Tons of AI guides have dropped lately.

Most repeat the same ideas, but a few are genuinely worth your time.

Here are 8 must-read guides from Google, Anthropic, OpenAI, and Kaggle 🧵 ↓

1/ Prompt Engineering – by @Google

Hands-on strategies to craft powerful and reliable prompts.

→ https://www.kaggle.com/whitepaper-prompt-engineering

2/ Prompting Guide 101 – by @Google

A beginner-friendly resource to master the fundamentals of prompt design.

→ https://services.google.com/fh/files/misc/gemini-for-google-workspace-prompting-guide-101.pdf

3/ AI in the Enterprise – by @OpenAI

A framework for embedding AI into real business operations.

→ https://cdn.openai.com/business-guides-and-resources/ai-in-the-enterprise.pdf

4/ A Practical Guide to Building Agents – by @OpenAI

Step-by-step methods for creating and deploying AI agents.

→ https://cdn.openai.com/business-guides-and-resources/a-practical-guide-to-building-agents.pdf

5/ Building Effective Agents – by @Anthropic

Core principles for designing agents that are dependable and efficient.

→ https://www.anthropic.com/engineering/building-effective-agents

6/ Identifying & Scaling AI Use Cases – by @OpenAI

How to discover AI opportunities and expand them at scale.

→ https://cdn.openai.com/business-guides-and-resources/identifying-and-scaling-ai-use-cases.pdf

7/ Agents Companion – by @Kaggle

A playbook filled with tools and reference material for agent builders.

→ https://www.kaggle.com/whitepaper-agent-companion

8/ 601 GenAI Use Cases – by @Google

A comprehensive library of real-world industry applications.

→ https://cloud.google.com/transform/101-real-world-generative-ai-use-cases-from-industry-leaders?hl=en

AI Guides Worth Reading

🚨 @AnthropicAI just made Claude Code remote-controlled.

Meet Channels.

You can now:
→ Connect a persistent local Claude session directly to a Discord or Telegram Bot 
→ Message tasks to your bot on the go
→ Come back to finished code 🔥 

Here’s how to set it up:

1️⃣ Create a bot (via Telegram’s BotFather or the Discord Dev Portal) and copy the token
2️⃣ Inside Claude, run /plugin install telegram@claude-plugins-official (or discord)
3️⃣ Configure the token and relaunch your session with the --channels flag
4️⃣ DM your new bot, grab the 6-character code, and pair it to your session!

Message the bot to fetch files, read history, and execute code locally while you are on the go.

The flexibility is unmatched.

Time to spin up your bot 👀

Read here on how to setup Telegram: 
→ https://github.com/anthropics/claude-plugins-official/blob/main/external_plugins/telegram/README.md

https://github.com/anthropics/claude-plugins-official/blob/main/external_plugins/telegram/README.md

Full step-by-step Discord guide here: 
→ https://github.com/anthropics/claude-plugins-official/blob/main/external_plugins/discord/README.md

https://github.com/anthropics/claude-plugins-official/blob/main/external_plugins/telegram/README.md

Channels is rolling out as a Research Preview. 

If you want to explore the access controls, allowlists, and the underlying architecture, Anthropic has it all documented.

→ https://code.claude.com/docs/en/channels
https://code.claude.com/docs/en/channels

Unrolled thread here: https://typefully.com/DataChaz/eCMWzLr

Claude Code Goes Remote

With Voicebox, @ElevenLabs just lost its moat.

→ Powered by Alibaba's Qwen3-TTS for near-perfect cloning
→ Ships with a DAW-like "Stories Editor"
→ No cloud, runs locally on your machine

100% Open Source. 100% Local.

Link to repo in 🧵↓

It features a full-blown "Stories Editor" (DAW stylee!):

→ Drag & drop multi-track timeline 🎚️
→ Complex conversation mixing
→ Precise inline trimming

Perfect for creating podcasts or multi-speaker narratives locally!


Massive shoutout to @jamiepine for shipping this in open source!

→ https://voicebox.sh/

Mac & Windows builds are already available.

