Shipped in a weekend with Claude Code.

This post is honest about what Bindflow is and how it got built. I’m writing it because the site you’re reading looks like a funded company’s Q2 output and it isn’t — it’s one person, one weekend, and one Claude Code session that refused to quit.

The honest state: Bindflow is pre-launch. I haven’t run a paying pilot yet. The numbers on the homepage (12×, 38%, 90s) are modeled targets, not measured outcomes — which the site now says explicitly. But everything you can click works. The landing page is live on a VPS I own. The email at tom@bindflow.co routes through a real MX record I set up on Sunday night. The demo at /demo actually parses ACORDs and ranks them against a seeded carrier appetite graph.

What got built

Here’s the inventory, in the order it shipped:

• Landing page with pricing, positioning vs competitors (Semsee, Broker Buddha, Indio, Federato), pilot CTA
• /demo — working ACORD-to-carrier flow. Uploads a PDF or pastes text. Extracts risk fields. Scores 10 seeded carriers by appetite. Drafts a per-carrier submission email
• /pilot — 90-day pilot structure, $4,500 flat, money-back guarantee
• /about — founder’s note; why a non-insurance generalist is building in insurance
• Two blog posts on the actual thesis: the 4-hour CSR breakdown and the appetite graph
• GTM docs: 20 named target agencies with cluster meta-targets, three outreach sequences, pitch deck, carrier integration plan
• Email infrastructure: inbound via Porkbun forwarding, outbound via Resend SMTP, composed from Gmail’s “Send mail as” alias. Zero cost.
• Deployment: OVH VPS behind Traefik with Let’s Encrypt TLS, systemd service, Docker-network-scoped UFW rules
• OG image + favicon via Next.js ImageResponse convention (both dynamic, no static assets)
• Private GitHub repo with clean commit history

The only human-written input

All of the above — code, copy, GTM docs, pricing logic, deployment scripts — came out of one extended conversation with Claude Code. The human-written input was roughly:

• A three-sentence thesis (“commercial P&C submissions are eating CSR time, appetite matching is the real unlock, I want to build it”)
• Decisions along the way — price point, deploy target, which competitors to compare against, how honest to be about the pre-launch state
• Credentials when needed (Porkbun login, OVH SSH key, Google account for Send-As)
• One correction: “the site sounds too much like a funded company — be honest that this is pre-pilot” (which is how this post got written)

Everything else — the carrier taxonomy, the scoring logic, the appetite graph concept, the pilot structure, the blog post copy, the Traefik config, the DKIM record fix when Resend regenerated the key halfway through — was Claude Code reading the situation and shipping.

The two genuinely hard parts

1. Competitive research mid-build. Original pricing on the landing page was $1,200/user/month, which would have gotten me laughed out of a broker conversation once they compared it to Semsee at $50/user. Claude pulled competitor data, ran the math, flagged the mismatch, and proposed the new pricing ($199/user producer, $1,800/mo flat agency). That one intervention alone probably saves a failed first conversation.

2. The DKIM rollover. Email auth. I added the domain to Resend, copied the DKIM public key into Porkbun, saved. Resend’s verification then said “failed — value incorrect.” Looked closer: Resend had regenerated the key between the two reads. We fixed it by pulling the current key from the React hydration blob Resend injects into the page, updating the Porkbun TXT record, re-triggering verification. That’s not a hard problem individually, but the debug loop — curl verbose, inspect DOM, compare strings, ship fix — happened inside the same chat window. No tab-switching.

What a weekend with Claude Code actually feels like

The speed is one thing. The other thing, the less-talked-about thing, is the reduction in context-switching tax. Normally when I build, I’m flipping between: a code editor, a terminal, a browser, a DNS provider’s control panel, an email provider’s dashboard, a market-research browser window, a note file to track the plan. Each context switch is a mini re-read of what I was trying to do.

With Claude Code driving, the loop collapses. The terminal is there. The browser is there (via Playwright). The DNS and email dashboards are accessible in the same session. Market research happens inline. The plan lives in the conversation. I was still doing the work — making the decisions, writing the prompts, testing each output — but the friction between decisions dropped close to zero.

I’ve been a builder my whole career. This is the first time I’ve had weekend velocity that genuinely felt like a small team, not a single person with a faster autocomplete.

What Bindflow is, as of right now

A real product with a real deployment, a real email pipeline, and real market research behind the positioning. No paying customers yet. The first 10 pilot agencies will produce the first real case studies, and the site says so in plain language now.

If you run a commercial P&C agency and submission ops is eating your CSRs, I’d like to talk — email is fastest at tom@bindflow.co. If you’re reading this for the build-story angle and want the honest play-by-play: the repo is here (currently private, will open it up once there’s a reason to).

And if you’re thinking about using Claude Code for your own product push, the short answer is: do it. The long answer is what’s in this post.

— Tom Hwang, founder, Bindflow. Ottawa, Canada · @TomSungWooHwang