←
StartupHub.ai2,892 posts
StartupHub.ai avatar

StartupHub.ai

@StartupHubAI

The World’s #1 AI Startup Directory.

◷ Joined May 2019▧ 385 Photos and videos
StartupHub.ai avatar
StartupHub.ai@StartupHubAI·1h

Everyone benchmarks models. Almost nobody benchmarks the harness. We sent 5 real AI agent harnesses (Cursor, Claude Code, Codex CLI, OpenCode, Gemini CLI) across 878 websites to answer the same 8 questions, that's 5,779 runs total. A plain model fetched 1.3 pages and stopped (2.49/8 answers). Cursor fetched 8.2 pages and found 71% more (4.25/8 answers). For founders, the takeaway is simple: Agents only explore 5–8 pages per run. Bury your pricing or docs deeper than a few links from the homepage, and you're invisible to them. Full breakdown + data table in the reply ↓

Post attachment
StartupHub.ai avatar
StartupHub.ai@StartupHubAI·3h

We ran 5,779 AI agent tests on 878 websites. The harness decided what agents found Five real agent harnesses and a plain model answered the same eight questions about 878 websites. The harness decided how much of each site the agent reached.

Post attachment
StartupHub.ai avatar
StartupHub.ai@StartupHubAI·3h

What AI assistants actually read on 25,000 startup profiles We tracked AI assistant traffic across 25,672 startup profiles for four weeks. Claude fetched 37,748 pages live. ChatGPT fetched 4,976. No assistant requested llms․txt.

Post attachment
StartupHub.ai avatar
StartupHub.ai@StartupHubAI·4h

Startup signals, 18 November: New launches: Fincero, Resans, Setup Dream, Contrefort Coaching Hiring jumps: Fireworks AI (104 roles), Flatpay (98 roles) Pricing changes: Tesorai

Post attachment
StartupHub.ai avatar
StartupHub.ai@StartupHubAI·4h

Startup funding, 18 November: 25 rounds, $2.2B raised. Biggest: Shield AI $1B (Series H) Payhawk $280M (Series C) Helical Fusion Co., Ltd. $126M (Grant) Ramon․Space $124M (Series B) Boltz $110M (Series A) Variational $110M (Series B)

Post attachment
StartupHub.ai avatar
StartupHub.ai@StartupHubAI·4h

The biggest tech stories of 18 November: 1. Neon brings cheaper frontier defense 2. Cloudflare released open-weight Clairic extraction models 3. Databricks makes serverless move a review step 4. Your AI Second knows before you have to ask

Post attachment
StartupHub.ai avatar
StartupHub.ai@StartupHubAI·Nov 17

ChatGPT for Sales. ChatGPT can build a target list from Salesforce, HubSpot, Close CRM or monday CRM. It adds signals from Apollo, Clay and Common Room, then drafts outreach in Outreach or Salesloft. After the call it pulls Gong, Otter or Granola notes to draft follow-ups and push next steps to Attio. NeoBid co-founder Gunnar Gombert, running real estate prospecting on HubSpot, puts that saving at around 65% less manual effort. Your dot can reuse the same plugins to prep briefs and flag deal changes for review. startuphub.ai/ai-news/artifi…

StartupHub.ai avatar
StartupHub.ai@StartupHubAI·Nov 15

The #1 market map maker. ✔️ 2x2 matrix. ✔️ AI auto update. ✔️ Publish with backlinks.

2:07
StartupHub.ai avatar
StartupHub.ai@StartupHubAI·Nov 11

If you are asking which website builder do AI crawlers read, the answer depends on what is served before scripts run. Most AI built sites look great in a browser. Many are almost invisible to AI crawlers. I reviewed StartupHub.ai research on which website builder do AI crawlers read before scripts run. startuphub.ai/ai-news/artifi…

AI.Engineer avatar
⟳ StartupHub.ai retweeted
AI.Engineer@aiDotEngineer·Nov 8

One. Million. AI Engineers. thank you.

