2287
Reading reports. Guessing the future
LinkedIn has built its Top Startups ranking for the tenth time: 50 US companies by its own data (headcount growth, jobseeker interest, ability to pull people from LinkedIn's top companies). No. 1 is Anthropic, then Mercor and Perplexity.
- of 315 startups from earlier lists, 163 (52%) became unicorns and 85 (27%) went public or were acquired
- only 16 shut down, which is 5%
- $16.5B went into space startups this year, according to LinkedIn
The ranking measures the inflow of people, not money or revenue.
https://news.linkedin.com/2026/top-startups-2026
📎 Read also:
→ Hurun Global Unicorn Index - value up 43%
→ Founder Collective - a $1B exit is 8x harder than Harvard
→ StartupBlink 2026 map - Beijing down 21.4%
Articos MCP puts synthetic buyers inside your agent: paste your site, it builds the people who buy from you, and you ask them questions.
- 12 personas per study, five of them set up as skeptics and late adopters, so not everyone loves your idea
- a report in about 30 minutes: motivations, objections, prioritized recommendations, PDF export
- from $8 per study on the monthly plan, about $10 for a one-off pack; traditional research starts at $5,000 and takes 4-8 weeks
The 86% accuracy (46 studies, checked against Baymard and Nielsen Norman publications) is the company's own estimate, with no independent verification.
https://articos.com/
📎 Read also:
→ IdeaProof - database of documented company failures
→ Databox MCP - ask business metrics in Claude
→ Clueso MCP - product videos from agent chat
Fireround - weekly funding rounds and investors, asked in plain English inside ChatGPT or Claude
Tracking who raised what usually means three newsletters and a database tab. Firecrawl built a funding feed from announcements and put it behind the chat window. Each round comes with the amount, lead investors and a link to the source article.
- Covers $10B+ in weekly deals and 8,000+ investors; try "who led AI Series A rounds this week" or "what has Sequoia led recently"
- Browsing fireround.dev is free, no account; asking in ChatGPT, Claude, Grok or Cursor runs through the Firecrawl plugin on your Firecrawl account
- Open source: deploy your own copy with a Firecrawl key and get new rounds in Slack, RSS or your own API
Only announced rounds are counted, so stealth deals won't show up. For scouting investors and peers it is a good free starting point.
https://firecrawl.dev/tools/fireround
📎 Read also:
→ ExploreYC - open data on 5,773 YC companies
→ Alumni Founder - map any company's founder network
→ Startup funding ranked by country
Several coding agents on one repository usually step on each other's files. Offrun gives each one its own git worktree
- A second agent reads the diff first, picked from your own subscriptions
- Runs on Mac through accounts you already connected: no new account and no API keys, with several logins per agent to cover usage limits
- Free and unlimited, no card: you only pay for your own agent subscriptions
https://offrun.dev/
📎 Read also:
→ Kanban board that sends each card to its own agent
→ Harden - a local guard for coding agents
→ A linter for what coding agents leave behind
Most teams write agent infrastructure themselves before they get to the product. Opengeni ships it ready-made and open
- Durable sessions with replay, sandboxes for agent code, managed credentials and memory, human approval of steps
- Multi-tenancy: every customer gets their own data and their own tokens
- Apache-2.0 license: self-host with Helm or Terraform for Azure, AWS and GCP, or use the managed cloud version, free to start
https://opengeni.ai/
📎 Read also:
→ 2,000+ production-ready API integrations for agents
→ AnySearch - a search API built for agents
→ Zero-touch OAuth for MCP servers
Europe closed its strongest quarter in four years: $25B in venture in Q3 2026, up 77% year over year (Crunchbase)
- Mistral's $3.5B round became the largest in European venture history, and four rounds of $1B or more made up about 40% of the region's capital
- AI took $18.8B, or 75%, a record share; roughly half of the money also went into physical tech: defense, data centers, energy, space and robotics
- The UK stays first with $7.5B, while Germany ($5B) and France ($4.8B) posted their best quarters since 2021-2022
https://news.crunchbase.com/venture/q3-2026-europe-strong-quarter-ai-uk-germany-france/
📎 Read also:
→ €9T Europe misses on research commercialization
→ The 2026 startup map, city by city
→ What happens between Seed and Series A in Europe
An agent's access to a database is usually all or nothing. Monospace, from the Directus team, puts one permissions model over databases, REST APIs and SaaS tools, and the data stays where it is.
