Good morning. I'm JB, and this is Familiar — Agent Native Compute. Over the next twenty minutes I'm going to make one argument: that the way technology reaches small businesses is broken, that agents change the economics of fixing that, and that I've already built the machine that does it. The structure is three flywheels on one axle: a conceptual flywheel — how the method works; an economic flywheel — how it makes money; a social flywheel — why keeping it open makes the business stronger, not weaker. Then Spain: the market, the wedge inside the wedge, the company structure, and — honestly — what I still have to prove. Let's start with the customer, not the technology.

Start with the customer. A twenty-person company doesn't lack software — it drowns in it. What it lacks is capability: the ability to make software work. Systems are fragmented; the important knowledge lives in the owner's head; integration is expensive; automation is brittle; and senior technical expertise is priced for enterprises. AI code generation doesn't fix this — it produces more software and leaves the operating problem exactly where it was. I spent years inside SAP-scale enterprise programmes, and the punchline of this whole plan is democratising what I learned there: the machinery a £25 million budget used to buy, delivered at SME prices. That is the constraint Familiar attacks.

Here's the insight everything else rests on: agents don't primarily need to be smarter. They need an environment in which their intelligence can compound. Seven things make that environment: persistent context, so memory outlives the session; coordination, so attention is decided rather than lucky; bounded authority, so an actor with your credentials has stated limits; accessible knowledge; observable work; executable proof; and reliable handover. Most of the industry is scaling model quality. I'm building the environment — because if it's only in the context window, it effectively didn't happen. Familiar is that environment, and the next slide shows why the environment, not the model, is where the business lives.

This loop is the business — everything else in the deck is a snapshot of it turning. Consulting work finds real problems. Solving them codifies a practice. The practice becomes open software. People use it and fork it, and the trust and capability that creates comes back as better consulting — which finds the next real problems. Every engagement makes the next engagement easier. Every piece of software makes the practice more legible. Every adoption makes the commons more valuable. Notice what this says about open source: it isn't a giveaway bolted on the end — it is a load-bearing stage of the loop. Next slide: the evidence that this flywheel is already turning.

And it is turning. This isn't a business-plan diagram — the estate already exists: fifty-plus repositories across seven capability regions, built since January, by one operator. More than sixty projects covering most of an enterprise engineering practice; over a hundred agents — familiars as continuous partners, residents standing guard on each project, burst agents for elastic labour — moving through four billion tokens a day. But hear the right number: the impressive figure isn't the tokens, it's that fifty-odd repositories are the *designed cost base* of one person, not a heroic feat. Every session starts from the accumulated record of every session before it. I'll show you crown jewels — deterministic simulation, file-native data with measured benchmarks — not as a catalogue, but as proof the layers exist.

The stack, in one breath. Carapace is the substrate: a production-grade Odoo base image — 17, 18, 19 — with backup, restore and upgrade as first-class operations, because deployment is where Odoo projects go wrong. odoo-assistant-mcp is the bridge: an MCP server giving an agent natural-language access to Odoo's five hundred–plus models — and, load-bearing for SMEs, SOP capture: workflows saved as reusable procedures in git. Same steps, every time; knowledge that doesn't walk out the door. Ninety-nine integration tests run against Odoo 14 through 19 in CI on every change. And a familiar is the agent driving it all. One lineage note I'm proud of: the MCP predates the practice — a working, tested product delivered *before* the method was codified. Old proof, new discipline, one stack.

So the commercial wedge: make existing businesses agent-native, starting with SME operations — ERP, finance, inventory, purchasing, documentation, customer operations. What Familiar sells is senior engineering capability plus agent labour on proven open infrastructure — not asking a twenty-person company to hire an AI team it can't afford, and not selling it another dashboard. The economics are the point. A traditional consultancy is a senior consultant plus juniors plus PMs plus manual documentation plus repeated discovery — high fixed cost, every time. Familiar is one senior operator plus agent capacity plus a reusable estate plus a compounding operational record — lower marginal delivery cost, every time. That's the innovation, and Carapace is where I prove it first.

Carapace in one sentence: don't replace the ERP — give the business an operator. Odoo provides the system; Familiar provides the agentic operating layer around it. Concretely: the familiar can read invoices, classify transactions, draft supplier orders, create draft journal entries. And here is where the open-source estate meets the product: submitting an actual payment? That requires a warrant — permission plus predicate plus mechanical enforcement. The Spanish SME doesn't just need an agent that can touch Odoo; it needs an agent it can safely permit to do increasingly consequential things. That's the difference between automation and an operator, and it's why the estate isn't separate from the commercial work — it's the safety architecture.

Why Spain, in numbers — and every figure has a source. SMEs are 99.8 percent of Spanish businesses, and 77 percent of firms have fewer than five employees against 71 percent across the EU — Bank of Spain and Eurostat. Labour productivity ran roughly 7 percent below the EU average in 2024 — OECD. AI in production: 8.7 percent of Spanish SMEs against 49.2 percent of large enterprises — DESI 2024. And the state is already paying to close that gap: Kit Consulting offers up to €24,000 per SME for digital and AI advisory; Kit Digital pays €3,000 to €29,000 by segment through Red.es. Tourism is 12.6 percent of GDP — INE. Contrast with where I sit today: German ICT vacancies fell 22 percent in 2025, and a visa status change takes twelve to fifteen months. Spain is where the money already points.

