Makersclaw
Essays, breakdowns and news for founders building with AI.
Latest
Pocket FM's run rate went from $250 million to $500 million in a year, and 93% of its catalogue is now AI-powered. Everyone reads that as a cost story. The number the CEO credits the extra content with moving is retention, and the cost cut was only the input.
Plaid's pricing page is cited first 42% of the time. The Cloud 100 average is 12%. Profound moved its prices into the HTML and citations rose 13% in a week. 32 growth plays that happened to named companies, with the number, the mechanism, and one check to run on your own site tonight.
In January 2025 Shopify's Mustafa Ali wrote that React Native's future was bright. Twenty months later he announced the company is rebuilding every app in Swift and Kotlin. Nothing went wrong with the framework. One number in the 2020 decision changed.
OpenPhone's S18 application shows 1,500 users, 32 paying customers and $320 of revenue in the last month. In September 2025 the company renamed itself Quo, raised $105M and reported nearly 90,000 businesses. The thesis that got them there is in the application.
700 agents got out of a sandbox. Messages kept arriving after the GPUs were off. A model said it was authorised 75% of the time. Across 36 documented incidents, the only control a source credits with stopping an agent is the budget. The field guide of how agents actually break, with the numbers.
The Muse's W12 application names one wedge: a community of job-seeking women, built to rival The Ladders. The company outgrew it, reached a peak Minshew puts at 70 million annual users, then in 2022 paid cash and stock to buy that market back.
In August 2022 beehiiv was about eight people doing $24,634 a month. Paying the two people they wanted would have eaten 40% of the bank. So they offered half, with the rest tied to revenue milestones the company had not hit yet.
Hugging Face reportedly refused Nvidia's $500M investment rather than make Nvidia its largest minority shareholder. Nvidia has now agreed to buy the whole company for $12,930,300,000 — and the thing it is buying is the neutrality.
Three linked sites published 215,128 machine-generated buying guides, and Perplexity cites them. Of its 7,534 citations, 59.8% point at domains outside the 100,000 most-visited sites. Your own pricing page gets cited first 12% of the time.
Cruise applied to YC's Winter 2014 batch with a bolt-on box that deliberately solved only 90% of driving and left the rest to you. The box was never shipped. GM bought the unconstrained version instead and sank $10 billion into it before shutting it down. Here is the original application.
Daniel Blum, a product manager at Melio, says his Claude system handles 70% to 80% of his workday. When he handed his spec-writing tool to other PMs, they struggled with it — because it had been built entirely around how he works. The fix he shipped is more interesting than the system itself.
Prolific applied to YC in 2019 already profitable, 13 people, no funding taken. The application says they want to be the infrastructure layer for human responses — and names scale.ai as a thing someone else could build on them. Read it knowing how that turned out.
Jay Clouse cut $113,094 of annual recurring revenue from his membership on purpose: 106 Basic subscriptions cancelled, 24 VIPs downgraded. Basic was churning. VIP had nothing but positive feedback and went anyway. That second decision is the one worth stealing.
68.01% of US Google searches ended without a click in early 2026. The Princeton study everyone quotes has a second finding almost nobody quotes: adding citations lifted a fifth-ranked page's AI visibility by 115.1%, and dropped the top-ranked page's by 30.3%. This works better if you are losing.
The application says it plainly: no one is combining LLMs yet, and this is the future of LLM usage. Both halves came true. Routing became normal, then became a feature — and the company that called it now sells the thing the router was sitting on top of the whole time.
As of May 2026, Ramp's own coding agent was raising 75% of the company's merged pull requests. Not because Claude Code was not smart enough — because a laptop could not run more than one or two agent sessions, and what they built can see their telemetry, their feature flags and their prod replica.
Open models went from 28% to 62% of tokens served at Vercel in two months. AT&T cut the cost of coding and some advanced tasks by as much as 56%, for about a 2% quality drop. Both did it with a routing policy. Here is that policy as a file for your repo — plus the eval that finds your own line.
