Te ensino como usar ferramentas para ser mais produtivo, organizado e ter sucesso naquilo que busca.
Belo Horizonte, Brazil
Joined May 2013
- Tweets3.9K
- Following317
- Followers360
- Likes3.5K
Agents writing the game is half of it. The other half is agents playing it: running matches overnight and logging every decision they make. That's the side I built Age of Mist (ageofmist.com) for, a strategy game AI agents play over MCP or REST.
I'm building a cartoon FPS with a team of AI agents.
First commit Sep 29, free demo on Steam Oct 23, 4,558 commits. I wrote almost no code.
@GodotEngine + Claude Code @claudeai + @ElevenLabs
One person can build a game now. Workflow and agent brief below (1/8)
Useful if all your revenue runs through Stripe. Many info-product sellers also use Hotmart or Kiwify, so an agent querying Stripe sees only part of the month. I built Revenue Planet (revenueplanet.studio) to put Stripe, Hotmart, Kiwify and others in one view.
Analyze your Stripe data with your agent. Now available to all businesses.
• Build custom reports, understand and prevent fraud, and track MRR, churn, and retention.
• Ask questions in natural language.
• Connect your agent via the Stripe MCP or by using the Queries and Reports APIs directly.
Queries are free through January 31, 2027.
"I turned off and deleted all 4 of my keyword responses nearly a year ago. They're still firing."
That's a ManyChat review on Trustpilot.
Comment-to-DM breaks in two ways: it stops when it should run, or it runs when it should stop.
So I built my own.
Postgate: comment-to-DM for Instagram on the official API. Every automation gets simulated before it goes live, pause stops it for real, and each run shows up in a log.
postgate.studio
4,558 commits in 24 days shows agents can build a game. Next test: can they play one well? I built Age of Mist (ageofmist.com), a strategy game where every empire is run by an AI agent over MCP or REST. Planning across turns is a different skill from shipping commits.
I'm building a cartoon FPS with a team of AI agents.
First commit Sep 29, free demo on Steam Oct 23, 4,558 commits. I wrote almost no code.
@GodotEngine + Claude Code @claudeai + @ElevenLabs
One person can build a game now. Workflow and agent brief below (1/8)
Competitors as proof works for your own stack too. I was paying $400/month in posting tools, so I built Postgate (postgate.studio): a posting API with MCP, plus comment-to-DM. Today it costs me nothing to run for my own accounts.
After failing for 7 years, I finally went from $0 to $30k MRR.
Here's the advice I wish I had:
𝟭) 𝗣𝗶𝗰𝗸 𝗮 𝘃𝗮𝗹𝗶𝗱𝗮𝘁𝗲𝗱 𝗻𝗶𝗰𝗵𝗲 𝗶𝗻 𝗮 𝗴𝗿𝗼𝘄𝗶𝗻𝗴 𝗺𝗮𝗿𝗸𝗲𝘁
Competitors are proof people pay. Go B2B if you can: B2C acquisition is more expensive and the LTV is lower.
𝟮) 𝗩𝗮𝗹𝗶𝗱𝗮𝘁𝗲 𝗱𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗶𝗼𝗻 𝗯𝗲𝗳𝗼𝗿𝗲 𝘆𝗼𝘂 𝗯𝘂𝗶𝗹𝗱
DM people who have the problem. Make the offer. If they say no, ask why. Get yeses before you write code.
𝟯) 𝗦𝗵𝗶𝗽 𝘁𝗵𝗲 𝗠𝗩𝗣 𝗳𝗮𝘀𝘁, 𝗻𝗮𝗶𝗹 𝗼𝗻𝗯𝗼𝗮𝗿𝗱𝗶𝗻𝗴
Solve the core problem, nothing else. Then make sure users hit the value in their first session. If they have to figure it out alone, they won't.
𝟰) 𝗕𝘂𝗶𝗹𝗱 𝗼𝗻𝗲 𝗰𝗵𝗮𝗻𝗻𝗲𝗹 𝘁𝗵𝗮𝘁 𝘀𝗲𝗻𝗱𝘀 𝟭𝟬𝟬+ 𝘃𝗶𝘀𝗶𝘁𝗼𝗿𝘀 𝗮 𝘄𝗲𝗲𝗸
Not a launch. A system you can repeat every week. Mine was Reddit: 10 minutes a day.
𝟱) 𝗧𝗮𝗹𝗸 𝘁𝗼 𝘂𝘀𝗲𝗿𝘀 𝗲𝘃𝗲𝗿𝘆 𝗱𝗮𝘆
Ask why they signed up. Ask what confused them. Ask the ones who canceled why they left. The feedback you get will improve your offer like nothing else.
𝟲) 𝗧𝗿𝗮𝗰𝗸 𝘁𝗵𝗲 𝗳𝘂𝗻𝗻𝗲𝗹, 𝗳𝗶𝘅 𝘁𝗵𝗲 𝘄𝗼𝗿𝘀𝘁 𝗻𝘂𝗺𝗯𝗲𝗿
Visitor → signup → trial → paid → retained. Find the biggest drop-off, fix it, repeat. This step alone took me from $1k to $15k.
𝟳) 𝗦𝘁𝗮𝗰𝗸 𝗰𝗵𝗮𝗻𝗻𝗲𝗹𝘀
Only once the funnel holds, start adding new channels on top. Don't stop doing what already works.
---
It took me 7 years to learn this. You get it for free. Use it.
André Lug retweeted
After failing for 7 years, I finally went from $0 to $30k MRR.
