MillionPlus took enquiries through its website, WhatsApp, Instagram, a phone line and email, and each one arrived with no knowledge of the others. A prospect who messaged on Instagram and then called was a stranger twice. The founder, carrying a 20-year network of relationships in his head, personally qualified every lead. That worked at a certain volume. The volume kept growing.

p0stman's answer was to rebuild the UK luxury asset finance broker as a multi-agent operation: one central AI orchestrator routing every channel, every interaction and every piece of content, with Attio CRM holding the canonical record. One person built it, in weeks.
Four bottlenecks, one root cause
p0stman's write-up names four problems. Read together, they share a shape.
- Enquiries arrived across web, WhatsApp, Instagram, voice and email in isolation, with no shared context between them.
- Two decades of relationships lived in one person's head, invisible to the wider team.
- Content production (articles, social snippets, video scripts, email campaigns) absorbed serious time every week.
- Lead qualification depended on the founder personally, which capped growth at his calendar.
Each problem is a missing shared layer. Context wasn't shared between channels, relationship knowledge wasn't shared with the team, and qualification wasn't shared with anyone. Patching any one of them alone would have left the other three intact.
High-touch firms hit a founder ceiling
Luxury finance is among the hardest kinds of business to automate because every interaction carries real weight and clients expect to be recognised. For years that meant the founder stayed in the loop on everything. The MillionPlus build tests whether an orchestrator with full shared context can keep that high-touch feel while taking the human out of the routing layer. If it holds, the same pattern applies to any relationship-driven firm hitting the founder ceiling.
One brain, many specialised agents
The architecture separates coordination from channel handling. A central AI brain owns the reasoning and the context. Around it sit specialised agents, one per channel: a web chat agent, a WhatsApp Business agent, a voice agent that works on both the site and the phone, Instagram DM routing, plus Slack, YouTube, email and an internal AI assistant for the team. Eight channels and counting.

A conversation that starts on WhatsApp and continues by phone is one conversation.
That's the whole point of the design. The agents don't hold their own memory; they hand everything to the orchestrator, which carries the thread wherever the client goes next.
Attio CRM as the layer every agent reads and writes
Shared context has to live somewhere, and p0stman put it in Attio CRM. Every enquiry, contact, deal and conversation thread lands there, and the sync runs in both directions: agents write what they learn and read what's already known before they respond. The founder's 20-year network moves from his head into a system the whole team and every agent can see.
The same brain writes the content and runs a 31,000-listing marketplace
The content engine draws on the same orchestrator. Articles, social snippets, email campaigns and video scripts generate on demand rather than eating a weekly chunk of founder time. Image and video generation is the next stage of the pipeline.
The public site grew into a luxury marketplace: 70+ pages and 31,000+ live luxury asset listings, synced daily. It ships with what p0stman calls the full agentic web stack, structured so AI search engines can find and cite it.

Shipped by one person, in weeks
The stack is conventional where it can be: Next.js 16, TypeScript and Tailwind v4 on the front end, Supabase for data, Vercel for hosting, Resend for email. The agent layer runs on Gemini 2.5, with Meta Cloud API for WhatsApp and Instagram, Twilio for voice, and the Slack and YouTube APIs for their respective channels. Attio sits at the centre.
The pieces are all off the shelf. The wiring is where the work went: a single orchestrator deciding what each channel agent should know and do, and a CRM that stays accurate because nothing bypasses it. The live site is at millionplus.com.
For any founder-bottlenecked firm, the practical version reads like this: pick the system of record first, route every channel through one reasoning layer that reads and writes to it, and only then add agents.