Sample project · Build & Launch Sprint
A website that talks to every visitor and runs the whole funnel.
The brief: rebuild a technology partner's site as an AI-native marketing surface. It composes itself around each visitor's situation, runs the commercial funnel end to end, and was built almost entirely with AI, in one codebase, by a small team.
Client details are anonymized. The demo uses a pseudonym and illustrative evidence.
The business
Twenty-one years of signed work.
A technology execution partner of 150+ people, remote-first since 2012. They deliver engineering, design, and growth work for organizations whose websites and digital estates are business-critical: universities, manufacturers, global consumer brands.
The site itself is the first proof. If you claim you build AI-native digital products, your website has to be one.
The buyers
Five people. One site. Each in their own register.
Strategic sponsor
Executive
Risk, cost control, and not funding a stalled program.
Executional owner
Tactician
Delivery dates, team burnout, and scope that keeps expanding.
Experience champion
Creative & Strategist
Design systems, governance, and editorial control.
Revenue driver
Demand Generator
Pipeline, discoverability, and conversion.
Measurement architect
Analyst
Data, attribution, and evidence.
The goal
Not more traffic. Qualified conversations.
A visitor states their situation. The site composes a dossier back to them. They leave their work email. The full journey (source, pages, what was composed, what they submitted) lands in the CRM and the team's inbox the same day.
- Cited by ChatGPT and Perplexity, not just ranked in Google.
- Marketing ships content without engineering.
- The site doubles as the delivery evidence library.
The build
Four layers, one codebase.
The site
Home, four lens pages (Strategy, Engineering, Design, Marketing), industries, case studies, and a thinking engine backed by a headless CMS. One modern stack: React, server functions, edge hosting.
The AI concierge
A persistent command bar. The visitor types their situation; an AI function reads the site, classifies intent, and recomposes the page in place: a headline, four lens-tagged reads, evidence from real engagements, and one next action. Every composition is recorded against a lead journey, so sales opens the call already knowing what was asked.
The connected systems
CRM for contacts and attribution, a database for lead journeys and audit runs, a CMS with an admin console for self-serve publishing, email delivery, LinkedIn conversion tracking, Slack alerts to the team, a consent layer gating all tracking, and a partner portal where vendors apply, get scored, and get onboarded.
The lead magnet
A seven-criteria AI Visibility Audit that checks any domain live, benchmarks it against a published dataset, and hands back a branded report on the page. Work email required. It runs the entire loop on its own.
The stack
Three tools. One builder.
Lovable
Build and ship
Product, database, and edge hosting in one place. The entire site and its systems were built and deployed here.
Claude
Direction and review
Specification, code direction, and review. Every layer was shaped and checked in conversation before it shipped.
OpenAI
Intelligence
The models behind the AI concierge and the Visibility Audit: reading the site, classifying intent, and scoring domains.
We ship on this stack ourselves. The person who scoped this sprint built it with the same three tools, in one codebase, with no front-end team and no handoffs.
The sprint
How it moved through our loop.
- 01 UnderstandMapped five buyers, their language, and the one conversion that matters: a qualified discovery conversation.
- 02 PrioritizeChose the concierge and the audit as the two moves with the most leverage. Everything else serves them.
- 03 BuildSpecified in conversation, written and reviewed by AI agents in one codebase, directed by a human. No traditional front-end team.
- 04 ValidateEvery composition, page, and submission is tracked from source to CRM, so evidence replaces opinion.
- 05 CompoundMarketing publishes without engineering, and each engagement becomes new evidence the concierge can cite.
"Everything you just saw, the site, the AI concierge, the audit, the CRM wiring, was specified in conversation and shipped in weeks. You are looking at the evidence."