Don't forget to give a ⭐ on GitHub to support Jamie!
→ https://github.com/jamiepine/voicebox 

Unrolled 🧵
→ https://typefully.com/DataChaz/2Gi4J88

Voicebox: Game Changer for TTS

Looking for bulletproof security for AI agent access?

@DescopeINC just changed the game with their Agentic Identity Hub 🔥

→ MCP server authentication backing OAuth 2.1, PKCE, DCR, and CIMD
→ A secure vault stocked with 50+ templates and short-lived tokens
→ Strictly scoped, short-lived creds assigned to each agent
→ Deep tool-level permissions, user consent, and full audit logs

Let’s break it down 🧵↓

1/ 

First, let's look at Descope's MCP integration.

When giving AI agents access to external tools, securing that connection is critical.

Descope handles this using:
→ Full MCP server auth
→ OAuth 2.1 support
→ PKCE validation

Your agents only operate within strict, tool-level scopes, backed by a full audit trail.

2/ 

DCR is incredibly simple, but seamless shouldn't mean unsecured!

A client can discover the endpoint and initiate OAuth instantly, but you still hold the reins.

Before handing over any credentials, you can run checks against:

→ IP reputation
→ Geolocation
→ Custom logic

This lets you instantly flag every new agent as verified, unverified, or completely blocked 🛡️

Scaling distributed agents has never been this smooth, or safe 🦾

3/ 

Now here's the best part → CIMD 🔥 

Under the new MCP auth model, a client hosts a metadata.json file over HTTPS and uses that exact URL as its Client ID.

This means significantly less pre-registration and a fundamentally cleaner trust model.

That's a wrap 🙌

Jump in and test it out using their free tier:
→ https://descope.com/use-cases/ai

Thanks again to the Descope team for teaming up with me on this thread, I genuinely learned a ton putting it together!

Unroll'd version:
→ https://typefully.com/DataChaz/yYzcQm4

Descope Just Fixed AI Agent Security

ElevenLabs just lost its moat 🤯

Voicebox just dropped:
→ Powered by Alibaba's Qwen3-TTS for near-perfect cloning
→ Ships with a DAW-like "Stories Editor"
→ No cloud, runs locally on your machine

100% Open Source. 100% Local.

Link to repo in 🧵↓

I’ve been testing a new AI agent that actually remembers who I am.

Meet @getsurething.

It hits 59.3% on the `REAL-Agent Benchmark`, crushing both OpenClaw and ChatGPT.

And unlike them, it has Persistent Memory, SOC 2 security, and connects to 1000+ apps 🔥 

Let's dive in 🧵↓

1/ 

The biggest friction with AI? 

Amnesia!

SureThing is SOTA with 59.3% on REAL-Agent Benchmark
(ChatGPT is 15.1%).

With 84.8% Memory Depth, it beats @OpenClaw on context:

→ Recalls autonomously
→ Learns patterns
→ SOC2 Certified

Read the paper: https://www.surething.io/research

2/ 

Is that SOTA status just hype?

I connected my Gmail to see. 

@getsurething indexed my workflow instantly.

Instead of searching, I queried:
"What is urgent?" or "Did I miss a reply?"

It understands priority based on my history.

FINALLY, an AI agent that gets the context!

3/ 

It doesn't stop at emails.

I linked Google Calendar to test it.

@surething's agent cross-referenced my schedule against requests.

It automatically:

→ Checks availability 
→ Drafts invites 
→ Manages time zones

It handled the scheduling logic for me 👌

4/ 

The killer feat?

Proactive Execution!

surething.io works while you sleep.

I woke up to:
→ drafted replies
→ a scheduled meeting 
→ priority threads flagged

It checked my calendar, prepped the actions, and waited.

I just had to hit "Approve" 🔥 

5/ 

The integration stack is HUGE.

It connects to 1000+ apps:

- Slack
- Notion
- HubSpot
- Socials
... and more!

It creates a unified memory layer, connects the dots across your stack.

eg if I update a Notion doc, the agent knows about it when drafting a Slack message.

Ace!

That's a wrap 🙌 

SureThing is invite-only.