Inside Spain, the wedge sharpens to hospitality — the sector Spain cares about most. Tourism was 12.6 percent of GDP in 2024 — €200.7 billion, INE — and drove roughly 41 percent of that year's growth, per the WTTC. Hospitality alone is around 6 percent of GDP and 1.7 million direct jobs: exactly the fabric this plan serves — independent, family-run, seasonal, paying double-digit commissions to booking platforms for the privilege of staying visible. For restaurants: POS, inventory, suppliers, rotas. For hotels and guesthouses: the back office, integrating with whatever booking system they already run. And hear the offer correctly: I am not selling AI. I'm selling less administration, fewer disconnected systems, better operations — the agent is the mechanism, not the product category. MINTUR and SEGITTUR's Última Milla programme has €25 million behind precisely this sector.

The second flywheel is how this makes money. Services fund software. Software makes services more valuable. Open source creates trust. Trust creates adoption. Adoption grows the market for both. Same loop as slide four — seen through the ledger instead of the practice. And it reframes the key question. It isn't "how much software can Familiar sell?" It's "how much economic value can one senior operator plus an agentic workforce deliver?" — and the estate is the running evidence that the answer is well above a consultant's billable hours. The next slide makes the revenue lines concrete, and I'll be explicit about which parts are solid today and which are still owed.

Four routes, sequenced by what exists today. Consulting and advisory — now; the founder's seniority is the bankable line. Early access — near; the access machinery is built and tested today: one NDA, named grants, revocable in sixty seconds. Hosted and managed deployments — scale; the estate's infrastructure becoming someone else's production dependency. Patronage and grants — parallel, funding the commons specifically. The honest framing is "open implementation, paid operation": everything released is MIT, and the customer pays for implementation, integration, hosting, operation and expertise — so the customer can leave, their software and data remain, and the relationship earns renewal. And the gap, plainly: the projection — rates, utilisation, cohort sizes, the founding-capital figure — is still owed. Consulting is what underwrites the plan today.

Third flywheel: why opening it doesn't destroy the business. Four properties do the work. Legibility — the work is inspectable. Operability — ordinary organisations can actually run it; "adopt one property in an afternoon" is the release standard. Interchangeability — the Relay handover as the SLA: a new agent plus the state directory equals useful work. Institutionality — something that outlives its founder. If someone forks a component and spawns a company, the component had value beyond its author: that is validation, not loss. Take the map, leave the institution. What cannot be forked: the operational record — forty-eight recorded sessions in threshold alone — the decision corpus, and the consulting practice where the seniority monetises. And the honesty: institutionality is the one property still untested. The enterprise form on this slide is the test.

Two ways to fund the mission. Option A, founder-funded: JLDB capitalises the operating company and the development programme — maximum control, fastest execution, simpler governance, a larger initial capital requirement, and the financial risk sits with me. Option B, co-funded: a strategic partner brings capital, customers, local market access, sector expertise, institutional credibility, training capacity — JLDB brings the technology, the operating practice, the estate and the senior engineering capability. Deliberately, the question is not "who pays the bill." It's "which combination produces the strongest flywheel." A co-funder with market access may accelerate the loop far more than money alone — and that trade against control is exactly the conversation I'm here to have.

The structure is two entities. JLDB Ltd — the existing UK company — is the founding and funding vehicle: group IP, stewardship of the category, consulting outside the EU. The Spanish entity, a standard S.L. at formation, is the EU vehicle: Spanish tax domicile, local hiring, SME engagements, and access to Spanish and EU programmes a post-Brexit UK parent cannot touch directly — including becoming an accredited agente digitalizador, so the market's own subsidy becomes the revenue channel. The mission doesn't need a special legal form to be real: MIT on everything released, open description, narrow grants — that's the constitution, and both entities inherit the documentary governance on day one. Commercial close to the market; commons preserved.

Now the slide that earns the deck its credibility. The next phase is economic, not architectural. Five things need proving. Market: who buys first, how large is the reachable market, which channels actually work. Economics: what a customer costs to acquire, what an engagement produces, the gross margin of agent-assisted delivery. Productisation: which consulting patterns become repeatable products. Scale: how one operator becomes a team, and how much revenue each trained operator unlocks. Funding: what level of initial capital gets the flywheel turning. I want to be precise about the numbers in this deck: the market statistics are sourced; the business figures are the research programme, not claims. Every headline here becomes a section; every assertion gets a receipt.

So, the thesis. Familiar is building a new kind of technology company. Not an AI agency. Not an ERP vendor. Not an open-source foundation. Not a consultancy with a chatbot bolted on. A company where senior human capability is amplified by persistent agentic systems and continuously codified into an open, reusable commons. Most companies go market, then product, then build. I came at it backwards — production system, operational estate, proof of production, reusable infrastructure, then the commercial application — which is riskier, but it produced the one thing a conventional plan cannot fake: evidence that the underlying production model already works. The moat isn't "nobody can copy the code." The code is the residue of a production system still running and compounding.

Four lines to leave with. Start with the problems businesses already have — less administration, not more AI. Use agents to make expert capability scalable — one senior operator plus an agentic workforce. Turn every engagement into infrastructure for the next — that's the flywheel, and it's already turning across fifty-plus repositories. Keep the resulting knowledge open — because trust is the distribution channel, and the commons is the constitution. Familiar: Agent Native Compute. The immediate next step is the economic research slide sixteen commits to — rates, utilisation, cohort sizes, founding capital — and I'd like you in it. Thank you. —-