- Permissions are enforced inside the query, down to rows and fields, not as a plain allow or deny at a gateway
- The same data is served as REST API, typed SDK and MCP, with no migration or copying
- Free Starter plan: 1 workspace and 3 users
The free tier is enough for a trial project, but row-level security, audit logs and SSO start with the Business plan, which is custom-priced.
https://monospace.io/
📎 Read also:
→ Swytchcode - 2,000+ API integrations for agents with guardrails
→ MCP's zero-touch OAuth for enterprise servers is now stable
→ Harden - a local guard that checks what your coding agent is about to do
54% of angels have rejected a startup for AI-washing. That is what the 2026 Angel Investment Network survey shows.
- 47% of angels reject half to three quarters of pitches within the first two minutes
- The fastest routes to a no: the founder can't explain the problem (23%), no evidence of demand (23%), evasive answers (19%)
- 62% say that honesty about problems, before they become fatal, is what sets successful founders apart
https://www.angelinvestmentnetwork.net/half-of-angel-investors-have-rejected-a-startup-for-ai-washing-finds-ain-global-investor-survey/
📎 Read also:
→ Sam Altman's blog - the founder skill is selling the future
→ AI-for-X pre-seed decks from founders who never worked in the industry
→ Schrodinger's advisor - the name on the slide who never read the deck
WIPO Global Innovation Index 2026 - VC funding grew 28% to $510B, and more and more money is going into AI
Three numbers from the report show where the money went and how many deals it was split across.
- AI took 53% of global VC deal value in 2025 and 77% in the first half of 2026
- Overall startup funding posted its biggest annual gain since 2021
- At the same time, the number of VC deals fell for a fourth year in a row
https://www.wipo.int/pressroom/en/articles/2026/article_0014.html
📎 Read also:
→ Crunchbase Q1 2026 - four companies took 65% of VC
→ Fenwick x Carta Venture Beacon - AI's growing share of VC
→ Exponential View priced real AI demand at $175B
LvlUp Ventures read 25,000 applications - what the fundable seed startups had in common
Aaron Golbin (LvlUp Ventures, 1,000+ startups backed) sums up what his team saw across roughly 25,000 applications last year.
- Of the applicants still in business a year later, close to 82% had a strong go-to-market plan already in the deck
- Over 78% of founders use AI somewhere; the ones who struggle bolt tools onto messy workflows instead of designing for them
- Standard marketing tactics in a deck are an auto-reject; they want channels the team has already tested
- Distribution gets designed before the product: Slack and Teams apps, marketplace integrations, embedded fintech
The 82% comes from one firm's funnel, not a controlled study, but it matches a pattern: the deck is now judged on distribution, not on the product.
https://news.crunchbase.com/seed/startup-funding-rules-ai-gtm-golbin-lvlup/
📎 Read also:
→ The GTM Co-Founder - a go-to-market plan for technical founders
→ NBER measured 329 accelerators
→ YC quietly doubled to ~980 startups a year
System Design Atlas - 15 architectures worked end to end, free
Founders hiring a first senior engineer, or defending their own architecture, need a shared way to talk about scale. The Atlas walks through 15 classic designs, and every diagram is live: click a component to see what it does and what it trades off.
- Ranked by how often they come up; the first five (feed, chat, URL shortener, rate limiter, notifications) cover most of what gets asked
- Each design starts from a hard number: 150k reads per second for a home feed, 50 million open connections for chat
- Also covers payments, Ticketmaster-style booking, Google Docs-style co-editing and a distributed cache
A free alternative to paywalled system design courses. The author says up front that the text is LLM-written and the numbers are interview-grade estimates, so treat it as a map, not a spec.
https://atlas-sysdes.vercel.app/
📎 Read also:
→ San Fran Sim - learn SaaS unit economics as a game
→ The Claude Code config fields the docs skip
→ IdeaProof - a free database of 1,091 company failures
Sai by Simular - an agent with its own computers for your screen work
Much of a small company's back office lives in apps with no API: supplier portals, spreadsheets, admin panels. Sai clicks and types through them the way a person would, on your machine or on cloud computers that keep going after you close the laptop.
