David Sánchez

Engineer by trade. Builder by instinct. Bringing ideas to life.

[David Sánchez @d4vsanchez]

I feel Codex limits are depleting significantly faster than they were last week 🤔

[David Sánchez @d4vsanchez]

@cardotrejos Recomendado seguir a @abuiles, está haciendo cosas muy pro en top de CF

[Freddy Montes @fmontes]

Un buen SKILL.md no se escribe. Se borra.

Claude te va a escupir 200 líneas. El skill bueno son las 20 que quedan después de borrar todo lo que el modelo ya sabía.

Mi loop después de escribir varios para producción:

→ Primera pasada, le digo: "Quita la sobreingeniería. No le enseñes lo que ya sabe."

→ Agarro un café y leo línea por línea. Si una frase no cambia el comportamiento del agente, fuera.

→ Pruebo en escenarios reales, no en el ejemplo que el modelo me dio,

Si tu skill no te está dando resultados, no es que los skills no sirvan. Es que dejaste a Claude escribirlo y le creíste a la primera.

¿Cuál fue el primer skill que te dio resultados reales? Curioso de ver qué están construyendo.

[Oisin @OisinO]

My company, Recharge, just acquired Skio for $105m. This is the largest private acquisition in the space ever.

Just 5 years ago I got a crazy phone call that nearly killed us...

Here’s the story I've never told anyone before:

[David Sánchez @d4vsanchez]

I think Claude Code changed its behavior when using the `--resume` or `--continue` flags.

It's now opening the last session regardless of the current working directory, I think it was previously showing and continuing the sessions from cwd only.

[David Sánchez @d4vsanchez]

Claude Code 2.1.120 has the resume feature completely broken 🙃 @bcherny

[David Sánchez @d4vsanchez]

@wesbos @internetofshit had to switch to SteerMouse because the Logi app was utterly bad. I realized that when my battery usage went up, it was mostly because it started consuming over 100% CPU doing god knows what.

[Wes Bos @wesbos]

Logitech's servers are down, so LogiOptions+ doesn't load, which means my mouse no longer works. @internetofshit

Took 5 mins and ported everything over to Better Touch Tool. Uninstalled that awful app.

[David Sánchez @d4vsanchez]

I'm currently testing T3 Code, and oh boy, it's very well crafted, years ahead of Claude's official app.

I'm just wondering one thing. Is there a reason for the sessions created in T3 Code to not appear in Claude Code (CLI) list when using `/resume`, or just Claude quirks? @theo

[David Sánchez @d4vsanchez]

@fmontes @orca_build @hector_reyes_ Estos tweets son oro puro!

[kate @whoiskatrin]

we’ve just released memory for agents

primitives in Agents SDK when you need them, managed when you don’t

[Claude @claudeai]

Introducing Claude Design by Anthropic Labs: make prototypes, slides, and one-pagers by talking to Claude.

Powered by Claude Opus 4.7, our most capable vision model. Available in research preview on the Pro, Max, Team, and Enterprise plans, rolling out throughout the day.

[Felix Rieseberg @felixrieseberg]

Hi! I'm here with *another launch*, it just happens to be extremely niche, nerdy, and probably only for a handful of people.

In the desktop app, Claude Cowork and Code now have a little Bluetooth API for makers & developers, allowing you to build hardware devices that interact with Claude.

I, for instance, built a little desk pet that alerts me whenever Claude is waiting for permission.

[ClaudeDevs @ClaudeDevs]

We fixed a bug where rate limits on Claude subscriptions weren't properly adjusted for long context requests in Opus 4.7.

We've reset 5-hour and weekly rate limits. Enjoy Opus 4.7!

[David Sánchez @d4vsanchez]

Totally agree. I've seen this more and more, watching @abuiles' projects.

[Ashley Peacock @_ashleypeacock]

Today is the point where Cloudflare becomes the full cloud platform. Agent Week day 4 is packed:

👨‍💻 Git-style, versioned code storage (Artifacts)
✉️ Send emails with 1 line of code
🧠 One API to access any AI model (AI Gateway)
🔍 Better RAG: Hybrid Search, Relevance Boosting & more
🌎Streamlined PlanetScale integration inc. billing

Artifacts are a big deal for humans & agents - versioned Git repos on demand.
Email Service plugs the final hole in the developer platform.

Here's the breakdown:

👨‍💻 Artifacts: Versioned Storage that Speaks Git 👨‍💻

The surge of AI coding agents has had effects on the software industry far and wide. The amount of code being written and stored is increasing exponentially, and existing solutions just aren't built from the ground-up with agents in mind.

That's where Artifacts come in - a brand new product from Cloudflare that allows you to create an unlimited number of objects to store your code, versioned and Git-compatible. It's releasing in private beta today.

At a high-level, Artifacts will support the following from day one:

- Create new repos with scoped tokens
- Fork a repo
- Import an existing repo (e.g. from GitHub)
- Manage repos via bindings or HTTP
- Compatible with the Git CLI

This will allow you to create as many repos as you need. An agent, or human, can import a repo straight into Artifacts, add the remote URL and git clone to their local storage. Make changes, commit and push the changes to Artifacts.

