Fable 5.2 Prompting: Drop the Old Prompt Tricks

I’ll admit it. For a long time, I started my prompts the same way every time. “Think step by step.” “Take a deep breath.” “Do not hallucinate.” I typed them out of habit, a bit like knocking on wood before hitting enter.

Then I read a LinkedIn post from Ruben, and it stopped me. He says the biggest mistake you can make when prompting Fable 5.2 is prompt engineering itself. He doesn’t mean bad prompt engineering. He means all of it.

The author says he doesn’t care what other AI creators are teaching. His point is simple. Stuffing your prompt with “Act like Albert Einstein” or “think step by step” makes you feel like a prompting wizard. The model just doesn’t need it anymore. In his words, prompt engineering is long gone, and models are smart enough to reason.

Old habits can also work against you. You can pay for the max subscription of the smartest AI on earth and still get Sonnet-level answers because your prompts are stuck in the past. That line hit home for me.

Why Old Tricks Backfire on Fable

It comes down to how literally Fable follows instructions. According to the original poster, Fable strictly obeys every “never” you write, even after you say yes later. A rule from the top of a chat can quietly override permission you gave ten messages later.

That’s a big shift. Tricks don’t help much anymore. What helps is clear, specific direction about permissions, scope and when the job counts as done. Here are the four fixes the expert shared, plus my notes on using each one.

1. Swap “Never” for “Ask Me First”

A hard “never” becomes a wall Fable won’t climb, even when you later want it to. The fix is to turn bans into checkpoints and give the reason behind them.

  • Old: Never change my prices.
  • New: Ask me before you change a price: my clients have signed quotes.

Look at the second half. The reason (“my clients have signed quotes”) tells the model what’s at stake. That way it can handle edge cases with some judgment instead of just following a rule. I’d use this pattern for anything sensitive, like contracts, published content, live code or budgets.

2. Spell Out What You Approve

“Go” feels efficient. But to a model that takes you literally, “go” is vague. Does it mean everything you discussed, or only the last item? The creator recommends naming exactly what’s approved and what’s off-limits.

  • Old: Go.
  • New: Yes, rewrite the 3 welcome emails. Leave the invoice template alone.

This takes five extra seconds and saves you from undoing changes you never asked for. A quick tip: when you approve a plan, list what’s in scope and name at least one thing to leave untouched.

3. Tell It to Finish the Job

This one is sneaky. You give Fable a long task, go grab a coffee, and come back to find it waiting at “Want me to continue?” Nothing happened while you were gone.

The fix is one line you can add to any long task:

Finish the whole job before you stop. Don’t offer to continue: continue.

I love how blunt it is. It’s perfect for batch work like rewriting a set of pages, going through a list of files or drafting a full content calendar.

4. Don’t Switch Models Mid-Chat

This tip surprised me the most. Each model keeps its own copy of your chat, called the cache. The cache is how Claude rereads a long chat without burning through your usage. But if you switch from Fable to Opus in the same chat, Opus has to reread everything from zero.

Put simply, switching models mid-chat quietly eats into your limits. The smarter move is to summarize your context while you’re still on Fable. Then open a fresh chat on the other model and hand it a clean brief.

The author shared the start of a handoff prompt for exactly this:

“Write a brief so a new chat on Claude Opus 5.5 can take over this work. It won’t see this chat or your thinking. Use these parts:
1. Goal: one sentence, then “Done when:” with the checks I’d use.
2. … “

He says the full prompt has 9 parts, and the post only shows the first one. Even that part teaches a lot. A one-sentence goal plus a “Done when:” checklist gives the new model a clear target and a clear finish line. If you want to build out your own version for now, I’d add sections like key decisions already made, files or sources in play, and open questions. That’s my suggestion, not the author’s.

How I’d Put This Into Practice

Here’s a simple routine you can run this week, based on the original poster’s advice:

  1. Search your saved prompts for “never” and rewrite each one as “ask me before…” plus a short reason.
  2. Delete filler like “take a deep breath” and “act like a world-class expert” and see if your results change.
  3. Replace one-word approvals like “go” or “ok” with scoped ones that say what to touch and what to leave alone.
  4. Add the “finish the whole job” line to any task you plan to walk away from.
  5. Stick to one model per chat, and when you need to switch, hand off with a written brief instead.

The idea behind all four fixes is the same. Smart models don’t need tricks. They need clear permissions, clear scope and a clear finish line. Talk to Fable like a capable colleague, not like a machine that needs magic words.

I think this is one of the most useful prompting posts I’ve read in a while, mostly because it’s about unlearning. Check out the full LinkedIn post for Ruben’s complete take, then see how many old prompt habits you can drop this week.

Scroll to Top