Five days inside your company. 3 to 5 real processes from different areas become agents that run with human approval, access policy and a test suite that proves they are working — with your IT and legal teams in the room from day one.
I want the executive diagnosis → Starts with half a day with leadership. No hiring commitment.Your company probably already has AI. Distributed licenses, a pilot that impressed in a presentation, teams using it on their own. And yet, three questions remain unanswered:
No one can answer with a number. The assessment is “the team liked it” — which doesn’t survive a committee question.
It hasn’t been tested against attack. No one knows what the agent could do if fooled by a received document.
The enthusiast who built it moved to another department. There’s no routine, no owner, and it rots in silence.
None of the three is a model or vendor problem. They are problems of method and governance — and that’s exactly what this program installs in your team.
Foundations in the morning, hands‑on in the afternoon, on the your processes and the your systems. No one starts from an empty folder: the team starts from a kit that already works and evolves it for the company — which is the path with roughly double the success rate.
Leadership workshop: area maturity, process inventory with A/B/C rating, and prioritization of 3 to 5 pilots. Leaves with an executive report and proposal — and serves as the investment decision even if the program doesn’t happen.
Executive report + prioritized pilot portfolioAnatomy of an agent, division of labor between person and agent, and pilot selection criteria. In the first hour the team already sees it working; by the end of the day each area has its first own skill written.
Chosen process + 1 skill per areaLayers, narrow‑scope tools, MCP, orchestration and proactivity. Each team connects a real company system — via IT, with minimal scope — and runs the end‑to‑end process.
1 real integration per area + demonstrable flowCorporate identity, real RBAC/ABAC, data classification and LGPD with legal present. Live attack demonstration, and teams attack each other’s agents. Every failure becomes a fix and a test.
Permission matrix + tested threat model + audit trailThe agent test suite, the root‑cause table (where to fix without inflating the prompt), cost per task and the weekly operation routine. It’s the day no competitor teaches — and what decides whether the project lives.
Eval suite with scoreboard + business case + defined routineMorning of assisted implementation with real data. In the afternoon, each area presents to the panel and sponsor: problem, before/after, live demo, governance, scoreboard and business case. Closes with a 30‑day commitment, with owner and date.
Agents in assisted operation + signed commitmentsOnly the minimum the task requires, and nothing more. The program teaches narrow‑scope tool — “query request X”, never “database access”. The rule that underpins this: a tool that doesn’t exist can’t be fooled. The role‑based permission matrix is filled on Day 3, by your team, with your IT together.
It will happen, and there is no complete defense against it — anyone promising otherwise is selling something. That’s why the design doesn’t rely on the agent resisting: it relies on it not having power to cause damage on its own. On Day 3 your team sees the attack succeed, sees the defense hold, and leaves with the threat model tested and documented.
Classification in four levels, identity‑based access policy, and a simple operational principle: less data flowing means less risk. The block is led with your legal team in the room — the policy that comes out of it is yours, not a generic template.
Irreversible actions — pay, delete, sign, publish — are never executed by an agent, in any configuration. Everything that leaves the company goes through identified human approval and is logged. The incident response plan fits on one page and is written on Day 3.
What your team builds are files in open formats, today governed by a neutral foundation with competing companies together. Switching models is editing a configuration file and running the test suite to validate. The program is design‑agnostic — not because it’s fancy, but because model deprecation is routine.
It’s not another course: it’s the same program tailored to the size, stack, and compliance requirements of your company.
| Scale | Open cohort | In‑company |
|---|---|---|
| Before | online diagnosis | Executive diagnosis half‑day with leadership |
| Processes | 1 per participant | 3 to 5, from different areas, prioritized in the diagnosis |
| Governance | simple roles (owner/team/customer) | corporate identity, RBAC/ABAC, LGPD with legal, formal audit |
| Stack | SMB tools | integration with ERP/CRM via IT, in a security‑approved sandbox |
| In the room | who does | who does + executive sponsor (required on Day 1 and on Demo Day) |
| After | community and mentorship | 90‑day assisted deployment, with Agent Ops accompanied |
| Investment | per seat | per project |
We surveyed the available programs — national platforms, postgraduate courses, the AI vendors’ own academies, and certifications. None teach evaluation and continuous improvement as a curriculum, and none bridge the technical and the operational (feedback, applied governance, organizational change).
A suite of test cases that can run at any time and reports, with a rating, whether the agent is good — covering the common path, edge cases, security, permission and escalation. It turns “I think it’s working” into a defensible report.
Thirty minutes per week with a fixed agenda: what it did, how much it cost, what it got wrong, what improves this week. It’s cheap and it’s what separates those who reap results from those who quit in the third month.
The access policy, the approvals traffic light, and the threat model are completed with your company’s reality, tested against real attacks during the program.
Open formats and centrally managed model configuration. Your team isn’t locked into a product — and knows how to validate a switch with testing, not faith.
• AI already exists in the company, but no one can prove results
• there are repetitive text processes in more than one area
• security and legal need to be comfortable before scaling
• there is an executive sponsor willing to be in the room
• the goal is internal capability, not outsourcing the build
• the expectation is to reduce staff — that's not what the program delivers
• no one can dedicate 5 days of agenda from the areas
• the decision is to buy an off‑the‑shelf product instead of building a team
• there is no sponsor — without one the project dies at the first competing priority
• results are expected without redesigning any process
The investment is per project and depends on the number of areas, the size and the integration scope. The path is the half‑day executive diagnosis: it defines how many pilots make sense and the proposal is built from it — with the prioritized portfolio in hand, not in the dark.
No. The program runs in a sandbox approved by your security team, and the connection order is always the same: read from file, read from system, write, and only then send. If IT does not release anything during the week, the teams work with exports and anonymized data — the infrastructure is ready to be connected later.
No. The program does not sell a platform: it teaches the method and the open formats that run on top of the one you already have. If your company has standardized a tool, it becomes the course environment.
The person who runs the process (who has the knowledge that becomes a skill), someone from IT for the integrations, someone from legal or compliance on Day 3, and the executive sponsor on Day 1 and on Demo Day. The sponsor is the least negotiable requirement: the difference between companies that reap results and those that don’t is organizational.
That’s the point of Day 4 and the 90‑day follow‑up. Each area gets an appointed owner, a weekly routine scheduled with day and time, a test suite running and a 30/60/90 plan. At the checkpoints the diagnosis is reapplied and compared with the initial one — maturity evolution is measured, not felt.
From 12 to 30, organized in area‑based teams — each team works on one process. Beyond that we split into cohorts to keep the mentor‑to‑participant ratio, which is what guarantees everyone finishes the week with something running.
It’s not what the program delivers, and the data do not support that promise: the most public case of replacing support with AI was reversed by the company itself due to quality drop. What changes first is the task — the work that falls into the gap between people.
Yes. The common path is to start with the executive diagnosis and, if it makes sense, run with two areas before scaling up. It’s also possible to send two or three people to an open cohort as an internal pilot, before deciding on an in‑company rollout.
Half a day with your leadership. We leave with a maturity map of the areas, a process inventory graded A/B/C and the 3 to 5 prioritized pilots — a document that serves as an investment decision even if you don’t hire the program.
30‑minute conversation to understand the context and see if it makes sense.
Half‑day executive diagnosis with leadership. The report and the pilot portfolio are delivered.
Proposal based on what the diagnosis showed — number of areas, scope and schedule.
I want the executive diagnosis →
Or, if you prefer to understand the method first: take the 5‑minute self‑diagnosis · see the open‑cohort version