- Built by ex-DeepMind researchers; first to reach human level on OSWorld, the benchmark for agents using real computers
- Recurring tasks get cheaper over time: a learned routine replays as code in Simulang, their open-source scripting language
- Starter is $50/month and invite-only for now; Premium with an always-on cloud computer is $200/month, Pro $500/month
The research record holds up (a Best Paper at an ICLR workshop, top OSWorld scores), but at these prices it pays off only for work someone now does by hand for hours every week.
https://www.simular.ai/
📎 Read also:
→ Prosus built 60,000 agents - what they found
→ Larridin priced the AI-native engineer at $920/month
→ OpenMarket - a marketplace where agents pick on evidence
Clueso MCP - product videos and help docs from a chat with your agent
Small teams ship features faster than anyone records a walkthrough, so onboarding videos and help articles fall behind. Clueso turns a raw screen recording, a slide deck or a PDF into a narrated video with captions plus a step-by-step article, and the new MCP runs it straight from Claude or ChatGPT.
- AI rewrites the script, cuts filler words, replaces your voice with a clean voiceover and zooms in on each click
- Voiceover, captions and docs translate into 37 languages in one click
- Free plan: 10 minutes of video exports a month, watermarked; Solo is $40/month and Business $120/month, both billed yearly
Rated 4.9 on G2, and customer quotes are all about speed: one team cut support videos from two days to two hours. The free tier is enough to test one onboarding flow, not to run a help center.
https://www.clueso.io/
📎 Read also:
→ DocsAlot - docs that stay in sync with the product
→ The GTM Co-Founder - a launch plan for technical founders
→ OpenSEO - an open-source SEO suite billed by usage
Sunk Cost - how many years until local AI hardware pays for itself
The case for running models locally is that tokens stop costing money once the machine is bought. This calculator puts a number on the word "once".
- the default scenario is unkind: a $3,499 Mac Studio running Qwen3 27B at a moderate coding-assistant load saves $0.22 a day, which works out to 43.6 years before break-even
- every assumption is editable - electricity price, tokens per day, context window, generation speed, or simply the monthly API bill being replaced
- the key-value cache math is shown alongside, so a context window that feels free turns out to cost gigabytes of memory and generation speed
The value here is the reframe rather than the verdict: it forces anyone arguing for local inference to state the workload they are assuming out loud, and 99 comments against 46 points on Show HN suggests that argument needed having.
https://sunkcost.ai/
📎 Read also:
→ Computable GPU Index - an open reference price for one GPU-hour
→ Larridin priced the AI-native engineer - $920 a month in tokens
→ PwC - where the $31.6tn AI buildout leaves compute prices
Capsule - a whole app packed into one file that sends like a PDF
Describe what is needed and it generates a single .capsule container holding the HTML interface, the schema and a local SQLite database. Send it over WhatsApp, AirDrop or email; opening the file launches the app with all data already inside.
- runs fully offline with no accounts and no servers, so there is no cloud storage to breach and no subscription attached to the data
- plain HTML and CSS inside, which means no vendor lock-in, and features or schemas can be reworked later through MCP coding tools
- free, with desktop players for macOS, Windows and Linux; iOS and Android are listed as coming soon, and the browser version previews but cannot save files
The honest use case is internal tools and one-off trackers that never deserved a cloud account in the first place, because anything genuinely multi-user is out of scope by design.
https://withcapsule.app/
📎 Read also:
→ Kanwas - open-source canvas where every doc is a plain .md file
→ KanBots - kanban board that dispatches agents into git worktrees
→ Harden - a local guard over what your coding agent is about to do
Cherry Ventures counted 10,000 European companies at the starting line, including those that never raised money: 89 cities, over €5B raised.