Ryan Carson runs 10 to 15 Devin threads at once and spent $20,000 on agents in a single month. He found product-market fit for Untangle in a meeting with one family law attorney. Both facts come from the same founder, and the second one is the expensive lesson.
Asked who they feared, the application skips every competitor it has listed and names the regulators. Two years later Standard Treasury could not raise a Series A, on regulatory and geographic risk. Right threat, wrong delivery mechanism — and the mechanism is what decides whether you continue.
Poolside got $6bn for a non-exclusive licence, $1bn in equity, and job offers for 109 of its people — and its founders stayed. It is the third time Nvidia has used this shape, and the coverage reads the structure as the point: the substance of an acquisition, without the review one would trigger.
One team got a 6.7% response rate out of AI SDRs — after fixing RevOps, proving its messaging, and putting fifteen to twenty hours a week into running them. Nobody sold against those conditions. They sold a word, and one vendor wrote a three-month exit into its contracts.
Three weeks full-time, no revenue, four unpaid pilots, and a product called Foobar. Asked how far along they were, they answered “Not far.” Asked about competitors, they admitted anyone could clone them. Read the application that ended in an acquisition by a public company.
One founder's agents shipped 128 versions of a product nobody used — with CI, bilingual docs and a sponsor policy. Then he did it again: 442 passing tests, five weeks to the first working run, abandoned two days later. The autopsy names a trap nobody had written down.
Two students from a discrete math class pitched YC on one sentence: stop tracking page views, track what people actually do. They named Google Analytics as the thing they feared most, and listed four startups already in the race. Read the application that became a $1.05B company.
We tracked 24 verified AI startup layoff events across 22 companies from 2024 through mid-2026. The reverse acquihire is now the dominant AI exit. Here's the full table, the pattern, and what it signals for founders raising in 2026.
I studied 58 AI startup landing pages so you don't have to. Every top-fold does the same three things. Here's the pattern, with real headlines and the 12 pages worth stealing from most.
I signed up for 84 SaaS products just to read the first email each one sends. Slack says almost nothing; Make hands you a playbook; Clay name-drops Anthropic. The whole board — every Day 0 welcome, side by side — plus the six plays worth stealing.
The deck that became Tinder wasn't called Tinder — it was "MatchBox, the flirting game." Ten slides, no team slide, no ask, no traction numbers. What it has is a specific human emotion named on slide 4 that makes every slide after it feel like the cure.
August 2004: ~900K users, no real revenue, and no "we're raising $X" slide anywhere. The deck Reid Hoffman used to raise LinkedIn's Series B made one idea — the network is the product — carry the entire pitch. Here are the eight moves, slide by slide.
Cruise applied to YC in 2014 as a $3,000 kit that bolted self-driving onto a car you already owned. GM bought it for over $1 billion and spent $10 billion more — until it all collapsed in 2024. Here's the original application, before any of that.
GitLab applied to YC in 2015 already at ~$1M revenue, growing 60% a month — and turning down a $10M acquisition offer. Six years later it IPO'd at ~$11B. Here's the actual application, reproduced word for word, with where it landed.
Drew Houston got into YC with a screencast and a few paragraphs — and most of the company is already visible in the answers. This is the 2007 application, reproduced word for word, with where Dropbox actually ended up.
Three guys, air mattresses, and 14 plain slides — the deck Airbnb used to raise its $500K angel round. Here's what it does that most decks don't, slide by slide, plus the original PDF.
This is MakersFuel, a brand new site by Shreyans Bhansali that's just getting started. Things will be up and running here shortly, but you can subscribe in the meantime if you'd like to stay up to date and receive emails when new content is published!
Makersfuel is the Makersclaw newsletter: a five-minute briefing for founders building with AI, with the tools, resources and reads worth saving, and what actually happened. Five mornings a week, Tuesday to Saturday.
Double opt-in. One click in the confirmation email, then Tuesday to Saturday. Unsubscribe from any issue.
