AI Engineer Guide
Articles about AI Engineer Guide
Pi Durable Handbook
There’s a handbook for Pi Durable. If you prefer to read them. It has good diagram and stuffs but it is little verbose. Pi Durable is awesome! Asked Opus 5.5 to make a deep technical manual on it. It’s really good and the diagrams work really well to explain things 🤌 https://t.co/SBjAR5Zcxj https://t.co/R9KKka2pUI — Caleb (@calebfahlgren) October 5, 2026
ACP support in Command Code
Command Code now has support for ACP - Agent Communication Protocol. You can now use command code via IDEs and tools that support ACP like Zed How to get started? cmd acp https://commandcode.ai/docs/acp https://x.com/CommandCodeAI/status/2107473388184936786?s=20
Traits of Slop UI
Came across this blog post on UI slop. If you look around, you’ll also be seeing these things everywhere. Common traits are: Gradients Too many random colors Badges - even if it doesn’t make sense Misalignment - I see that even in apps like Youtube sometimes :( Too many emojis Generic taglines You can read more about it here: https://godobject.dev/blog/10-tells-of-slop/
Cloudflare Web Search API
Cloudflare now has support for web search through their AI gateway. Right now it uses the following providers under the hood like Exa, Ceramic, Linkup with zero data retention. You can make request like this curl https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/ai/websearch/ \ --request POST \ --header “Authorization: Bearer $CLOUDFLARE_API_TOKEN” \ --header “Content-Type: application/json” \ --data ‘{ “query”: “What is the latest model from OpenAI for coding?”
Clef - Jev like decision model from Cloudflare
Cloudflare has released Clef - a decision making model similar to Jev. And it performs as good as Jev in their benchmarks. Unlike Jev, this model supports image/vision as well. And its available in their Cloudflare Worker AI And its Apache 2.0 licenses and weights are available in HuggingFace It supports Typesafe AI Jev like api
Decision models in llama.cpp
Llama.cpp has support for the decision models which you can use in the dedicated endpoint /v1/systemone It follows Typesafe.ai API format curl http://localhost:8080/v1/systemone \ -H “Content-Type: application/json” \ -d ‘{ “state”: “Customer message: I was charged twice for my order last week and nobody has replied.”, “questions”: { “route”: { “type”: “choice”, “instructions”: “Which team should handle this?”
Upsy.ai - Give Browser in cloud for your AI Agents
Upsy.ai by Upstash gives you a dedicated cloud computer with browser, connectors, memory, etc which you can wire up with any AI agents via MCP. For now it’s free. And under the hood uses Box feature by Upstash. Once you connect it with your agent/assistant, it can interact with browser and other tools. Here is an example that I asked it to find a bus ticket pricing
Cloudflare’s Agentic CLI
Cloudflare has released cf cli. I really like their search command. You can say something like: cf search “your problem statement here” and it’ll get back with the results Introducing "cf", an agentic CLI for the entire Cloudflare API. You can do nearly anything. Even buying a domain is incredibly simple from the CLI. Don't know what command you need? Run `cf cli search "look at messages stuck in my queue"` and it will return a list of relevant… pic.twitter.com/lif3KVoumP — Brayden (@BraydenWilmoth) September 28, 2026
Native MCP support in Pi Coding Agent
MCP is now supposed out of box in Pi Earlier we had to install a third party plugin to make it work which is not needed anymore. Reference https://earendil.com/posts/you-said-no-mcp/
Embedding based Emoji Search
We all know that Jev is being used for everything that is a classification based problem. But the problem is it is used for the sake of using it. And sometimes it may be overkill or might not be right tool for the job. Recently Linear team has used Jev for their emoji search feature. However Max has a interesting post on why a small model like bge-small is more than enough for such use cases in terms of accuracy, cost, latency, etc. https://maxleiter.com/blog/embedding-emoji-search
GLM 5.3 on Mistral API
Z.ai’s GLM 5.3 is now available on Mistral’s API platform It’s a good call by them, serving powerful open weights so that anyone who is worried about the privacy and things can use them instead. https://x.com/MistralDevs/status/2103896309493080336?s=20
Zed - Single No AI toggle
Almost all the IDE are bloated with AI feature (VS Code and it’s forks) If you prefer to not have a any AI feature, you can toggle it off using a single switch in Zed A single setting removes every AI-related feature in Zed. Not a VS Code fork, btw... https://t.co/O42VUB2WhZ pic.twitter.com/dp1LVb2czW — Zed (@zeddotdev) September 25, 2026
Anthropic’s developers blog
Anthropic team has a developer blog claude.dev where they post interesting engineering posts like this https://claude.dev/blog/how-we-made-claude-ai-faster/
Fine Tuning your own Jev on Together.ai
Hassan has a detailed post about fine tuning your own Jev (just in case if you need to) using Together.ai It is fine tuned on top of Qwen3.5 4B. https://x.com/nutlope/status/2102881280115249597
What People are building with Jev
There’s a directory for Jev use cases Ship With Jev Some of the projects looks interesting which you can use it as reference and obviously there are lot of hypes as well.
