|
Issue #001 · Fri, Oct 16, 2026 · 5-min read
🤖 Bots & Builders
AI, agents, games & vibe coding, explained in plain English.
|
|
Hey builder 👋 Welcome to the very first issue of Bots & Builders, your daily 5-minute roundup of what changed in AI, coding agents, and games, and what it means for the stuff you're making.
It was a big week: two major new models, a pile of updates to the tools you vibe-code with, and a game studio rethinking how it works. Grab a coffee and let's go.
|
|
🚀 Top Ship
Google releases Gemini 4 Argon
On October 13, Google released Gemini 4 Argon, its newest top-end model, to developers, businesses, and paying Gemini users at the same time. Google says it's built for long, hard, multi-step work like refactoring big codebases (reorganizing code without changing what it does) and running agent tasks, and that it ranks #1 on the Vals Index, a benchmark that scores models on real work in finance, coding, law, and tax. It can handle up to 1 million tokens of input and output combined (a token is roughly a chunk of a word, so that's a lot of code in one go). Launch pricing is $2 per million input tokens and $10 per million output tokens until November 30, then $4 and $20.
Why it matters to you: If you build on the Gemini API, you can use a frontier-class model at half price for the next six weeks, so this is a cheap time to point it at your messiest repo. Just remember to plan for the price doubling in December.
Read Google's announcement →
Oct 13
|
|
⚡ Quick Hits
|
1. Meta releases Muse Spark Open, a big open-weight model · Oct 15
Meta published the weights for Muse Spark Open: 816B total parameters (42B active at a time), a 1M-token context window, aimed at coding and agents, and able to run on two Apple M5 Ultra Mac Studios.
Why it matters: "Open weights" means you can download the model and run it yourself, and it's already on hosts like OpenRouter, Groq, and Together AI if you don't own two maxed-out Macs. Check the license, though: it's free for many developers, but the commercial terms are limited.
2. Codex CLI 0.163.0 gets a searchable settings panel · Oct 15
OpenAI's coding agent for the terminal now has a /config panel you can search, lets you send a follow-up to steer it while it's still answering, and can run read-only lookups in parallel. Separately, GPT-5.5 was retired from Codex for ChatGPT sign-ins on Oct 14; OpenAI suggests switching to GPT-6.1 Sol.
Why it matters: If a script or saved setting still asks for gpt-5.5 under a ChatGPT login, update it now. And steering mid-answer means you can stop a bad approach without starting over.
3. Lovable adds GPT-Live, Claude, Gemini, and Perplexity to the apps you build · Oct 15
Apps made in Lovable can now use OpenAI's GPT-Live for real-time voice, the latest Claude models, Gemini 4 Argon, Gemini Omni (video), Lyria (music), Perplexity (cited web answers), and scraping and search tools like Firecrawl and Parallel.
Why it matters: Adding "talk to my app" or "summarize the web with sources" is now a sentence in the chat, not a weekend of setting up API keys. You can even pick the model by name, like "use Perplexity."
4. Lovable will now help you find customers, too · Oct 14
Ask Lovable how to grow your app and it drafts a growth plan, then helps with SEO, Google Ads, and Facebook and Instagram campaigns from the same chat. No campaign goes live until you approve it.
Why it matters: For a lot of indie builders, shipping isn't the hard part; getting your first 100 users is. Ad spend is billed by Google or Meta, and Lovable says it doesn't add a fee on top, but planning and making ads uses your regular Lovable credits.
5. Cursor opens an app marketplace for Origin, its code-hosting service · Oct 13
Origin, Cursor's place to store code and review pull requests (still in early beta), now works with 20+ apps, including Vercel, Netlify, Cloudflare, Sentry, Linear, CodeRabbit, and Semgrep. It's also open to public submissions.
Why it matters: If you've wondered whether you could keep a project in Origin instead of GitHub, the deploy and CI tools you rely on are starting to show up there. If you make dev tools, you can now list one.
6. Replit Desktop is coming to Windows, plus one chat for all your projects · Oct 9
Replit opened a private preview (with a waitlist) of Replit Desktop for Windows, which builds and runs apps on your own PC in an isolated sandbox. You can now also ask one chat to find, read, and send tasks to any of your existing projects.
Why it matters: Moving a feature from Project A into Project B is now a single request, and Replit says nothing gets published automatically, so you review the changes first.
|
|
📦 Repo of the Day
heygen-com/hyperframes
"Write HTML. Render video. Built for agents." HyperFrames is an open-source (Apache 2.0) framework that turns HTML, CSS, media, and animations into MP4 videos. It comes with skills (installable instruction packs) for Claude Code, Codex, Cursor, Gemini CLI, and other agents, so you can ask for something like "a 10-second product intro with a fade-in title" and get a rendered video. It was on GitHub's weekly trending list today with about 64.6k stars, roughly 5.3k of them from this week.
Try it if: you need a launch video for your app and you'd rather describe it than learn video-editing software.
View on GitHub →
|
|
|
💡 Vibe Tip
Write the "done" line before the code
AI coding agents are great at writing code and not so great at knowing when to stop. Before you ask for a feature, give the agent a finish line it can check for itself:
"Before writing any code, restate what I'm asking for in 3 bullets, list the files you plan to touch, and write a short 'Done when…' checklist I can test by hand (e.g. 'Done when: clicking Save with an empty title shows an error and nothing is saved'). Wait for my OK. Then build it, and at the end, go through the checklist and tell me which items you actually checked."
Why it works: You catch misunderstandings before they turn into 400 lines of code, the agent has a clear stopping point, and you get a ready-made test list. It works the same in Cursor, Claude Code, Codex, Replit, or Lovable.
|
|
|
🎮 Game Dev Corner
Epic goes "AI first" on its Disney game
According to a Bloomberg report covered by Ars Technica on Oct 12, Epic Games is moving its delayed Marvel/Star Wars shooter with Disney to "AI-first" development: "no more hand-written code," with art "generated with full AI workflows" and humans directing the work. CEO Tim Sweeney confirmed a "huge pivot" toward "maximum AI acceleration" and said the goal "isn't to do the same stuff with fewer people." Plenty of players and developers pushed back hard.
Why it matters to you: The company behind Unreal Engine is betting its own big game on AI tools, so expect a lot more of them aimed at game makers. Small-scale example: on Oct 14 Ludo.ai released a command-line tool that lets coding agents like Claude Code generate sprites, 3D models, and sound effects right from your terminal (press release).
Ars Technica → · BBC →
|
|
|
That's the issue! Go build something weird this weekend, and hit reply to tell us what you're making. 🛠️
See you next issue,
Bots & Builders
|
|
|
|
You're reading the web version of Bots & Builders #001. · botsandbuilders.com · @BotsAndBuilders · Advertise: sponsors@botsandbuilders.com
|
|
