Waimia.
FR / EN Book an audit →
Vol. 03 · № 01 Paris / Genève Bilingual EN · FR ◉ Booking Q3 2026

Revenue operations,
rebuilt with AI —
without the theatre.

§ 01 · Lede

Waimia is a Claude-native AI agency for companies whose revenue engine cannot afford a rewrite. We deploy strategy, operations, and our own platform — VirtuoseOS — as one continuous layer.

Book 45 min → Meet VirtuoseOS
§ 02 · Latest shipments
04.2026 CASE Halcyon Health — AI triage routing in 6 weeks. ↗ 04.2026 RELEASE VirtuoseOS 2.4 — multi-workspace kernels, French locale. ↗ 03.2026 FIELD NOTE HubSpot + Claude: the unglamorous pattern that actually ships. ↗
¹ All figures are re-measurable under NDA. ² Claude Opus 4.7 · Sonnet 4.6 · Haiku 4.5 · open-source fallback. ³ French typographic rules observed: espaces insécables and « guillemets ».
№ 04
120+
enterprise agents shipped
№ 03
6 wk
median time-to-production
№ 02
€14M
pipeline recovered · 2025
№ 01
4.2×
avg. ROI year 1
Ayla · Veriform · Northbound · Plateau · Halcyon · Opalfield · Stratoclay · Meridian&Fils · Vireo · Caserne · Ouest.io · Nimbra · Kilnwright · Hestia Labs · Obsidienne · Brume · Perla Finance · Verger · Ayla · Veriform · Northbound · Plateau · Halcyon · Opalfield · Stratoclay · Meridian&Fils · Vireo · Caserne · Ouest.io · Nimbra · Kilnwright · Hestia Labs · Obsidienne · Brume · Perla Finance · Verger · Ayla · Veriform · Northbound · Plateau · Halcyon · Opalfield · Stratoclay · Meridian&Fils · Vireo · Caserne · Ouest.io · Nimbra · Kilnwright · Hestia Labs · Obsidienne · Brume · Perla Finance · Verger
FIG 1.1

The average RevOps stack holds 43 tools. Fewer than 9 touch the same customer record consistently.

FIG 1.2

A generative model without a clean warehouse is a very expensive guess.

FIG 1.3

Agents shipped under 6 weeks succeed four times as often as those shipped under 6 months.

FIG 1.4

The replatforming instinct is almost always wrong. The question is rarely which stack. It is which handoff.

IVAct IV — How we ship

Four steps. No intermediate slide deck.

Each step delivers something concrete. You only pay the next one if the previous one passed its audit.

i ii iii iv
I

Diagnostic

We map your SOPs, identify repetitive workflows, quantify impact potential. Output: a process map + a priority.

II

MVP Sprint · 4 weeks

One agent, one job. We ship a working POC in four weeks with full audit trail and before/after metrics.

III

Production rollout

Team adoption, onboarding scripts, observability, governance. The agent joins the runtime — managed Claude or self-hosted based on your constraints.

IV

Maintenance & extension

SRE 24/7, model upgrades (Opus 4.7 → 4.8…), new use cases, ongoing AI Act compliance. The system learns; you keep control.

FIG II · 2.0 · Act II — The method
6 SERVICES · ONE CONTINUOUS LAYER

Six services, taught by doing — not by slides.

FIG 2.01

AI strategy

Six-week diagnostic that separates theatre from leverage. Ships an ordered backlog, not a slide deck.

→
FIG 2.02

RevOps rebuild

Funnel, handoffs, attribution. We touch the CRM ourselves.

→
FIG 2.03

HubSpot operations

Certified partner work — data model, automations, custom objects, GDPR-clean.

→
FIG 2.04

Data foundation

Warehouse, semantic layer, lineage. The layer every AI claim sits on.

→
FIG 2.05

AI automation

Agents that do one job well, with audit trails humans can read.

→
FIG 2.06

Platform engineering

We build custom tooling when the market doesn’t. See VirtuoseOS.

→
IIAct II — The engagement pyramid

Four tiers, one continuous architecture.

From process diagnostic to sovereign infrastructure — each tier sets up the next. You climb when you're ready.

I CONSULTING

Audit & Strategy

Map the SOPs before you buy any model.

  • Process Intelligence
  • AI Act · RGPD
  • Hybrid architecture
Open tier → II LIGHT

Claude at every desk

Cowork · Skills · MCP · Routines. Adoption over deployment.

  • Claude Cowork rollout
  • Custom Skills
  • MCP connectors
  • Routines & plugins
Open tier → III ORCHESTRATION

Autonomous agents & MVPs

One job, one agent, one audit trail humans can read.

  • Business agents
  • Multi-agent (LangGraph/CrewAI)
  • Tiered routing
  • 4-wk MVP sprint
Open tier → IV ENTERPRISE

Scale & sovereignty

Anthropic-managed async · or fully self-hosted on your hardware.

  • Claude Managed Agents
  • Self-hosted (vLLM, Ollama)
  • Open-source models
  • SRE · 24/7
Open tier →
§ Manifesto · Viewport 01

We are suspicious of AI theatre. We distrust the purple-glass aesthetic, the brain-circuit illustration, the bento grid of empty affordances. We write French, we read Stripe docs, we ship code on Tuesdays, and we measure pipeline on Fridays.

FIG III · 3.0 · Act III — The platform
VIRTUOSEOS · v2.4

The operating layer we ship under.