1:47
Mila Chervenkova 🍩 avatar
⟳ StartupHub.ai retweeted
Mila Chervenkova 🍩@MilaChervenkova·Nov 6

A profile link is not a parasite page. Both are published on third-party domains, but that is where the similarity ends. Here is the difference (+examples): 1. A Parasite Page targets a specific search query. 2. It contains substantial, standalone content that fully addresses the search intent. 3. It is created to rank in Google and attract its own organic visitors. 4. The link to your website supports the content but isn’t the page’s only reason to exist. Examples of potential Parasite Pages include: 1. A complete guide published on Hashnode 2. A hosted page or microsite on Neocitiesne 3. A hosted page or microsite on Neocities 4. A public document on HackMD 5. A substantial public note on Evernote 6. A creator article on Buy Me a Coffee 7. A post on Tumblr A Profile Link is different: 5. It primarily identifies your brand, business, or website. 6. It usually contains a short description, basic information, and a backlink. 7. It isn’t built around a specific search query or intended to attract its own organic traffic. Examples include: 1. Link-in-bio pages on mssg .me, start .me and heylink .me 2. Profiles on ArtStation, Flickr and Dead .net 3. Directory listings on SoMuch, ProvenExpert, Find-Us-Here, StartupHub, DesignRush and similar platforms Keep in mind that public visibility doesn’t mean Google can index the page. Always confirm that the individual page is crawlable, indexable, and capable of appearing in search results. Always use an indexation tool for profile links and parasite pages.

MacroGuru avatar
⟳ StartupHub.ai retweeted
MacroGuru@macroguru9·Nov 2

Revolut Says Self-Hosted GLM Cut Its AI Coding Costs 90% Revolut moved 2,000 engineers off metered AI APIs onto self-hosted open models running on 56 Nvidia B200 GPUs, and says that cut its AI coding costs 90%. David Abitbol of Revolut gave the figures in a talk posted on the AI Engineer YouTube channel on 29 October. Abitbol said the bank serves GLM 4.5 at about 30 cents per million tokens, which StartupHub reports as one tenth of its prior Anthropic and OpenAI invoice.

Michal Bláha avatar
⟳ StartupHub.ai retweeted
Michal Bláha@michalblaha·Nov 3

The primary source is a conference talk by David Abitbol, who leads AI operations at Revolut. He gave it at AI Engineer NYC 2026 on October 13, 2026, under the title "Shipping like a startup, regulated like a bank". The AI Engineer YouTube channel posted the recording on October 29 as "How Revolut Cut AI Coding Costs 90% With Self-Hosted Open Models – David Abitbol, Revolut". Talk page with transcript: ai.engineer/nyc/2026/talks/r… Video: youtube.com/watch?v=Ok6QdZHC… The post on X appears to be built from secondary write-ups. StartupHub.ai published the first one, and FourWeekMBA, Trending Topics and others repeated it. What the transcript supports: 👉 The spending problem: Revolut started with strict budget limits on Claude Code. Abitbol said engineers were spending "a thousand a month, 10,000 a month with not so much ROI." 👉 Claude and GPT kept as exceptions: He said "the frontier is the exception" and that Revolut keeps Codex and Claude Code as exceptions. 👉 The cost: "Self-hosting GLM, our blended cost of token is 30 cents per million, and so our invoice is one-tenth of what we were spending depending on Anthropic and OpenAI." 👉 The metrics: self-hosting the GLM model resulted in a 60% improvement in coding time and a 70% improvement in lead time. Abitbol credited the larger lead time gain to tasks outside pure coding, like inbox triage and meetings, that engineers now automate when tokens are not rationed. What the post gets wrong or adds: ❌ Together AI isn't in the talk. The transcript never names a GPU provider and never says the GPUs are rented. It only says to serve 2,000 engineers, we roughly need 56 B200. And that's to host GLM 5.3. Abitbol calls it self-hosting on dedicated GPUs. "Rents them from Together AI" seems to be the author's addition. ❌ "All it needs is GPUs" is too simple. The talk spends a lot of time on tuning. Revolut's tuning slide shows 76% higher requests per second and 88% lower time to first token on the same 56 B200 GPUs. StartupHub also notes that the 90 percent savings exclude the costs of cloud, staff and monitoring. ❌ The $200B valuation doesn't come from the talk. Trending Topics, reporting on the same talk, gives a recent valuation of about $115 billion. I didn't check Revolut's current valuation. ❌ The GLM version varies between sources. The transcript mentions GLM 5.1 for the productivity gains and GLM 5.3 for the current deployment. StartupHub wrote GLM 4.5, which looks like an error. ❌ The conclusion is the author's own. The claim that companies will copy this and that frontier labs face a revenue problem is the X author's opinion, not something Abitbol said. Sources: AI Engineer NYC 2026 – Shipping like a startup, regulated like a bank (David Abitbol, Revolut) YouTube – How Revolut Cut AI Coding Costs 90% With Self-Hosted Open Models StartupHub.ai – Revolut says self-hosted AI cut coding costs 90% Trending Topics – Revolut Cuts A.I. Coding Costs 90% With Open-Weight GLM BigGo – talk summary with quotes