God of Prompt

🔑 Your key to AI independence

Get powerful prompts, custom workflows, guides and essential info on all you may need in AI

[God of Prompt @godofprompt]

https://t.co/aTFlJd0FvZ

[God of Prompt @godofprompt]

You can build a working budget or client tracker in Claude this week. No Excel, no formulas. 24,000 people saved the announcement overnight, and Microsoft dropped 3% in 30 minutes.

Claude Sheets shipped on Monday (beta, paid plans). You ask for a tracker or a budget and you get a working spreadsheet you can keep building.

Here's the trap. You type "make me a tracker" and you get five columns and no logic. The sheet is exactly as good as the brief you gave it.

Two prompts, in this order:

Prompt 1: The brief (Opus 5.5, in Claude Sheets)
"Before building anything, ask me 6 questions, one at a time: what I'm tracking, where the numbers come from (manual, export, an app), how often I update it, who else reads it, the 3 decisions I make from it, and what breaks in my current setup. Then propose the structure: tabs, columns, which cells are inputs and which are formulas, and 3 checks that catch my mistakes (totals that must match, dates out of range, duplicates). Wait for my OK, then build it. No sample data that looks real. Put every assumption in a Notes tab."

Prompt 2: The audit (Fable 5.2, once. It shipped a day earlier, caught a bank statement where deposits didn't match the balance in Anthropic's tests, and counts for up to 50% of your weekly Pro limit, so one pass is enough)
"Here is my spreadsheet. Don't rebuild it. Check every formula against what its column header promises. Find circular logic, hard-coded numbers that should be inputs, and totals that silently skip rows. Give me a table: cell, problem, fix, severity. Then list the 3 assumptions most likely to be wrong for a business like mine. Max 300 words."

Excel didn't lose to a chat box. It lost to people who never liked formulas and now don't have to write them.

You can grab more business prompts in the God of Prompt library. 7,200+ prompts, free, no signup. Link in the first reply.

Would you move your business tracker out of Excel this month, or wait until Sheets is out of beta?

[Robert Youssef @rryssf]

so chatgpt is the new google business profile now

the business agent answers customers off whatever info you put in there.

worth 20 min writing that context properly before your competitors figure it out

[God of Prompt @godofprompt]

@DeItaone two frontier models shipped today and the market read the token bill

[God of Prompt @godofprompt]

Finally it’s official!

[Stripe @stripe]

OpenRouter is now part of Stripe.

Stripe is building the economic infrastructure for AI. Together with OpenRouter, we’ll help businesses explore new models and optimize token routing and usage.

[God of Prompt @godofprompt]

@unusual_whales clippy waited 30 years for this

[God of Prompt @godofprompt]

@Polymarket the grow lights were already pulling data center power anyway

[God of Prompt @godofprompt]

@unusual_whales called claude. it asked if i'd like that formatted as a table

[God of Prompt @godofprompt]

@trq212 one shot demos sell the model. turn forty is where people actually live

[God of Prompt @godofprompt]

@GoogleDeepMind verification loops are great until nobody tells it what done looks like

[God of Prompt @godofprompt]

@gdb whoever writes their business description like a system prompt wins local search

[God of Prompt @godofprompt]

@unusual_whales softbank pulling back is the most bullish ai signal of the year

[God of Prompt @godofprompt]

this is basically context engineering for small businesses

> claim the listing
> write hours, pricing, faqs like you're briefing a new hire
> test the agent with the 10 questions customers actually ask

most people will fill 3 fields and wonder why it answers wrong

[OpenAI @OpenAI]

Now in beta: Businesses in ChatGPT.

Claim your business, manage your information, and get insights on how people engage with you.

Eligible businesses can also build a Business Agent to answer customers’ questions and help them explore what you offer.

https://t.co/N2GEZX0mFT

[God of Prompt @godofprompt]

@wallstreetbets cafeteria lady vesting schedule: four years, one-year cliff, unlimited pudding

[God of Prompt @godofprompt]

@unusual_whales the first ipo where retail already knows the product's limits

[God of Prompt @godofprompt]

@Polymarket a flip phone at least never hallucinated a phone number

[God of Prompt @godofprompt]

@claudeai @kevin_t_ngo same prompt across three models is the only benchmark that fits in a tweet

[God of Prompt @godofprompt]

https://t.co/aTFlJd0FvZ

[God of Prompt @godofprompt]

You can build a AAA-looking shooter with Fable 5.2 from a two-sentence prompt. It will cost you $550 and 8 hours. Or you can spend 5 minutes on a spec and pay a fraction of that.