Btw, the onboarding is pretty unique. 

It interviews you to decide if you get an invite code.

Basically a reverse Turing test! 😅 

Go see if you can pass the interview → surething.io

Unrolled version of this 🧵 
→ https://typefully.com/DataChaz/WNK5ESI

Meet SureThing: Memory-Driven AI

ICYMI Google's MCP Toolbox for Databases is now open source 🔥 

A single backend for connections, security, and schema-aware SQL tools, built for AI agents.

Compatible with Python, JS, Go, LangChain, and more.

Repo in 🧵 ↓

1/ 

Features include:

→ Declarative tool definitions (<10 LOC)
→ Secure by default with integrated authentication
→ Auth, pooling & fast queries out of the box
→ Native observability (OpenTelemetry)
→ Postgres, MySQL, Cloud SQL & more supported


2/ 

Repo: 
↳ https://github.com/googleapis/genai-toolbox
https://github.com/googleapis/genai-toolbox

3/ 

Note: MCP Toolbox for Databases is currently in beta and may introduce breaking changes until the first stable release.

Full docs are here:
↳ https://googleapis.github.io/genai-toolbox/getting-started/introduction/

4/ Unrolled thread: https://typefully.com/DataChaz/68ymsiA

Google Open-Sources MCP Toolbox

Your @FastAPI app is powerful.

But imagine if you could turn your endpoints into agent-ready tools.

Now you can, with just 3 lines of code via @makhlevich’s FastAPI-MCP! 🤯

It exposes your endpoints as MCP tools, ready for agent use.

Free and open source too (repo in 🧵) ↓

Here’s why it changes the game:

✅ Native to FastAPI → Works with the auth system you already built.
✅ Zero settings → Just plug it in and it’s live.
✅ Saves your existing work → Models, responses, docs: all preserved!
✅ Blazing fast & adaptable → Mount or run standalone with pure ASGI performance.

The code snippet you need:

# pip install fastapi-mcp
from fastapi import FastAPI
from fastapi_mcp import FastApiMCP

app = FastAPI()

mcp = FastApiMCP(app)

mcp.mount()

More about `FastAPI-MCP` here:

- repo: https://github.com/tadata-org/fastapi_mcp
- docs: https://fastapi-mcp.tadata.com/getting-started/welcome

Unroll'd → https://typefully.com/DataChaz/A2LLsaJ

FastAPI-MCP: Turn Your Endpoints into Agent-Ready Tools!

So cool.

@Tiny_Fish just shipped Mino, a way to programmatically interact with *any* website through browser automation and AI-powered navigation.

Give it a URL and a goal in plain English, it does the work and returns structured JSON.

No API? Not a problem anymore 👊 

🧵↓

1/ 

Just watch Mino casually running on 100+ Japanese hotels and fetching prices for Jan 15-16, 2026.

5 hotels done in less than 90 seconds.

That's 18 seconds per hotel 🤯

Imagine what you could build if you had 100+ Minos working on hotel websites? :)

2/ 

Mino's workflow is brutally simple:

→ Playground to validate
→ Batch Upload to scale
→ API to automate

Here you can watch it handle urgent care availability across 14 sites in parallel.

Different layouts, but same output schema 🔥

3/ 

Check out Gemini tries the same task and fails.

Wrong clicks.

No progress.

Why?
→ Needs vision inference every scroll
→ Rethinks each screenshot
→ Zero determinism

Mino learns the workflow once, then runs it like code.

I’ve barely scratched the surface in this thread.

With Mino you get:

→ 85-95% success rate on complex workflows
→ Parallel execution across multiple sites
→ Structured JSON outputs every single time.

With the link below, you also get 50 complete runs FREE:
→ https://www.tinyfish.ai/?utm_source=charly&utm_medium=x&utm_campaign=dec15

Scaling Real Web Workflows with Mino AI 🔥

Google's MCP Toolbox for Databases is now open source 🔥 

A single backend for connections, security, and schema-aware SQL tools, built for AI agents.

Compatible with Python, JS, Go, LangChain, and more.

Repo in 🧵 ↓