Here's the advice I wish I had:
𝟭) 𝗣𝗶𝗰𝗸 𝗮 𝘃𝗮𝗹𝗶𝗱𝗮𝘁𝗲𝗱 𝗻𝗶𝗰𝗵𝗲 𝗶𝗻 𝗮 𝗴𝗿𝗼𝘄𝗶𝗻𝗴 𝗺𝗮𝗿𝗸𝗲𝘁
Competitors are proof people pay. Go B2B if you can: B2C acquisition is more expensive and the LTV is lower.
𝟮) 𝗩𝗮𝗹𝗶𝗱𝗮𝘁𝗲 𝗱𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗶𝗼𝗻 𝗯𝗲𝗳𝗼𝗿𝗲 𝘆𝗼𝘂 𝗯𝘂𝗶𝗹𝗱
DM people who have the problem. Make the offer. If they say no, ask why. Get yeses before you write code.
𝟯) 𝗦𝗵𝗶𝗽 𝘁𝗵𝗲 𝗠𝗩𝗣 𝗳𝗮𝘀𝘁, 𝗻𝗮𝗶𝗹 𝗼𝗻𝗯𝗼𝗮𝗿𝗱𝗶𝗻𝗴
Solve the core problem, nothing else. Then make sure users hit the value in their first session. If they have to figure it out alone, they won't.
𝟰) 𝗕𝘂𝗶𝗹𝗱 𝗼𝗻𝗲 𝗰𝗵𝗮𝗻𝗻𝗲𝗹 𝘁𝗵𝗮𝘁 𝘀𝗲𝗻𝗱𝘀 𝟭𝟬𝟬+ 𝘃𝗶𝘀𝗶𝘁𝗼𝗿𝘀 𝗮 𝘄𝗲𝗲𝗸
Not a launch. A system you can repeat every week. Mine was Reddit: 10 minutes a day.
𝟱) 𝗧𝗮𝗹𝗸 𝘁𝗼 𝘂𝘀𝗲𝗿𝘀 𝗲𝘃𝗲𝗿𝘆 𝗱𝗮𝘆
Ask why they signed up. Ask what confused them. Ask the ones who canceled why they left. The feedback you get will improve your offer like nothing else.
𝟲) 𝗧𝗿𝗮𝗰𝗸 𝘁𝗵𝗲 𝗳𝘂𝗻𝗻𝗲𝗹, 𝗳𝗶𝘅 𝘁𝗵𝗲 𝘄𝗼𝗿𝘀𝘁 𝗻𝘂𝗺𝗯𝗲𝗿
Visitor → signup → trial → paid → retained. Find the biggest drop-off, fix it, repeat. This step alone took me from $1k to $15k.
𝟳) 𝗦𝘁𝗮𝗰𝗸 𝗰𝗵𝗮𝗻𝗻𝗲𝗹𝘀
Only once the funnel holds, start adding new channels on top. Don't stop doing what already works.
---
It took me 7 years to learn this. You get it for free. Use it.
"We can't get it to move 50 units per turn. It can create a lot of illegal moves."
That's the creator of an AI mod for Civ V, on Reddit.
LLMs are bad at micromanaging a game built for humans.
So I built a strategy game built for agents.
For a small SaaS or infoproduct business, the slow pass is the first, gathering inputs: revenue sits in Stripe, Hotmart and Kiwify, each with its own export and refund logic. I built Revenue Planet (revenueplanet.studio) to put those in one view before the model starts.
I built more 3-statement models by hand at Morgan Stanley than I'd like to admit. Claude in Excel is a game changer .... here's how I build any model in 5 passes:
The process is all about context window management to minimize hardcodes + hallucinations and maximize control over model style.
(1) Gather the sources. Filings, earnings transcripts, raw financials and any model or template I already have. I have it make a clean folder I point Claude to.
(2) Turn the sources into a JSON or .md input. A short inputs file with historicals, segment KPIs, accounting definitions and operating drivers. This step converts many megabites of PDF's into a liteweight doc for RAG.
(3) Build the structure (in your firm style). Building the shell lets you reverse engineer your firm model style .... which Claude will never have context for.
(4) Add the forecast. I take the same historicals JSON or .md and run a deep research workflow to fill in those parameters but for forward assumptions. Obviously with sources so I can check assumptions. Then I layer in that file to project out forward assumptions.
(5) Audit it in a fresh chat. New chat is necessary to hard reset the context window (improves quality). A simple prompt even to check for hard + soft errors can rank issues by severity (eg broken formulas and things that don't tie to assumptions that look off)
If you're really advanced at Claude Code - you can wrap this into a programmatic skill package in your firm build style .... maybe I write a thread about that next time.
In the meantime, I've put together this AI for Excel build guide here: davewang.ai/resources/how-to…
Roblox opening Studio to agents covers building games. The other side is agents as players. I built Age of Mist (ageofmist.com), a strategy game where AI agents play through MCP or REST. Building for agents and playing with agents will need very different APIs.
Quem já fez mais de 4 bilhões de views para marcas como Buldak e Replit explicou como viralizar sem depender de sorte.
São 8 passos 🧵
6. Antes de postar, peça para um amigo explicar o vídeo em uma frase. Se ele não conseguir, ninguém vai compartilhar.
7. Corte todo segundo em que nada novo acontece e legende tudo. Muita gente assiste sem som.
8. Quando um vídeo pegar, refaça com outro ângulo. Um vídeo de 1,8 milhão rendeu mais dois, de 600 mil e 300 mil.
Para o passo 3, fiz o AI Maestro: você cola a estrutura e ele devolve roteiro e carrossel. aimaestro.studio