Fable 5.2 landed in Claude plans last night. The first thing that went viral: an Anthropic engineer, Felix Rieseberg, one-shot a full game with it. Models, textures, sounds, from scratch, no feedback. 1M views overnight.

His whole prompt: "You know, you should also build me a FPS. Insanely detailed quality pls, something that looks visually like it can keep up with AAA. Make it small in scope (like a tiny level) but otherwise impressive photorealistic graphics."

His verdict: "the age of prompt engineering is basically over."

The bill under the video: 445M input tokens, 2.7M output, about $550 at API prices, 8 hours.

Here's what you're paying for with a two-sentence prompt. The model writes the whole spec itself: scope, stack, asset list, what "AAA" means. Then it rewrites parts of it, over and over, and every pass goes on your meter. On the API that's $550. On your Pro plan that's your week, and Anthropic caps Fable at 50% of your weekly limit anyway.

Anthropic's own launch post tells you how to use it: Opus 5.5 is the daily driver, Fable is for the hardest problems. They even name the pattern, Advisor: a cheaper model does the work and asks Fable only when it's stuck.

You can run the same pattern with two prompts:

Prompt 1: The spec first (Opus 5.5)
"Before writing anything, write the spec for this: what's in scope, what's out, the stack, the file list, and 5 checks that prove it works. Keep it under 300 words. Stop and wait for my OK. After my OK, build file by file and run the checks after each one."

Prompt 2: The advisor (Fable 5.2, one call)
"You're the advisor, not the builder. Here is a task and the spec another model wrote for it. Don't write code. Find the 3 decisions most likely to be wrong, slow or expensive, explain why in two lines each, and give the fix. Then list the questions I should answer before anyone builds. Max 300 words."

Two sentences work when Anthropic pays the bill. Your five-minute spec is what saves you the second run.

You can grab more prompts like these in the God of Prompt library. 7,200+ prompts, free, no signup. Link in the first reply.

Is prompt engineering over, or did it just move into the spec?

[God of Prompt @godofprompt]

https://t.co/Yge05bthCV

[God of Prompt @godofprompt]

You can audit your whole codebase with one prompt. You can also do it without losing 20% of your weekly Claude limit.

The prompt everyone saved last night (14,000 bookmarks, from Alex Albert at Anthropic):

"Read this entire project and find everything that's broken, inconsistent, inefficient, or no longer needed. Fix what you're sure about and list the rest for me to review."

You can paste it as is. The top replies tell you what happens next: "RIP weekly usage limit" and "20% weekly limit gone in a single prompt."

That's not a bug. "Read this entire project" puts every file into context. And Fable 5.2, out the same day, can only use up to 50% of your weekly Pro limit. One audit on a mid-size repo and half your week is gone.

You can get the same audit for a fraction of the tokens. Three passes instead of one:

Pass 1, the map (Opus 5.5):
"Don't read file contents yet. From the folder tree, file names, sizes and imports, give me one line per file: what it does, and a 1 to 5 guess of how likely it hides dead, duplicated or inconsistent code. Flag anything over 500 lines or not imported anywhere. Table format. Then ask me which folders to open."

Pass 2, the audit (Opus 5.5, one folder at a time):
"Read only [folder]. Find everything broken, inconsistent, inefficient or no longer needed. For each one: file:line, what's wrong, severity (high/med/low), your confidence (sure/unsure). Fix only the sure ones. Keep existing behavior and tests green. No style refactors. Then give me the unsure list as a table."

Pass 3, the second opinion (Fable 5.2, once):
"Here is a list of possible issues another model wasn't sure about, with file paths. Don't re-read the project. For each: fix, ignore, or need more info, one line with the reason. Be blunt."

Why you pay less: pass 1 reads almost nothing, pass 2 only reads what pass 1 pointed at, and the expensive model never sees your repo, just a list.

The one-prompt version is fine when the limit is someone else's problem. On your $20 plan, the prompt has to say what NOT to read.

You can cut your token costs further with our free guide: caching, routing, effort settings, 10 before/after examples at current prices. Link in the first reply.

Do you run one big audit prompt, or chip at it folder by folder?