VirtuoseOS is the kernel that sits under every Waimia engagement — workspaces, agents, auditable command trails, a French-first UI.

See product tour → Read changelog
virtuoseos · workspace/plateau
[kernel] multi-workspace/init ok
[agent:revops] 14 handoffs routed · 0 dropped
[hubspot] 482 contacts synced ◉
[audit] trail open · last 12 min
[locale] fr-FR · typographic rules on
$ virtuose ship pipeline.agent --workspace=plateau
Kernel
Multi-workspace
Agents
Audited, reversible
Data
CRM-native + warehouse
Locale
FR-first · EN parity
IIIAct III — Department solutions

Six departments. Six fields. One method.

Each department has its own constraints — regulation, data, internal culture. Our method stays the same: process, agent, audit, measure.

Reconciliation · memos · Excel

Finance

Open department → Lead scoring · content · Design

Sales & Marketing

Open department → Triage · resolution agents

Customer Support

Open department → Contract review · Word

Legal

Open department → Onboarding · CV · SOPs

HR

Open department → Legacy · Claude Code

IT & Engineering

Open department →
FIG IV · 4.0 · Act IV — The proof
4 CASE FILES · DATA BLOCKS

Four files. Sector · duration · stack · impact.

CASE 01 Plateau Sector SaaS B2B Stack Claude · HubSpot · dbt Impact · 10 wk +€2.4M pipeline recovered ↗ CASE 02 Halcyon Health Sector Healthtech Stack Claude · Snowflake Impact · 6 wk −38% triage time ↗ CASE 03 Northbound Sector Fintech · EU Stack Claude · Segment · dbt Impact · 14 wk ×3.1 SQL conversion ↗ CASE 04 Caserne Sector Industry · FR Stack Claude · custom RAG · SAP Impact · 8 wk 1 200 h/year reclaimed ↗
All case files →
04.2026 CASE Halcyon Health — AI triage routing in 6 weeks. ↗ 04.2026 RELEASE VirtuoseOS 2.4 — multi-workspace kernels, French locale. ↗ 03.2026 FIELD NOTE HubSpot + Claude: the unglamorous pattern that actually ships. ↗ 03.2026 ESSAY Against the brain-circuit aesthetic. ↗ 02.2026 CASE Plateau SaaS — RevOps rebuild, €2.4M pipeline recovered. ↗
VIIAct VII — Offices

Two cities. One time zone. Two languages.

Francophone presence Paris–Geneva — same team, sovereignty logic when data must stay in Switzerland.

Paris
Headquarters · creative
15 rue de la Paix 75002 Paris France
Genève
Sovereignty · finance · pharma
8 rue du Rhône 1204 Geneva Switzerland
§ FAQ

Questions we get asked

Seven honest answers — the ones we send by email before the first call.

What is a 'Claude-native AI agency' and how is it different from a generalist AI agency?
We focus 100% of our R&D on Anthropic's ecosystem — Opus 4.7, Sonnet 4.6, Haiku 4.5, Skills, MCP, Managed Agents — and on compatible open-source frameworks. It shows in time-to-production: 6 weeks median for an agent in prod, vs 14+ for generalist teams. We write French, read Stripe docs, ship code on Tuesdays, and measure pipeline on Fridays.
What is the difference between Claude Skills and MCP?
A Claude Skill is a defined behaviour — a precise workflow (e.g. 'generate a finance memo') with its steps, constraints, output formats. MCP (Model Context Protocol) is the open standard connecting a model to data sources and external tools (Notion, Drive, Slack, HubSpot). Skills = reusable behaviour. MCP = standardised connectivity. They complement each other in every Waimia agent.
Can you deploy self-hosted, with no Anthropic dependency?
Yes. We deploy open-source models (Llama 4, DeepSeek V3.2, Qwen 3.5, GLM-5) on customer infrastructure via vLLM or Ollama, orchestrated with LangGraph or CrewAI. This path is recommended when GDPR, ACPR, SecNumCloud or HIPAA constraints require data to never leave your perimeter. We always document total cost and the migration path back to Claude managed if relevant.
How much does a first engagement cost?
A Process Intelligence Audit starts at €12,000 over 5 days. A 4-week MVP Sprint (1 agent in production, full audit trail, before/after metrics) starts at €24,000. For a precise quote, use our TCO calculator or book a 45-minute call — we come back with one diagnosed problem, or we don't follow up.
Are you a certified Anthropic partner?
Our Anthropic partnership application is under review (Q3 2026). Our architects are Claude Certified and we are Francenum activators (AI experts recognised by the French government). All numbers and certifications are public on our Trust Center.
How do you ensure AI Act and GDPR compliance?
Every agent we ship includes by default: signed logs of all actions (AI Act Art. 12), CSV/JSON export, rollback capability, automatic risk-level classification. For GDPR/LPD compliance, we run a PIA (Privacy Impact Assessment) before the MVP sprint. Our Trust Center details the full list of controls.
What is VirtuoseOS and do I have to buy it to work with you?
VirtuoseOS is our internal platform — the multi-workspace kernel under every Waimia engagement. You don't have to buy it to work with us. For Enterprise engagements, we offer it self-hosted on your infrastructure or as managed SaaS with French parity.
FIG V · 5.0 · Act V — Act

Book a working session.
Not a pitch.

45 minutes. We bring one diagnosed problem back, or we don't follow up. Either is fine.

Book 45 min → hello@waimia.fr
Waimia.