🧭 Choose the right task + explain why

Fable 5 was built to maintain context, make chained decisions and sustain reasoning over long runs. That takes focus: when you point this capability at a two-minute task—summarizing a short text, answering a quick factual question—there’s no “extra” capacity being used. You’ve just paid for an advanced model to do a simple model’s work.

The real benefit shows up at the other end of the spectrum: when the task requires keeping the whole picture in mind at once. Review an entire project's architecture, assess product/technology/business risks, turn a loose idea into an execution plan, analyze long documents, review code/security/UX/databases, organize complex strategies, create diagnostic and tracking systems — all of this calls for sustained depth, not a quick answer.

This is a prompt tested in practice, not an official documentation quote—it’s a recommended template for when the task has already passed the filter from the previous topic. It works because it replaces “share some ideas” with a numbered list of deliverables: the model tends to cover exactly the items you numbered, in the order you numbered them.

Notice the order of the 6 items: diagnosis first (what's wrong), then decisions and risks (what to do and what could go wrong), only then the plan and deliverables, ending with what to do now. This sequence avoids the classic trap of jumping straight to a solution without understanding the problem.

Estou trabalhando em <projeto maior> para <público>. O resultado final precisa
permitir <objetivo real>.

Analise o projeto inteiro como um agente sênior. Não me dê ideias soltas. Entregue:
1. diagnóstico do problema central;
2. decisões estratégicas necessárias;
3. riscos;
4. plano de execução em etapas;
5. entregáveis concretos;
6. o que deve ser feito agora, nesta ordem.

The model responds better when it understands the intent behind the request: what the bigger goal is, who will use the result, what context it fits into, and what the answer needs to enable. Without that, it fills in the gaps with the most generic interpretation possible for that kind of request—which is rarely exactly what you needed.

This isn't a new rule: it's an established prompting practice, tested again here in the specific context of using Fable 5 as a work agent—not a definitive citation from the documentation. Treat it as a solid recommendation validated in practice, not a rule set in stone.

See the same request in both versions. The difference isn't in the "what" — both ask for a 30-day learning path. The entire difference lies in the reason: who the audience is, what they need to achieve, and how that translates into a 6-part structure instead of a loose list of days.

❌ FRACO:
Crie uma trilha de IA.

✅ FORTE:
Estou criando uma trilha para profissionais que querem aprender IA de forma
prática e aplicar no trabalho. O objetivo é fazer a pessoa praticar todos os
dias e construir projetos reais.

Com isso em mente, crie uma trilha de 30 dias com:
1. diagnóstico inicial;
2. prática diária;
3. projeto simples por semana;
4. conteúdo de apoio;
5. critério de evolução;
6. resultado final esperado.

Tips 1 and 2 aren’t sequential steps — they’re the same opening sentence. You choose a task that really calls for depth (it doesn’t ask for a summary) e explain the reason/context right there in the first message — not later, as an addendum after the answer has already come back generic.

Realistic example: a solo founder about to hire their first team member needs to restructure the entire project beforehand—including architecture decisions, priorities, and what to delegate. This is a task where Fable 5 is worth it (tip 1), and the reason (“I’ll use this to make a hiring decision in the next 4 weeks”) completely changes what the response needs to prioritize (tip 2).

Estou reestruturando <nome do projeto> antes de trazer a primeira pessoa
do time. Preciso que esta análise sirva de guia pras próximas 4 semanas de
decisões — não é curiosidade, é o documento que vou seguir pra contratar e
priorizar.

Analise o projeto inteiro como um agente sênior, considerando que quem vai
usar isso sou eu sozinho, tomando decisões de arquitetura, prioridade e
contratação. Não me dê ideias soltas. Entregue:
1. diagnóstico do problema central;
2. decisões estratégicas necessárias;
3. riscos;
4. plano de execução em etapas;
5. entregáveis concretos;
6. o que deve ser feito agora, nesta ordem.

The most common mistakes aren’t conceptual—they’re habits. You know you should provide context, but you’re in a hurry and send only the task. You know the task is large, but you use the same generic prompt as always. The list below is a quick antidote: review it before you hit send, not after you get a generic answer.