Codito Sunday

The Anthropic-certified AI agency serving company directors

Why read this issue

I spent months, working 70-hour weeks, learning to lead 15 AI agents. This issue gives you the shortcut: the working framework that came out of it, ready to put in place in your own business.

My account in Le Figaro on 13 October: read the article (subscribers only).

Hello everyone,

On Tuesday, Le Figaro published my account of what Americans call AI brain fry: a brain frazzled by AI. In it, I describe a spring Friday when, after months of running my AI agents in parallel, a violent headache forced me to take my first real weekend off in a long time.

Above all, that Friday says something about what my job is: exploring tomorrow's ways of working before anyone else. I live them first, run into their problems and look for solutions until they hold up, so that what remains in the end is a working framework rid of its pitfalls, which I then put in place for my clients.

Where this exploration cost me months, your teams can start straight from what works, and that is a golden shortcut. This week, Anthropic published figures showing just how much it will matter for every company.

I · Market Analysis

The age of agents begins: your attention becomes the company's foremost asset

The hourglass: agents produce without limit, while human attention lets the work through one file at a time.

The fact.  On 16 October, Anthropic's team of economists published a study on how work is changing within the company itself. In March 2026, the working hours of AI agents there overtook those of humans for the first time, and by August, the agents directed by each employee were clocking up 6 times as many hours as the employee (Anthropic, “How employment is changing at Anthropic”, 16/10/2026; Trending Topics and IBTimes, 17/10/2026).

The Codito analysis.  The same report shows where the limit now lies: outside engineering, reading everything the agents produce would take up more than half a working week, and the share of time spent steering them (writing their instructions, checking their work, having one agent review another) rose from 10% to 22% between autumn 2025 and spring 2026. A BCG study published in March in the Harvard Business Review measures the cost of ignoring this: 14% of users say they have already experienced brain fry, perceived productivity peaks at 3 tools and then declines, and those who feel they use AI too much report an error rate 39% higher (Axios and The Register, March 2026). The machine has therefore stopped being the limit, and the next competitive edge will go to companies that organise their teams' attention as carefully as their agents.

What it means for you.  Multiplying agents without redesigning the role of humans is like piling files onto a desk that never gets any bigger, and that is exactly the dilemma I described to Le Figaro: check everything at the cost of your sanity, or lower your guard at the cost of errors. There is a 3rd way, and it starts with a simple step this week: list the AI tools and agents each member of your staff actually uses, then, if there are more than 3, decide with them which ones they keep.

II · Codito Expertise

The golden shortcut: the working framework I have tested for you

15 threads, just 1 read: the agent office framework.

The laboratory.  Before offering a method to a client, I test it on Codito. After that spring Friday, I faced the 2 options described in the article, slowing down or lowering my guard, and I found a 3rd: organising my agents the way you organise an office, with a hierarchy, written rules and a daily review point. The overall picture I used to scribble on paper, like a pilot without an instrument panel, is now held by the system.

1.  An office manager. An agent I have named Napoléon gathers what the agent team has produced every evening and presents it to me in a single summary, so I read one report instead of 15 threads.

2.  Sentinels that stay silent. Every evening, Vidocq checks that the tasks have run and that the deliverables exist, and only writes to me if he finds an anomaly: silence becomes information.

3.  Rules read before acting. Colbert has every agent read our written procedures before any client operation, so I no longer have to re-explain the context, and my attention goes to decisions and clients.

In your business.  This framework, tested in-house, is what we put in place with your teams through the Codito AI Partnership. Each month, we build agents with them that connect to their software and train them on their real files, setting out from the start who reviews what, when and how. Your teams are spared months of trial and error, and the time saved by the machine genuinely comes back to them.

That is what Codito Ergo Sum, an Anthropic-certified partner agency, does: turning AI into a lasting gain, for your teams and your results alike.

Discover the Codito AI Partnership

III · Inside Codito

Codito joins the Hub France IA's 2026 AI Map

On Friday, I received an email from the Hub France IA: Codito Ergo Sum is now listed in the AI Map, the catalogue of AI solution providers for SMEs and mid-sized companies, alongside more than 1,000 start-ups and suppliers that shape the French ecosystem. Companies are selected on the basis of their European roots and verified client references, and Codito now carries the “Trusted AI Partner 2026” badge.

This listing matters, because AI also attracts providers who emerge from a few days' training and deliver without any technical foundation, exposing their clients to risk. For us, technical skill is a point of honour: everyone in our technical teams can write code by hand and understands what it does, because that is what it takes to deliver serious, secure work.

More than 1,000 companies listed in the Hub France IA's 2026 AI Map.

Until next Sunday. In the meantime, bear in mind what this week has shown: machines now produce more than we can read, and that is excellent news for the companies that learn to lead this abundance. I will keep exploring the ground first, so that each time you can take the shortest route to what AI has promised from the very beginning: time to think, create and decide.

“You made it to the end: which task will you hand over first to an agent that only speaks up when something is wrong? Reply in a word; I read every reply.”

Emile

President of Codito Ergo Sum

LinkedIn  ·  YouTube  ·  Instagram

Reply to Emile in a word  ·  Share on LinkedIn

You are receiving this email because you subscribed to the Codito Sunday newsletter.

Unsubscribe

Codito Ergo Sum · France