There's pricing available too, the first 10,000 operations per month are free on Workers Paid, then $0.15 per additional 1,000 operations. Storage is charged based on gigabyte-months, the first 1 GB per month is free + $0.50 per additional GB-mo after.

The development flows you're used to, built for the scale of the modern AI era. I'm very excited to try this one out!

✉️Email Service Public Beta ✉️

One of the gaps in the Cloudflare platform has always been the lack of email sending capability. Yes, you could use external providers, but everyone yearned for a Cloudflare-native email solution, and it's now available to all paid customers in beta.

Email Routing has existed in Cloudflare for a while, which allows you to programatically handle inbound email using a Worker. It's really quite clever, with a few clicks to setup in the dashboard, and a single method definition in a Worker for setup.

You could send responses to inbound emails using Email Routing too, but that restricted it to use cases such as handling support cases or lead ingestion.

Today, that changes. Email Sending is now available to send transactional emails, both via a new binding as well as REST API.

Setup is easy, you just onboard a domain in the dashboard, add the email binding, then you can send emails with a single line of code. Keep in mind that only transactional use cases are allowed, so you can't send marketing emails as of today. You must be using Cloudflare DNS to use Email Service.

On the free plan, you can only send emails to verified emails in your account. You'll need a paid account to send emails more widely, this is to prevent spam.

I've used this personally during the beta and it's effortless to use, and I know it's going to make a lot of people happy!

🧠 AI Gateway: Route Across LLM Providers Without Managing API Keys 🧠

I've posted about Cloudflare's AI Gateway quite a few times, as it's such a good product and it's insane to me that it's completely free to use.

You simply sub in an AI Gateway URL instead of a specific provider, and get a ton of features right out of the box.

You can then route to pretty much any LLM provider, cache at the gateway, get insights into billing and requests, implement retries and even dynamically route requests (e.g. free plan to one model, paid plan to another or fallback to a secondary provider if the first fails).

Starting today, you can call pretty much any LLM provider using the same Workers AI binding that natively-hosted Cloudflare models use - without needing an API key for individual providers.

Cloudflare will bill you for the usage, and use their own API keys for the providers to make the calls. There are 70+ models across 12+ providers available right now from all the names you'd expect.

Additionally, in the future, you'll be able to run AI models directly on Cloudflare that you have built. This will be done using Replicate’s Cog technology to help you containerise machine learning models - all using bindings and APIs as you've come to know with Cloudflare.

Recently, Cloudflare added Kimi K2.5 to its roster of hosted models. There's a deep dive available today too into how this was achieved, as it had some unique challenges due to the size of this particular model - recommend reading it on the blog!

🔍 AI Search: Hybrid search, relevance boosting & direct content upload 🔍

AI Search allows you to create knowledge bases for your AI agents to use. You supply it with data sources (e.g. from R2, from a website URL) and it handles everything needed to provide RAG to your AI agent, or even non-agent use cases like website search.

This makes it effortless to ingest data, as AI search handles storage, chunking, indexing and more. On top of that, it naturally provides an interface to query data with numerous options to customise the search, and you can even have a model generate a response based on the retrieved data.

From today, you have more fine-grained control over how data is supplied to AI Search. You can still use data sources such as R2, but you can now ingest data directly using the new Items API. This makes it trivial to setup ingestion of your own data sources, rather than relying on Cloudflare adding connectors for it or using R2 as a middle man.

Furthermore, there's a new ai_search_namespaces binding that allows you to create, modify and delete AI search indexes programatically. The AI Search binding that lets you query your instance also supports cross-instance search, so you're no longer confined to searching a single instance.

On the search side, there are two new ways to modify how your searches are handled.

Hybrid search combines vector search with BM25 keyword search. Both types of searches are executed, and the results merged together. You can configure the tokenizer, keyword match mode and how results are combined.

Relevance boosting allows you to modify how a search is handled based on document metadata, potentially putting more weight on certain results. For example, you can prioritize recent documents by boosting on timestamp.

🌎 Direct PlanetScale Integration in the Cloudflare Dashboard 🌎

Cloudflare and PlanetScale are deepening their partnership to simplify how developers connect to external MySQL & Postgres databases.

Users can now create and manage PlanetScale Postgres and MySQL databases directly through the Cloudflare dashboard and API. Starting next month, billing will be unified, allowing self-serve and enterprise customers to pay for their databases via their Cloudflare account. This integration also supports Cloudflare credits, such as those from the startup program or committed spend, for database usage.

That's it for day 4, what do you think of the announcements today?

[David Sánchez @d4vsanchez]

@bcherny It's very neat. Only issue right now is a noticeable lag when sending a file to the context with `@`. The "autocomplete" popup gets slower the more characters of the filename you write.

[David Sánchez @d4vsanchez]

I like the new Claude Desktop

[David Sánchez @d4vsanchez]

@felixrieseberg I always found weird that Claude Code Desktop didn't had the sessions from Claude Code CLI. Does this new version merge both?

[David Sánchez @d4vsanchez]

How is it that it takes MINUTES to check whether there's a new version of Claude Code when running `claude upgrade`?