debugging at scale.
🔸 AI-native enterprise software
A new Salesforce or ServiceNow, but built for the AI era — tools that actually assist teams, not just log data.
🔸 Ultra-lean, AI-leveraged startups
YC believes a 10-person team can now build a $100B company — if it uses AI to move fast and stay small.
🔸 AI replacing legacy govtech
$100B/year goes to old software in the US government. YC wants startups like Caucus or Archon that cut this cost with modern, AI-native tools.
Applications for the fall batch close August 4.
And yes, they still offer $500K for 7% — plus another slice after your next raise.
The message is clear: build with AI at the core — or get left behind.
📈 Figma stock pops on IPO debut, hits $47B market cap
Figma finally went public—and Wall Street couldn’t get enough. The design platform opened at $33, surged to $124, and closed at $115.50, ending the day with a $47B market cap.
🔸 IPO demand was so high, trading was briefly halted due to volatility
🔸 Retail buyers joked online about being allocated just 1 share
🔸 Market cap doubled Adobe’s failed $20B acquisition offer from 2023
🔸 After-hours trading stayed hot, with continued price action
Figma didn’t just rebound from a blocked deal—it sprinted past it.
🚀 Australia's first orbital rocket lasted just 14 seconds in the air
Eris, the 30-ton rocket from Gilmour Space, launched from Bowen Spaceport — and promptly crashed. But the company is calling it a win.
🔸 First Australian-made rocket to launch from local soil
🔸 Flew for 14 seconds, with 23 seconds of engine burn
🔸 Crash caused no injuries or environmental harm
🔸 Valuable flight data collected for next test
🔸 Prior attempts were delayed by power surges and high winds
🔸 CEO: "Getting off the pad is a huge step forward"
💡Even failure is a milestone when you're building a space industry from scratch. Australia's now officially in the orbital launch game — and they’re just getting started.
🫂How small teams build big revenue
A new report from Growth Unhinged breaks down how early-stage SaaS startups structure their teams.
The takeaway?
You don’t need a huge headcount to hit $1M ARR.
Startups under $1M typically run lean — with just 7 people.
That’s your classic "two-pizza team."
Here’s what that usually looks like:
🔸 3 engineers
🔸 1 product/design
🔸 1 sales
🔸 1 support
🔸 1 admin/ops
Some skip marketing hires early on — founders often handle GTM themselves. Others outsource one-off tasks instead of hiring full-time.
The key is not just who you hire, but how clearly each role maps to revenue. If a founder understands that every teammate should drive profit, not just burn cash — investors will notice.
📊 Powered by Crypto Insider