Usewright · Practical AI Guides (@usewrightapp) header
Usewright · Practical AI Guides (@usewrightapp) profile picture

Free AI how-to guides you can actually follow. 300+ step-by-step playbooks with copy-ready prompts and worked examples, plus plain-English AI news.

Joined April 2012
Photo by Usewright · Practical AI Guides (@usewrightapp)Photo by Usewright · Practical AI Guides (@usewrightapp)Photo by Usewright · Practical AI Guides (@usewrightapp)Photo by Usewright · Practical AI Guides (@usewrightapp)Photo by Usewright · Practical AI Guides (@usewrightapp)
Usewright · Practical AI Guides retweeted
Another official Anthropic plugin worth adding to your Claude Code stack… especially if you’re building MCP servers mcp-server-dev includes 3 skills: - build-mcp-server: Build MCP servers with tools and authentication - build-mcp-app: Create interactive UIs for MCP - build-mcpb: Package MCP servers for easier distribution Install it: /plugin install mcp-server-dev@claude-plugins-official A useful addition for anyone developing custom MCP integrations. GitHub: github.com/anthropics/claude…
26
13
2
161
13,034
Usewright · Practical AI Guides retweeted
a good tip that i figured out while working with skills in claude code: if you create a skill and you are unsure how to make it work better and better every time you trigger it, you should be able to improve upon that through the results that the skill gave. in the skill itself you embed the ability for it to run on a feedback loop and do recursive self-improvement on every run it does. it is optional: if you tell it to, it can ask you questions after each result it gives for you to review.
1
6
396
When an AI gives you a step-by-step answer, ask it which step it's least sure about. That's usually the one worth checking yourself.
8
Usewright · Practical AI Guides retweeted
Game Dev Tip #2: When running a team of AI agents to help develop your game it is important that you have the Game Editor (e.g. Roblox Studio, Unreal Editor, Unity Editor, etc.) always at the forefront of the screen. This is necessary so the QA Agent can properly run its tests on new updates, which always includes a performance check. If the Game Editor is in the foreground, then your computer automatically throttles the frame rate which leads to inaccurate test results. It is also important that you do not use your computer while the QA is running tests, but this creates a bottle neck. How can you continue brain dumping ideas all day to the Orchestrator Agent (i.e. the highest level agent)? One solution that has helped me (especially since I’m limited to one screen during the work week), is using the /remote-control command in Claude Code. This feature allows you to access your same Claude Code instance from your phone wherever you might be. Hope this is helpful. 🤝
Game Dev Tip: When using AI to help develop your game, prompt your LLM to analyze your game and create a lineup of multiple different permanent agents (all Opus 5.5 for my setup) that can be assigned to work on different aspects of the game. Then tell the LLM to have each agent to review the project to gain an understanding of what the game is and what the tech stack is like, then do deep research on their domain of expertise. The ability to now have a team of agents working on many different tasks in parallel has saved me significant time. I can freely brain dump all the different ideas I have through out the day without having to wait for my single LLM to free up their context. Different projects will have different types of agents spun up, but for Impact Soccer it made me 2 different game play coding agents, a graphics designer, a social media manager, a gameplay testing agent, multi-platform performance reviewer, sound and audio agent, project manager/planner, platform compliance agent, and a few others. Hope this is helpful 🤝
11
5
2
101
10,025
Usewright · Practical AI Guides retweeted
Claude Code tip: put "ultrathink" in a prompt and that turn runs at extra-high effort. Type it while Claude is mid-task and it raises effort for the rest of the turn. Spend it where thinking is the work: the plan, the gnarly bug, the migration. A copy change doesn't need it.
1
15
Usewright · Practical AI Guides retweeted
If you want to master AI agents in 2026, here is the complete 8-week curriculum (100% free): • Week 1: Hugging Face Agents Course (Unit 1) → Time: 1 week → Milestone: Trace a complete ReAct loop (thought, action, observation) on paper. Know exactly when an agent is overkill. • Weeks 2-3: DeepLearning .AI, "AI Agents in LangGraph" → Time: ~1h 45m → Milestone: Build a research agent with graphs, state loops, and human-in-the-loop gates. It resumes from checkpoints if crashed. • Weeks 4-5: Microsoft's AI Agents for Beginners (18 lessons on GitHub) → Time: 2 weeks → Milestone: Master multi-agent swarms, memory, planning, and agentic RAG. Defend pattern choices for 3 distinct production tasks. • Week 6: DeepLearning .AI, "Evaluating AI Agents" → Time: ~2h 35m → Milestone: Trace every tool call, score path lengths, and catch hallucinations before users report them Z
1
95
Usewright · Practical AI Guides retweeted
Recording videos but don’t have time to edit them? This guide shows how to set up an automation with Make, so every recording you add to a Google Sheet comes back as a captioned vertical clip, edited in your brand’s style.
