Brief IA
›

Fable, ChatPRD, Claude Code: AI Put to the Test of Use Cases

AI Tools·Tom Levy·

Fable, ChatPRD, Claude Code: AI Put to the Test of Use Cases

Fable, ChatPRD, Claude Code: AI Put to the Test of Use Cases
Key Takeaways
1Claire claims that token fraud has cost ChatPRD tens of thousands of dollars.
2ChatPRD has removed its free trials and stopped issuing promotional codes.
3Scott Farrar uses Claude Code and Obsidian to organize information in his household.
Why it matters : at ChatPRD, every fraudulent call to a model incurs a real expense, even when the service already has security controls in place.

Claire reports convincing creative results with Fable 5.2, but still refuses to delegate a website solely to it. At ChatPRD, she has closed free trials after costly abuses. Scott Farrar, a mathematics educator, has chosen another form of control for his household: searchable files rather than a memory locked in conversations.

Claire reserves Fable for complex or creative tasks

A complex animation completed in 16 minutes, exercises often finished in under a minute: Claire reports these durations after trying Claude Fable 5.2 from Anthropic. She also appreciates its interventions being limited to moments when human action becomes necessary. Her judgment contrasts with her view of the first Fable, which was difficult to understand and unpleasant to use, and then Opus 5, which had left her less convinced.

For a marketing video, she provided it with a tweet and the graphic guidelines from CXO.dev. According to her, the result tells a story with customized invader characters capable of walking, jumping, or clapping, where other models tried had delivered nothing comparable. She now considers this exercise as a benchmark test. In another area, poorly defined photos of a disassembled piece of furniture, lacking instructions, were enough for Fable to recognize the parts, draw a diagram, and propose an assembly successfully followed by her household, after the failure of other models.

Claire also entrusted it with the resumption of a complex pipeline from ChatPRD and reports the creation of a functional 3D typewriter. She appreciates the humor in a game dedicated to bugs and engineering trade-offs, extended by a trailer where the finance department imposes a cheaper model after discovering token expenses. She finds this version more accessible, faster, and creative, yet still does not delegate the autonomous design of a website to it. Her preference remains Opus 5.5 for most coding and knowledge tasks; she would choose Fable for broader, difficult, or creative missions.

ChatPRD closes promotional access after abuses

At ChatPRD, Claire estimates the damage from token fraud at tens of thousands of dollars, alongside a significant portion of her engineering time. As an experienced solo founder with technical expertise, she leads a product compliant with SOC 2 and regularly subjected to penetration testing. Each fraudulent use of a model represents an actual expense for the provider.

She distinguishes six categories of abuse, including diverted trials, reselling promotions, card testing, product relays, and exploiting the MCP. According to her account, automated account creations provided access to the most expensive models to execute work unrelated to ChatPRD. Neither the requirement for a credit card nor paid trials sufficed: she has eliminated free trials, giving up a channel she deemed valuable for product adoption and enterprise prospects.

Promotions intended for partners were also stolen and resold to fuel other AI businesses, Claire claims. Estimating that about half of her codes had reached resellers, she stopped issuing them. She now recommends continuous monitoring of expenses rather than just counting requests, closing free entries, and verifying clients beyond their accounts. In her experience, abuses quickly cease once expensive models become inaccessible to fraudsters.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

ChatPRD classifies suspicious sessions with Jev

For sessions exceeding her abuse threshold, ChatPRD uses a Jev loop without data retention, with inexpensive decision models that classify activity in real-time. A confirmed fraud leads to a ban or a discreet switch to a cheaper model. According to Claire, one user had also transformed the custom MCP support into a 24/7 available computing resource; she limited the admitted hosts and blocked free tunnels as well as infrastructures with no apparent connection to a legitimate business.

After massive exclusions, clumsy appeals coexisted with polite AI-generated messages, she recounts. Her support agent now has anti-fraud tools and guidelines explicitly prohibiting the reactivation of accounts or the granting of promotions. Claire also reports a significant decrease in abuses after blocking certain countries with Vercel Firewall, a decision that led her to reassess her international strategy. She now believes that trust is earned rather than assumed.

Claire claims to have identified the same fraudsters behind rotating proxies on several major AI services. She has begun to publicly discuss these practices. With Codex, she has also prepared cybercrime reports filed with the FBI.

Scott Farrar retains context in verifiable files

Mathematics educator and parent, Scott Farrar has assembled Claude Code and Obsidian to manage a shared memory in the household. He prefers the context preserved in Markdown to lengthy conversations, which he criticizes for their memory drift: files remain readable, correctable, and reusable from one session to the next. He provides raw information and lets Claude organize it, while a custom linter prevents agents from overly extending documents, particularly the one for chores.

When asked about housing documentation, Claude suggested he photograph the appliances. These images fed into a searchable inventory including their references, serial numbers, and warranties. Scott Farrar sees this as a conversational version of his father's binder, who kept receipts and warranties and could explain repairs done decades ago. When the agent strays from the task, he restarts or inspects its reasoning before erroneous data enters the memory files.

The household shares tasks without imposing everything on GitHub

Scott Farrar and his wife initially listed household tasks that came to mind. Claude then prioritized them based on urgency, cost, and necessary information; according to Scott Farrar, the system highlights delays before they turn into emergencies. Family projects and chores are tracked in a shared list, with a wall screen displaying planned movements, remaining tasks, recent accomplishments, and daily suggestions.

The sharing did not fully follow Claude's recommendation, which suggested GitHub to prevent conflicts between files. His wife found this organization too complex, so the couple reserves GitHub for common scripts and uses Dropbox for everyday documents. For insurance, Claude also compares quotes to the characteristics of the home, flags possible coverage gaps, and prepares targeted questions. Scott Farrar thus feels less dependent on the explanations of the person selling him the contract.

Scott Farrar wants tutors that require reasoning

In teaching, Scott Farrar favors assistants that ask students to articulate their reasoning rather than solve exercises for them. He wants them to remain encouraging while maintaining a standard, without erasing difficulties that he considers useful for learning. He also believes that some children may find it easier to submit still-undeveloped ideas to an AI than to a human interlocutor.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.