- 79% of companies without institutional money have a single founder; among those that closed a first round, 46%
- only 7% of founders were ever in an accelerator (657 people, 185 of them from YC, 58 from Antler)
- co-founders are 3.2 times more likely to share a former employer than a university
Amsterdam and Zurich lead in founders per 100,000 residents: about 34, twice London's rate.
https://heartbeat.cherry.vc/
📎 Read also:
→ Antler European Founder Report - Seed to Series A
→ EUR 9 trillion left on the table by research
→ Dealroom - benchmark 325 tech hubs for free
Pancake gives your agent (Claude or Codex via MCP) a brain, data and tools for go-to-market: you write the idea, the agent builds a "Play" and runs it every day.
- reads your site and builds a GTM brain: who buys from you, what to say, where to find them
- 30+ data providers in the plan (Dealroom, TheirStack, Coresignal) and live signals: funding rounds, open roles, tech stacks
- you approve every lead yourself, messages go out from your own account; $99 a month per workspace, 60,000 credits, 7-day trial with 15,000 credits (card required)
Example from the site: "find founders who just raised a seed round and are hiring a marketing lead" - 148 rounds searched, 25 leads. The site names Kong and several YC startups among its customers.
https://pancake.ai/
📎 Read also:
→ The GTM Co-Founder - go-to-market plan for technical founders
→ Elentaria - scores 22 go-to-market channels
→ ICP says who to call, not when
IDC: PC shipments fell 20.1% year over year in Q3 to 62.7M units, and 9.1% below Q2, a quarter that is normally the bigger one.
- Vendors and channels stocked up early in the year ahead of memory-driven price hikes, so Q3 had little demand left to draw on
- Lenovo, HP and Dell fell faster than the market (-22.6%, -30.9%, -25.0%) and together lost 4.2 points of share; Apple fell 11.3% and gained a point
- Channels are carrying inventory, so promotions are possible, but IDC expects prices to stay elevated and the outlook into 2027 to worsen with the macro picture
Preliminary data from the IDC Quarterly Personal Computing Device Tracker.
https://www.idc.com/resource-center/press-releases/idc-pc-tracker-3q26/
📎 Read also:
→ PwC Global Data Centre Outlook - where the AI buildout leaves compute prices
→ Bluevine cost report - the expenses nobody budgets
→ Brex ranked the 25 fastest-growing vendors, 14 are not AI
IrisGo observes the work, learns from it and then runs those workflows itself, like an executive assistant for a solo founder
- Ranked #1 product of the day on Product Hunt: 524 votes
- Backed by Andrew Ng, covered by TechCrunch and Fast Company
- Currently a free beta; per Tech in Asia, the paid tier was planned at $20 a month
https://irisgo.ai/
📎 Read also:
→ Kanwas - a canvas shared by humans and agents
→ ChartMogul AI - an analyst that explains ARR moves
→ zero.xyz - agent access to ~8,000 tools
Viktor - an AI coworker that lives in Slack or Teams and does the work itself instead of suggesting what to do
- Connects to 3,200+ tools: CRM, ads, analytics, Notion, finance
- Free to start: up to $100 in credits, no card, credits never expire
- Paid plans start at $50 a month
69,000+ teams is the company's own claim on its site, not independently verified
https://viktor.com/
📎 Read also:
→ OtoDock - a self-hosted company OS with agents
→ Prosus built 60,000 AI agents in 18 months
→ OpenMarket - where evidence beats copywriting
A record 27 rounds of $1B or more in a single quarter: Crunchbase counted global venture for Q3 2026
- $159B across almost 6,000 startups: the lowest quarter of the year, but higher than any quarter since Q2 2022
- 27 companies raised $1B or more each (16 in Q2, 14 in Q1), together about a third of all capital in the quarter
- AI took $102B, which is 64% of global venture, 14 percentage points more than in Q3 2025
https://news.crunchbase.com/venture/q3-2026-global-startup-funding-ai-billion-dollar-rounds-exits-data/
📎 Read also:
→ Startup funding ranked by country
→ $3T of dry powder stuck in old funds
→ Where $6T of AI infrastructure money is heading
The possible sale of Monzo has again raised the question of who ends up owning the winners of Europe's venture race.