2
4
256
Usewright · Practical AI Guides retweeted
It took the AI era for me to finally find my design tool, and maybe that's the whole point: pushing pixels around with a mouse never felt right. Now I use @paper every day. I've been updating this guide ever since I started it, because new tutorials, tips and tools keep coming. Man, I can tell you, the Paper community is ON FIRE! 🔥 So here's my collection. It's not everything out there, but these are the ones I'd recommend. Have a look: tympanus.net/codrops/?p=1227… #webdesign #ai
6
7
101
6,324
Usewright · Practical AI Guides retweeted
The best AI agent isn’t necessarily the one with the most tools, the largest context window, or the greatest number of subagents. It’s the one whose work can be understood, evaluated, and trusted within the limits of its design.
1
2
52
Wednesday AI habit: before you ask for a rewrite, tell the AI who's going to read it. "For my manager, who has two minutes" changes the result more than "make it better."
9
Usewright · Practical AI Guides retweeted
Introducing Tembo Computer. Every session runs in a full Linux VM on our own bare metal compute, 20% cheaper than E2B and Daytona. Claude Code, Codex, gh, Docker and Postgres come preinstalled. tembo.io/blog/announcing-tem…
26
7
6
118
8,363
Usewright · Practical AI Guides retweeted
building an interactive 3D website with Claude Code + Tripo CLI📷
32
47
2
936
30,193
Usewright · Practical AI Guides retweeted
5 things to check before you leave an AI agent running overnight: 1. Which speed tier it's on 2. The spend cap on its key 3. How many tasks it can start at once 4. Who gets the alert when it spikes 5. How you shut it off from your phone Save this for the next big run.
1
1
37
Usewright · Practical AI Guides retweeted
Claude Code is running an open-source 4K sky camera that caught 17 meteors overnight and had them sorted before its owner woke up. OpenSkyCam records the sky 24/7 and keeps all footage locally. Claude Code can then operate the system through its MCP server: > sample frames from the footage > run detection models > connect detections into tracks > separate moving objects from stationary ones > classify meteors, planes and fixed lights > review detections and reject false positives The interesting part is that the agent also improved the detector itself. When stars kept getting classified as flying objects, Claude built a training dataset from previously approved detections and retrained the model. On unseen nights, star false positives reportedly dropped from roughly 50% to 7%, while the system still detected about 90% of meteors. Every morning it throws away the junk and sends the useful detections through iMessage. If the night is cloudy, it sends nothing. The system also has actual guardrails outside the prompt: > destructive actions require confirmation > edits fail instead of overwriting newer versions > API keys can't create more keys > architecture checks fail if Claude puts code in the wrong layer Basically, Claude Code isn't just analyzing sky-camera footage here. It's operating and maintaining the entire observation workflow.
5
2
1
11
699
Easy audit to run this week: list every key, folder and account your agents can reach, then cut anything they haven't needed in the last month.
The takeaway for anyone running AI agents: they'll use whatever access you leave open. Worth a quick audit of what yours can reach.
2
Usewright · Practical AI Guides retweeted
tool of the day: orca free open source app that runs claude code, codex and opencode side by side, each agent in its own git worktree. there's a phone app to check on them and send follow-ups github.com/stablyai/orca follow along for a useful new tool every day
1
1
1
84
Usewright · Practical AI Guides retweeted
Google just gave coding agents a safe way into your databases. Data Agent Kit is GA now — free, open-source, plugs into Cursor, Claude Code, Codex and VS Code. 37 skills for best practice, and IAM rules that actually block a DROP TABLE. The skill asks nicely. The IAM policy doesn't.
3
2
124
Usewright · Practical AI Guides retweeted
Opening the repo was enough. JFrog found 42 real GitHub repos armed by the testmetest npm worm, with hooks that run the moment a project is opened in Claude Code, Gemini, Cursor or VS Code. The payload was too big for GitHub code search to see. Check your repos for hidden .claude, .cursor and .vscode hooks. Are you prepared for the cyber war? research.jfrog.com/post/beyo…
1
1
1
53
Usewright · Practical AI Guides retweeted
I have lived long enough to watch a note-taking app turn into an AI agent company. Notion is basically competing with Perplexity now.
Your Notion Agent can now use a computer to create and edit files. Go from a conversation to a finished Excel spreadsheet, Word doc, PowerPoint deck, or PDF. All without leaving Notion.
1
10
1,567
Before you paste a long document into an AI, tell it what you'll use the answer for. "I'm deciding whether to sign this" gets a more useful read than "summarize this."
2