Talk to Me, Arthur.

No 104 | Meet Arthur, the AI we just hired. He's read every email, every meeting, every deck we have ever seen.

"Because I was inverted." — Maverick, Top Gun (1986)

Hawaii, June 1997. I am on vacation with Lisa. jcrew.com is a week old, built with Arthur Cinader Jr. and Ohad.¹ I pull it up on hotel dial-up, spot a typo on the homepage, view source, save the file, edit it, re-upload to the server. Ten minutes.

The site goes down in six.

Turns out the saved-source page has session IDs baked into every link. Every new visitor inherits somebody else’s cart. Havoc.

I have been building like this since 1980.

My dad brought home an Apple II Plus when I was six. Then a IIGS. Then a Mac. I have been an Apple user and a self-taught coder ever since, with a product manager’s brain bolted onto a designer’s eye and an operator’s stomach for shipping things that are not quite ready. I dropped out of college in 1995 to start Neptune Interactive, a dial-up ISP, because I had seen the web and could not sit in a classroom anymore. Two years later I was in Hawaii, trying to fix a typo and learning the hard way about sessions.

I’ll say this plainly. If I were in college today, I would drop out again.

The last six months are the reason.

At Sugar Capital, we just rebuilt our entire internal stack from scratch. Code name: Project C6. The chemical formula for sugar is C₆H₁₂O₆. My dad’s college nickname was C6. (Wish I had nerdier friends and that nickname. Instead I get B.Sug, which is fine. But how badass is C6?)

The name does two jobs at once. Version one was the duct-tape rig we had been running for the last few years, a PHP app with Airtable as the backend. Version two is not a CRM. It is a place where every thread, note, deck, meeting, and follow-up we have ever touched lives in one structured brain, searchable by vibe instead of keyword.

I built most of it at night, on planes, between meetings.

I want to be careful here, because this is the part people get wrong. I am not telling you AI wrote the software. I am telling you the software I have wanted to build for twenty years finally got built. AI collapsed the distance between the thing in my head and the thing on the screen. I still made every design decision. I still broke production three times. Five and a half weeks. 115,000 lines of code. About a year of work for a five-person engineering team, shipped by one product manager who still uses vi.

That is the leverage shift. It is not that AI replaces the builder. It is that AI lets the builder finally move at the speed of their own mind.

Upload your deck. Get roasted in 90 seconds. sugarcap.com/arthur

The thing at the center of Project C6 is an AI. We named him Arthur.

The name is a tip of the hat. Arthur Cinader Jr. is one of my oldest, closest friends. Lisa calls him the older brother she never wanted. He hired me to put jcrew.com online in 1996. His father, Arthur Cinader, started J.Crew, and was involved in jcrew.com as much as Arthur could get him excited about it. Arthur was a founder of POPSUGAR. He is an LP in Sugar Capital today.

He has four daughters. We have three. We both live in San Francisco now, on the same block, close enough that we do not have to cross a street to get to his front door.

The Cinaders come with a curmudgeon streak that sharpens with age. Arthur Sr. had it down. Arthur Jr. is getting there.

Our new Arthur lives in our group chat. In email. In chat. He came with it too. He launches with this essay. He has read every email since I registered the domain in 2019, every Granola meeting, and every deck we have ever received, which in retrospect feels like something I should have been more careful about.

He does not get tired. Every conversation makes him a little more useful. He forgets nothing about our deals, our preferences, or the times he’s been wrong.

The first deck I fed him was Grüns’, two and a half years after we invested.² I wanted to see what he would do with it. He came back with this:

The format insight is real. Athletic Greens proved people will pay $3.60 a day for a comprehensive greens product, and gummies are the fastest-growing supplement format at 17% CAGR. Grüns threads the needle, comparable nutrition density at $1.95 a serving with 60%+ gross margins at launch that expand to 72% at scale. A third-time founder with exits who has advised 10 profitable CPG brands is exactly the operator you want running a DTC subscription business.

He saw what we saw.

The stack is boring on purpose. Next.js on Vercel. Supabase for everything stateful. Anthropic for Arthur. OpenAI for embeddings.³ Aside from Vercel and Supabase, every service is swappable in an afternoon. Own your data, rent your tools.

We pulled him out of our internal stack and gave him a side gig, free for the nearly 10,000 of you who read AirSugar. We call it Not Sugar Coated.

You upload your deck. Arthur reads it. You get a verdict. A score. A steelman, which is the best possible version of your pitch, written back to you in clean prose. A pull quote you will either frame or delete. A shareable memo page with its own OG card, in case you want to prove to the world that an AI just roasted you in under a minute.

It is free. It is fast. It is a little mean.

Founders who have tried it say it is the most useful 90 seconds of feedback they have gotten on their deck. Not because Arthur is always right. He is not. But because he says the thing every investor is thinking and nobody says out loud.

Try it: sugarcap.com/arthur.

If you opt in, your deck lands in our deal flow. If you do not, you get the memo and we never see it. Your call.

Reading the code. Changing the code. Shipping the code. The distance between idea and product has collapsed twice in one lifetime.

If you are in college right now, and you can feel it, do not wait.

Drop out. Build the thing.

1. An order-taking page had shipped six months earlier, but you needed a paper catalog open next to you to enter item codes. June 1997 was the first browseable version of jcrew.com.

2. Grüns is a Sugar Capital portfolio company; Unilever announced the acquisition in April 2026. Unilever Ventures is an LP in Sugar Capital Fund III

3. The full stack: Next.js 16 with the App Router and React 19, TypeScript 5, hosted on Vercel with deploy previews on every commit. Supabase for Postgres, auth, storage, and row-level security, which replaced what used to be Airtable plus Auth0 plus S3 plus a prayer. pgvector for embeddings, so semantic search lives inside SQL. The AI layer runs on the Anthropic SDK for Arthur himself, memos, SMS replies, briefings, steelmen. OpenAI’s text-embedding-3-small builds the semantic index and GPT-4o handles deck text extraction. A custom Postgres function blends keyword and vector similarity so Arthur can find that founder I met last spring pitching a fintech thing without knowing the exact words. Frontend is Tailwind v4 and shadcn/ui, a design system we own. Tiptap for rich-text notes, @hello-pangea/dnd for the deal kanban, @tanstack/react-table for the 2,100-row company view. React PDF renders the investor memos Arthur emails to founders. @napi-rs/canvas renders the dynamic OG image on shared memos so when a founder drops an Arthur roast in iMessage it actually looks like something. Integrations: Google Calendar and Gmail, Granola MCP for meeting transcripts, Apollo for contact enrichment, Sendblue as the iMessage gateway so Arthur texts me in blue, Resend for transactional email, Cloudflare Turnstile for bot protection, GA4 through @next/third-parties. Dev and CI: Vitest for unit tests, Playwright for end-to-end, ESLint 9, GitHub Actions with type-check gated, every push auto-deploying to Vercel. Yes, Claude Code helped build the thing.