- This year non-European investors bought 173 European VC-backed companies for €18.8B
- That is 37.9% of deals by count, the highest share this decade
- Median deal: €120.6M with a non-European buyer against €22M with a European one
Part of the gap is maturity: non-European buyers take later-stage companies, and their share fell only at pre-seed and seed. Data from PitchBook research.
https://pitchbook.com/news/articles/ownership-of-europes-biggest-startup-acquisitions-is-shifting-abroad
📎 Read also:
→ AVP Transatlantic Index - four boxes EU founders need to raise in the US
→ CEPR - financing the startup J-curve when American money backs a European company
→ Tracxn - startup funding ranked by country
Deep Science Startup Tracker - WIPO and Dealroom built an open map of startups and university spinouts that grew out of frontier research: from semiconductors and robotics to quantum technologies and space.
- The share of such companies among all VC-backed startups hit 16% in 2025, a record, and their combined value is up 23% from 2024
- The US has 12,752 such companies, China 3,466, the UK 2,133; since 2020 the fastest growth is in Central and Southern Asia (118%)
- According to Dealroom, 10% reach late stage against 4% of all VC-backed startups
https://www.wipo.int/en/web/global-innovation-index/2026/deep-science-innovation
📎 Read also:
→ Redstone VC - the €9 trillion research-to-startup gap in Europe
→ AI unicorns publish almost nothing - someone finally counted
→ Hurun Global Unicorn Index - 903 of 1,603 did not move
McKinsey's State of AI 2026: 32% of companies skipped buying software because their coding agents can build it in-house.
The annual survey (1,719 respondents, 97 countries) finds AI everywhere except the P&L:
- 80% say AI made them personally more productive, yet only 37% see any EBIT impact, flat vs last year
- Large companies scaling agents went from 27% to 40%; smaller ones stayed at 22%
- About 1 in 5 say AI operating costs, tokens included, already cap their usage
- 39% expect AI-driven headcount cuts next year, though last year twice as many expected cuts as later reported them
For anyone selling software, the build-vs-buy number matters most: the customer's own coding agent is now on the competitor list. Free 31-page PDF, no email gate.
https://www.visualcapitalist.com/wp-content/uploads/2026/08/the-state-of-ai-in-2026-on-the-road-to-roi_final.pdf
📎 Read also:
→ Deloitte: only 11% run AI agents in production
→ WEF convergence report and why startups should cut
→ AISlop - linter for code left by AI agents
Most Hated Tools - a public vote on which work software people would delete first
Plenty of SaaS companies started with a tool everyone merely puts up with. This site pits two workplace tools against each other and asks which one you would rather delete from your life; 15,000+ votes so far.
- SharePoint is currently number one
- No sign-up, votes capped per browser; the makers call it unscientific themselves
- Anyone can suggest a tool, so the list keeps growing
Not research, but a quick read on pain: whatever tops this list is where users are already looking for a way out, which is exactly the market a new product walks into.
https://www.mosthatedtools.app/
📎 Read also:
→ Brex ranked the 25 fastest-growing vendors
→ Common Paper - what SaaS customers now refuse to sign
→ IdeaProof - a free database of 1,091 company failures
Whiteboard (YC W26) - a design session with your coding agent, on a canvas
Coding agents now write more code than a small team can read, and the risk is merging changes nobody understands. Whiteboard plugs into Claude Code or Codex and has the agent draw its work: sequence and entity diagrams you click through to the code behind them.
