01 · Monitor
It watches the situation. On a schedule.
Slack, your tools, the outside world — plugged in.
SchedulesConnectionsChannels
PMIA.agency · Agent systems for operators with a P&L
Agents that movethe bottom line.
Your team celebrates an AI win. The CFO's spreadsheet doesn't move. I build agents tied to what you measure — revenue, cost, risk — not a vanity metric.
Scroll
The Thesis
Most companies don't need another AI demo. They need systems that watch the world, form a view, and tell you what to do before it moves.
I · The world
Events, signals, and assumptions — scattered.
II · A brief
Clustered into a model you can read.
III · A decision
One call. Defensible.
Why me02 / The dots
Regulation, geopolitics, AI —it hits your P&L before your plan does.
Agents are the latest shock — faster than the last one.
You need someone who reads what's happening outside — and still watches the number on the page.
Engineer's discipline. A financier's eye for ROI. A decade in geopolitics. Years shipping product inside big companies. One question: where do agents actually pay?
The record
- 01
Engineer
CentraleSupélec (Télécom)
- 02
Corporate finance
ICCF @ HEC Paris — International Certificate in Corporate Finance
- 03
A decade in global governance
Board Member & Chief Marketing Officer, Institut Open Diplomacy
- 04
Enterprise product & digital transformation
France's #2 health insurer · a 350-year-old building-materials & glass leader · AI / hardware startups
- 05
Hundreds of iterations → a shipped simulation engine & agentic systems
From scratch — Bestie
What we do03 / Method
I build agent systemson one shared picture of the business.
Palantir built one vocabulary for the whole company — an ontology anyone could act on. I do that for teams that aren't Palantir-sized.
02 · Reason
It reasons from your principles — not generic AI takes.
A role, playbooks for how you decide, specialists it can call.
InstructionsSkillsSubagents
03 · Deliver
It acts — and you can see it on the metrics you care about.
Safe sandbox. One world model. Your company metrics.
ToolsSandboxWorld model
This is how Bestie runs today — and how I'll build yours.
The Show04 / Curriculum
I'm publishing the work as afour-season curriculum.
Interleaved, not linear.
Everything I learn building and deploying intelligence — in public, for free. Upvote what you want first; I publish to demand.
- The AI stack behind a real agent — and one gateway to rule the billsoon
- 1 billion tokens a month, under $100soon
- Why I built an agent framework instead of buying onesoon
- The day I thought I needed Kubernetes & GraphQL (I didn't)soon
- Why agent memory is still unsolvedsoon
- A workflow that pays for itselfsoon
- Why most corporate AI quietly diessoon
- Hire vs build an agent — the honest decision treesoon
- Who actually makes money with AIsoon
- Who underwrites an agent's execution risksoon
- What a world model really issoon
- From knowledge graph to assumption graphsoon
- How to price an eventsoon
- Claim, evidence, warrant, rebuttal — why your AI judge needs principlessoon
- Why you need a simulation enginesoon
- Sovereign AI and Europe's scramblesoon
- The supply chain of a single chatbot querysoon
- Private credit × AIsoon
- Why information is worthless unless it's earlysoon
- Why data is so hard to monetizesoon
Work with me05 / The journey
Four stages, fromaware to deployed.
Wherever you are, there's a clear next step.
01Awareness
Train
Your team gets fluent — they know what to use, when, and for which job.
Claude, Codex, Cursor — as a developer or a knowledge worker. I only teach what I do: I use these tools every single day to build Bestie and PMIA. No secondhand theory.
- Use-case first
- From a daily practitioner
- Custom format for your team
Shape your team's training
Tell me your role, your team, and what you want to get done. I'll build the format around it.
02Considering action
Consult
AI projects that show up in the P&L — CFO-ready, not pilots that quietly die.
This is the job I did for years: product manager in health insurance, transforming customer experience, data and digital products while driving measurable health and financial impact. Project management, scoping, opportunity mapping — done by someone who has shipped it inside a regulated enterprise.
