The method
I start from what you and your teams actually do every week. I hand everything that can run without you to agents, put the controls in place, then hand over. It is the method I first applied to my own job as a CEO.
Five steps, always in the same order. The price and the gain target are agreed before we start, and everything we build stays with you.
I go through your week and your teams’ week: every task, how often it happens, how long it takes, the data it uses. We put a number on what can run without you.
Deliverables
An agent is only as good as what it reads. I connect your sources: core software, banks, email, calendar, CRM. Read-only at first, with access you control.
Deliverables
One task chosen together, one agent in production within 14 days, used by your teams on a real workflow. It starts in observation mode: it proposes, you compare, then it goes live.
Deliverables
We rebuild one key function around agents: finance, steering or sales. Every agent is watched by controls. Only exceptions reach you, and the decision stays yours.
Deliverables
I document every agent, train your teams and hand over. Nothing is resold: the code, the instructions and the data stay with you.
Deliverables
Guaranteed return
Every engagement starts with a quantified gain target, agreed together and written into the proposal. It is measured on your own data, before and after.
Hours given back, costs avoided or extra revenue: we agree on the gain we aim for, and it goes into the proposal.
We measure on your own data, at the start and once in production. Tracking stays on a dashboard you keep.
If the target is not met by the agreed date, I keep working at my own cost until it is.
The test
Where does your company stand?
10 questions, 3 minutes, an instant result: your level and the 3 tasks to hand over first. Nothing is sent or stored.
Every one of my agents takes one of these three forms. The right choice depends on the task, not on the technology.
Someone triggers it with a click or a message. The agent does the work in seconds and hands back.
Example
Echo drafts the reply to a prospect, with open slots, when the salesperson clicks the button.
Best when
The task comes up often but needs a human look before it goes out.
It runs on its own, at a set time or every few minutes, day and night. It only disturbs you when it finds an exception.
Example
Every night, 4 diagnostics run about 190 checks on the business. In the morning, only the exceptions surface.
Best when
The work is regular, the rules are clear and the volume is more than one person can follow.
Several agents pass the work along: one spots, another drafts, a third checks. You sign off on what commits the company.
Example
Sorter tags a prospect’s reply, Echo offers time slots, Reader and Listener spot the agreed meeting, Verifier and Judge reject fake meetings.
Best when
The process runs across several teams or several tools.
The tools that run my agents every day. For you, every account is opened in your company’s name.
AI and agents
Claude Code
Builds the agents and runs them
Claude Opus, Sonnet, Haiku
The models, picked by how hard the task is
Voice models
Meetings run, calls, interviews
Data and memory
Supabase
Database, functions and app access
MySQL
The product’s business data, read by the agents
Google Workspace
Email, calendars and documents
Infrastructure
A Linux server
Runs the scheduled agents, day and night
Cloudflare Pages
Hosts the apps and dashboards
Discord
The channel between agents and the team, about 130 messages per working day
On my own CEO job, before any client.
No commitment. You walk me through your week, I tell you what AI can take on, with a first estimate of the gain.
Or email me: thibault@vivasync.io