Five AI plays — and the one idea underneath them.
A working draft to pressure-test with James and Leonides. Not a finished pitch. The point is to decide three things together: does this make a real product set, is the vision useful, and can we workshop it into prototypes.
Five plays, one asset
Pull on any one of the five plays and you reach the same place. Each one needs a structured, queryable picture of the business — its customers, its money, its history, its documents. None of them works without it.
So they are not five products. They are five views of one asset: the company's living knowledge graph. Gold Digger Pro is simply the first view we've actually wired. That reframe is the whole point of this draft — we're not proposing five half-built tools, we're proposing to build one operating system for the business and prove it one application at a time.
Thought piece — the living knowledge graph of an SME is its operating system
A draft thought piece, synthesised from the Atrapos.ai brochure (Leonides / Fosferon Ltd) and the Gold Digger Pro build. For discussion, not publication.
Two things run a business: its data and its judgment. The data is what happened — every customer, transaction, document, decision. The judgment is the practical knowledge built over years that decides what to do about it: which opportunities to chase, which to decline, what the business will and won't stand behind. In a small firm both live in a few people's heads. As it grows — new markets, new languages, more decisions than any team can hold uniformly — both fragment. That fragmentation is Atrapos's framing of the problem, and it's the right one.
Most "AI projects" only attack the first half. Connect a language model to the company's documents and it produces fluent answers — and hallucinations, and commitments the business can't honour. Fluency isn't judgment. As Atrapos puts it, the hard part is replicating organisational judgment, not just producing text.
So picture two layers, joined:
- The graph — a living, queryable model of the business: customers tied to transactions, documents and history, kept current. This is the half Gold Digger Pro already touches: CRM joined to real spend, mapped to the actual buyer.
- The judgment — a layer above the language model that holds the firm's principles, boundaries and decision logic. Atrapos calls an agent built this way a Cognitive AI Agent: governed (humans define its boundaries and authority), domain-grounded (it knows what it knows and what it doesn't), structurally honest (it says it's AI and names its limits), producing a structured brief at the end of every interaction so a human still decides, and earning depth only through progressive trust.
The graph is the memory. The governed agents are the workforce. Together they're the operating system.
Put the two layers together and you have something that behaves less like a chatbot and more like an operating system for the business: the graph remembers and senses; the governed agents act on it — consistently, in any language, at any scale — and always hand a structured brief back to a person.
Once that OS exists, the five plays stop being five products. They're applications running on it. Reactivation (Gold Digger Pro) is the same machinery as Atrapos's renewal-and-follow-up agent. Qualifying and delivery (the Multiplier) is the professional-services qualifier. Pricing, forecasting and visibility are queries and agents on the same graph, under the same judgment.
That's the vision worth testing, and it's the honest one: don't sell five tools — build one operating system (a data graph plus governed agents) and prove it one application at a time. Gold Digger Pro is application one, and it's already running. Atrapos already builds the governed-agent layer. What's new, and what these sessions are for, is joining them into one living graph for a real SME.
The five plays, honestly
Simplified, and scored on where each really is today. "What it needs" is the honest distance to a working version.
| Play | What it does | Maturity | What it needs to be real |
|---|---|---|---|
| Gold Digger Pro | Reactivate the revenue sitting dead in the customer base | BUILT · 6.5/10 | Hardening, and the invoicing + ERP rungs of the data ladder |
| The Growth Model | Budget in → growth curve and valuation impact out | CONCEPT · 3/10 | The model logic and a working app. Not built today. |
| The Profit X-Ray | Which products, customers and channels actually make money | CONCEPT · 3/10 | The data join (CRM + invoicing + ERP) and a margin engine |
| The Delivery Multiplier | More output from the same team on one high-volume job | CONCEPT · 3/10 | An AI delivery layer, proven on ourselves first. The "3×" is a hope, not a measurement. |
| The Answer | Get the firm cited when buyers ask AI about the category | CONCEPT · 3/10 | Owned domains and a repeatable GEO method with measured results |
Leonides — building the living graph
Leonides is Atrapos.ai (Fosferon Ltd). He builds the judgment layer of the operating system described above — Cognitive AI Agents grounded in an organisation's real domain, governed by human-defined boundaries, structurally honest, and producing a structured brief at the close of every interaction. It is the same human-in-the-loop discipline Gold Digger Pro is built on, already productised. The collaboration is the missing half of this draft: Articulate brings the sales-and-marketing domain, the data graph and the go-to-market; Atrapos brings the governed agents and the deployment method.
Atrapos agents are built on five principles — governed behaviour, domain-grounded knowledge, structural honesty, structured output, progressive trust — and deployed through a consulting-led process (Intake → Knowledge → Build → Review → Go-Live). Both map cleanly onto the three working sessions below.
His agents already map to the plays
| Dare to Dream play | Atrapos agent it resembles |
|---|---|
| Gold Digger Pro (reactivation) | Outbound Renewal & Follow-Up Voice Agent |
| The Delivery Multiplier (qualify & deliver) | Professional Services Qualifier |
| The Answer (guided discovery) | Vertical Market Sales Agent |
| Growth Model & Profit X-Ray | Graph / analytics side — agents read, humans decide |
Two things to explore together
- His agents on the graph. Extraction, enrichment, drafting, monitoring — the workforce that runs on the operating system. Which of his existing agents map to which play, and what's missing?
- How he'd build the live graph. The data model, the store, identity resolution (tying every transaction to the right account, department and buyer), and crucially how it stays live as the business changes. His architecture, our domain.
Open questions for Leonides
- What's the right graph store and retrieval layer for an SME-scale knowledge graph?
- How do agents read from and write back to the graph without it drifting?
- What keeps the graph current — event-driven updates, scheduled re-syncs, both?
- Which of your agents could run a real task in session three?
- How do we structure jointly-owned IP on what we co-build?
The workshops
The way to make this real isn't a spec — it's a demonstration the customer can see. We run short workshops that each turn one idea into a working demo on the customer's own data: what they see first, how we build it, and only then the tech underneath.
Open the workshops page → — built outputs-first, for James to use in the room.
What we want from this
By the end of the first workshops: a shared vision, one or two working demonstrations on real data, and a clear split of who builds what and how it's owned. Then — and only then — we decide what becomes a product.
For James, two simple questions: which client gives us a real dataset to build the graph on, and which app is play two?
For Leonides: your read on the graph architecture, and which of your agents plug in first.