- Opportunity mapping
- Scoped to ROI
- Enterprise-tested
03Decide to move
Coach
You want to deploy an agent without the learning curve — so you learn from my mistakes, not yours.
The track record talks: I built and deployed Bestie — a real multi-agent, multi-round system — and PMIA, hands-on. You skip the dead ends I already walked into.
- Learn from real mistakes
- Skip the dead ends
- Hands-on guidance
Don't leave with a clear, confident path? You don't pay.
04Move
Deploy
You're ready to build. Let's put a real agent to work.
Two ways in — start today with the flagship, or build something bespoke to your stack.
- Managed or custom
- Verifiable today
- Your stack
Today · Managed
Run it today with Bestie — fully managed or customized. See the simulation engine orchestrate agents across rounds, and the AI worker live in your Slack.
Custom · Bespoke
Specific requirements? I build agents and workflows around them — the same kind of systems I run internally to operate PMIA. See what I build.
Agents we build06 / Out of the box
The same toolkit I runinternally — built for you.
These aren't slideware. They run Bestie and PMIA today — wired to your stack when you hire me to build them.
Intelligence & Decisions
Watch the world, build the model, stress-test the call — the Bestie engine on your business.
- Monitoring agent — markets, news, X, YouTube, filings
- World-model builder — your assumption & ontology graph
- Stress-test agent — multi-round PRO/CON on any thesis
- Synthesizer — debate → one defensible, CFO-ready brief
- Research & report agent — analyst-grade, on demand
Outcomes
- See external shocks before they hit the quarter
- One model the team can argue from
Revenue & Growth
Find the right people, qualify them against your real ICP, and keep the pipeline moving.
- Lead agent — sourced + enriched + graph-qualified
- Outreach agent — personalized sequences that book calls
- ICP intelligence — who to talk to, and why now
- Competitor & market monitoring digests
Outcomes
- Pipeline without another SDR hire
- Outreach that sounds like you wrote it
Content & Operations
One operator shipping at team volume — posts, video, triage.
- Publishing agent — platform-ready posts in your voice
- Motion-design video agent — branded demos, no editor
- Triage & production agent — classify, route, ship
- Document & workflow agents across your stack
Outcomes
- Steady output without a content team
- Humans only see what needs a human
Have something specific in mind? That's the custom build — let's scope it.
Who it's for07 / Fit
I keep engagements where oneoperator can deliver excellence.
I don't have a big bench. That's the point — and it makes the fit easy to call.
A fit
- Founders who want agents that move revenue, not demos
- Operators carrying a P&L who need AI to show up on the number
- Product teams ready to ship agentic systems, not slideware
- Real-economy SMB and mid-market reacting to a changing world
Not a fit
- Very large enterprises needing a 600-person delivery bench
- Teams shopping for the cheapest prompt course
- Anyone who wants a logo on a deck, not impact on the bottom line
The Flagship08 / Bestie
Bestie — the flagship
An external intelligence deskthat watches your world
and briefs you before it moves.
Objections09 / Honest answers
The hard questions —answered on camera.
Every answer here becomes a short film — objections are fair game.
One agent optimizes its task. Another bottleneck pops up. The team celebrates; finance sees nothing. Tie agents to one shared model and the metrics you report — then a win shows up on the number.
Video — The local-optimization paradox
These systems guess. We're used to software that doesn't. Anthropic, OpenAI, Amazon, GitHub have all had incidents — the question is how you scope and recover. Bestie stays narrow: a few jobs, done well.
Video — Who should underwrite the execution risk — the provider or the user?
Start with Bestie if you want value this quarter. Build custom when only your stack and your rules will do. I'll tell you which — I make money on both, so I have no reason to oversell the build.
Video — Hire vs build — the honest decision tree
We pick a measurable outcome before we build. You should see it in the first deployment — not after a year-long program with nothing to show.
Video — Scoping AI projects the CFO will sign off