- A semantic diff viewer written in Rust shows only the changes that matter: big new functions as pseudocode, tests and docs collapsed
- A decision log shows which choices the agent made on its own and how they shaped the change
- Free and open source (MIT), runs on local checkouts; macOS only for now, Windows is on the roadmap
330+ points on Show HN this week. Worth a look for any founder who reviews agent PRs and wants to still understand the system six months from now.
https://github.com/devdotfast/whiteboard
📎 Read also:
→ Harden - a guard for what your coding agent is about to run
→ KanBots - parallel coding agents with a hard cost cap
→ Kanwas - a canvas where humans and agents share context
NOAN - one approved set of company facts for every agent you run
Once several agents draft proposals, answer customers and edit the site, each works from whatever it was prompted with months ago, and they start quoting different prices. NOAN stores company facts (pricing, positioning, objection answers) as approved, versioned records and serves them to any model or agent through one API and MCP.
- Change a fact once and every site section, app and agent that reads it gets the new version, no redeploys
- Agents can propose new facts, but nothing enters the base until a human verifies it; Time Travel shows what the company believed on any date
- Free for 1 project and 40 facts; Starter is $99/month with unlimited facts, Team $299/month for 5 seats
Social proof beyond Product Hunt is thin, so start on the free tier. The problem it names is real, though, and it shows up the week a second agent goes live.
https://getnoan.com/
📎 Read also:
→ OtoDock - a self-hosted company OS with agents in departments
→ Kanwas - a shared product context for humans and agents
→ DocsAlot - docs that agents can read
Financing the startup J-curve - what changes when American money backs a European company
High-growth startups lose money before they make it, and whether a company can hold that curve depends less on the idea than on how deep the pocket behind it goes. Three economists tested this on Sweden, where US and non-US investors operate under identical rules.
- 137 US venture funds above $1 billion between 2013 and 2023, against 11 in the EU and ten in the UK - the gap is the mechanism, not the symptom
- same country, same regulations: Swedish startups backed by US VCs run deeper operating losses, draw more follow-on capital and post higher sales later
- 134% more new investors join after the first round when the lead investor is American, and that is the part which compounds
The practical read for a European founder is that the handicap sits in depth of capital rather than in quality of ideas; the authors apply the same test to the Scaleup Europe Fund, asking whether it pulls private money in behind it rather than whether it manages to deploy its own.
https://cepr.org/voxeu/columns/financing-startup-j-curve
📎 Read also:
→ Antler mapped what happens between Seed and Series A in Europe
→ €9 trillion left on the table by Europe's research institutions
→ $300B in Q1 2026 VC - four companies took 65% of it
How Stale Is Your AI? - a live table of when each model was trained and how blind it is now
Twenty models from eight labs, each with its release date, its training cutoff, and a running counter of how many days of world events it has never read.
- median model age is 51 days, but the spread is brutal: Llama 4 has been blind for 747 days while DeepSeek V4.1-Flash sits at the fresh end
- only 10 of the 20 publish a cutoff at all, so for the other half the honest answer about what the model knows is that nobody outside the lab knows
- free, no signup, counters tick live in the browser
Worth a bookmark specifically when picking a model for a feature that depends on current facts, which is exactly where an unpublished cutoff turns into a support ticket three months later.
https://stale.jock.pl/
📎 Read also:
→ Larridin priced the AI-native engineer - $920 a month in tokens
→ Brex ranked 25 fastest-growing vendors - 14 are not AI
→ Exponential View priced real AI demand - $175B, 0.42% of GDP
Wisry - agents that study competitor ads and rebuild the winners for a store
Paste a store URL and it builds brand memory: products, voice, visual identity, audience. A research agent then reads competitor ads running on Meta and TikTok and pulls out what is already converting.
- every campaign angle ships with source citations - which live ad the hook came from, not a model's guess
- generates static and video creative from the winning patterns, pushes them to Meta and Google, and moves budget toward whatever performs
- $99 per month introductory instead of $199, for 2,000 credits: roughly 106 static ads, six videos and four research runs
Built for ecommerce and DTC rather than SaaS, and the research layer is the real product here; the advertised +200% performance boost is the vendor's own figure, so treat it as a claim and not as a benchmark.
https://wisry.ai
📎 Read also:
→ Omeda State of Audience - events beat ads, and only 9% use their data
→ OpenSEO - an open-source SEO suite billed by usage
→ Repaint - redesign your site without losing SEO, $20 a month