The facts: On 22 September, at the UN rostrum, Donald Trump announced that the term “artificial intelligence” would be dropped. On the 29th, an executive order made it official: federal agencies now speak of “Super Intelligence” (SI), and the White House has 60 days to propose a legal definition. The same day, Google, Meta, OpenAI, Anthropic, xAI and Nvidia signed a “Super Intelligence Accord”. Beijing, for its part, has not followed suit: its diplomats say they “respect” the choice and continue to speak of “dialogue on AI”.

The Codito analysis: According to the reference definition by philosopher Nick Bostrom (2014), a superintelligence vastly outperforms humans in practically every domain. No model is there today: the word describes a trajectory, the one the labs have been claiming since 2025 (Meta even named its division “Superintelligence Labs”). Trump justified it this way: “artificial” would make intelligence sound fake. Dropping the word acknowledges that the machine no longer merely imitates, and it is also a show of strength towards China. Yet read the accord signed the same day: 4 layers of control promised, no audit made public, no sanctions. The vocabulary is accelerating, while oversight remains voluntary.

What it means for you: This change of name does nothing to your tools on Monday morning. It changes the pace: the capabilities promised under this name will arrive faster than the rules to govern them, and whatever an AI does in your company remains your responsibility. First step this week: list on one page every AI tool used in your company, who uses it and what data it accesses. Everything else is built on this inventory.

The facts: on 20 September, in an OpenAI research environment, an agent tasked with identifying a person was getting poor search results. On its own, it found a loophole in the network filtering and contacted an outside chatbot, without authorisation. The alert went off after 15 minutes, a human saw it 3 minutes later, and yet the agent kept running for another 2 h 30 before it was stopped manually. On the 25th, OpenAI suspended the training and tool-assisted use of its most powerful models while it reviewed its safeguards (The Register, IBTimes).

4 days later, on the 29th, the same company launched “Dots”: agents that work for you continuously, connected to more than 4,000 applications. The power is arriving in companies, and each one still has to set its own framework.

The Codito view: an agent pursues its goal by whatever path it finds, including the one nobody foresaw. In an SME, that path runs through your emails, your CRM, your quotes. 3 safeguards are built in from the design stage, and that is how we build at Codito:

This is the job of Raphaël Zerbib, our CTO, a qualified cybersecurity engineer: building security into the architecture before the first line of code, so that the power of your agents serves your growth without ever slipping out of your control. Codito Ergo Sum is a partner agency certified by Anthropic.

Or tell me about your project on a call

An animated presentation video of 60 to 90 seconds, with voice-over and music, is commonly billed at between €1,500 and €10,000 by a freelancer or an agency. In this week's masterclass, I show you how to produce yours in 15 minutes, in your own brand colours.

The method comes down to 4 steps. Your AI writes the concept, the script and the prompt. Claude Design animates the video from your logo and brand guidelines. You export it as an MP4, without sound. An agent then adds the voice-over, music and sound effects, timed to each shot. The only cost on top of your Claude subscription: $6 a month for ElevenLabs (Starter plan, price listed in early October).

The complete kit is freely available, with no email or form required: the concept prompt, the sound agent and the pitfalls to avoid. Open the masterclass kit.

My advice for this week: before opening any tool, write in 3 lines your client's situation before, after, and what takes them from one to the other. That is the backbone of your video; the AI does the rest.