ossrules.md - Collection of AGENTS.md for reference
Ben Vinegar has recently launched ossrules.md a collection of AGENTS.md file from leading open source projects. 📜 introducing ossrules.md - learn how leading OSS projects use agent rule files - discover skill files that projects actually use - level up your own projects and workflows I’ve personally learned a lot just exploring pic.twitter.com/VvrZkV810X — Ben Vinegar (@bentlegen) September 18, 2026 You can also let your agent improve it’s AGENTS.md prompt by prompting this
Jev use case - Malicious link scanner
One of the best use case for Jev is using that to scan for malicious links if you’re getting input from the user For example, here is an example by Steven Tey who has used Jev to built url scanner for his short link service dub.sh The way it works is by passing the needed context in the prompt (state) to Jev and get the output from it. Based on it we can take a call whether it’s malicious or not.
Jev use case - Picking right model automatically
One of the interesting use case for Jev is figuring out the right model for the action you are trying to do based on the prompt Here’s a quick example by Kshitij because Jev is an instant classifier, you don’t have to pick the right model anymore. it just happens. pic.twitter.com/mazhn2TyF1 — Kshitij (@okkshitij) September 20, 2026
Jev Ultrafast - Really fast Agentic browser
jev-ultrafast is an agentic browser which under the hood uses Jev model to figure out the action. And only if it is needed for TYPE_TEXT operation it uses the LLM. Rest of the things are handled by Typesafe.ai’s Jev model In the demo video, they were able to book flights using it in less than 7.1 seconds from Zürich → London on Google Flights https://github.com/browser-use/jev-ultrafast
Typesafe.ai’s Jev on Cloudflare AI Gateway
Jev by TypeSafe is a model that is focused on solving structured evaluation really well and cost effectively. For example, instead of using regular LLMs for classifing support tickets, you can consider using it to get the fast + better output with cheaper price. Input tokens: $0.042 / MTok and Output tokens are free. It is been going crazy on X and people of building lot of cool things with it.
Install skills directly from Notion page
skills cli now supports directly installing via notion url. No need to worry about having to commit your skills to github and things like that. This notion based approach will be useful for non tech folks! How to use it? 𝚗𝚙𝚡 𝚜𝚔𝚒𝚕𝚕𝚜 𝚊𝚍𝚍 Under the hood it’ll use notion cli. If you don’t have it installed already you can do it using this command curl -fsSL https://ntn.dev | bash ntn login Reference https://vercel.com/changelog/skills-cli-notion-skills
Union Alpha stealth model on OpenRouter
Usually the model providers gives model inference for free during their alpha period however they will use you prompts for their further training. But this time that is not the case. Union Alpha which is available in openrouter, opencode, etc is just giving away the inference without any catch. They don’t trun on your data as well. It’s a multimodel and it claims to be good in agente coding and workflows 256k context window. Pricing and other things an not announced yet And this might be from the Chinese lab as well
Inspo MCP - Design inspiration MCP for AI Agents
Inspo MCP let’s you get some design inspiration from a live websites for your AI Agents How to get started? npx inspo-mcp install Reference https://x.com/nutlope/status/2099547343112564921
How to configure sub-agent model in Claude Code
Claude Code let’s you to customize the sub-agent model. This is especially useful if you’re using some powerful model like Fabel 5.1 You can just do it by setting the env variable CLAUDE_CODE_SUBAGENT_MODEL=opus By setting this, any new sub-agents that spawns will use that configured model instead. Reference https://x.com/lydiahallie/status/2099195114115707210
px0 - Lightweight Read Only IDE
Arpit Bhayani has built a lightweight read-only IDE in golang. The goal for this project is that, it is primirly intented for code review or reviewing AI generated code. Unlike VS Code/Cursor or simililar IDEs it does not take huge memory when you run it. And that’s it’s USP. How to get started? It is just a single binary curl -fsSL https://px0.ai/install.sh | bash https://x.com/arpit_bhayani/status/2099124464365756672