Thursday, October 22, 2026 AI makes promises. Le Recul verifies.
Newsletter
FR EN

AI tests · Clear verdicts · Updated ranking

AI media · Independent · No hype

AI makes promises. Le Recul verifies.

We test what AI promises, to separate what works, what gets in the way and what is oversold.

AI ranking

Full ranking →
  1. 01 97.37 Anthropic
  2. 02 96.32 Downward trend Anthropic
  3. 03 95.97 Downward trend Anthropic
  4. 04 95.27 Downward trend Anthropic
  5. 05 94.75 Downward trend Anthropic
  6. 06 94.71 Upward trend Anthropic

Top story — AI news

All AI news →
"No more hand-written code": Epic Games goes all in on vibe coding, independent measurements find a far smaller gain

AI news

"No more hand-written code": Epic Games goes all in on vibe coding, independent measurements find a far smaller gain

"No more hand-written code": that is the instruction given to the team behind Epic Games' next game, a Disney collaboration bringing together Marvel and Star Wars, revealed by Bloomberg on 12 October 2026. Tim Sweeney, the head of the company behind Fortnite, promises up to five times more productivity thanks to AI. We set those figures against the independent measurements published this year: the measured gain is far smaller, and the real cost shows up exactly where Epic is now placing its developers, in code review.

Comparisons

All comparisons →

Two tools, the same task, one verdict.

Claude Sonnet 5.5 vs Muse Wonder Full: level at 56 on paper, two very different machines in practice

Comparisons

Claude Sonnet 5.5 vs Muse Wonder Full: level at 56 on paper, two very different machines in practice

On the leading independent ranking, Anthropic's Claude Sonnet 5.5 and Meta's new Muse Wonder Full both score 56 points, and split the ten tests five wins apiece. But one is available from $20 a month, the other from $50, and not on the same terms for your code. We set the independent measurements, the first published trials and nearly a thousand user comments against each other: in practice, the two models no longer look alike at all.

Tested promises

All tested promises →

One promise. One test. One verdict.

Partly

Tested promise

Base44: the app really does arrive in minutes, the advertised price does not

The technical half of the promise is kept, and it is measurable: a four-screen app with a database, three user roles and authentication comes out in under six minutes and goes live in ten seconds. That is not marketing, it is reproducible. What is not kept is the price.

Read the test →

Overstated

Tested promise

AI text detectors: what the promised 99% reliability is really worth

The announced figures are not invented, but they describe a laboratory situation that is almost never the one in which the tool is used. Compilatio measures 94 to 99% reliability on homogeneous texts — entirely human or entirely generated — from questions put to the AI "without specific instructions on writing style".

Read the test →

Misleading

Tested promise

Can you really make €10,000 a month with AI?

The promise is misleading in its marketed form — accessible, fast, almost passive. A narrow path does exist, but it looks nothing like any of the adverts selling the promise.

Read the test →

Partly

Tested promise

Social media automated by AI: real gains, or a generic content factory?

AI genuinely speeds up production (scripts, voice-overs, visuals, editing, subtitles, multi-platform publishing): the "produce faster" promise is kept. But direct pay-per-view monetisation stays low and conditional — high thresholds, videos of at least 1 minute on TikTok, content judged original. The real revenue comes from what you sell behind the account, not from the workflow itself.

Read the test →

Latest analyses