Tomorrow is ship day!

Pablo Picasso said: 'The meaning of life is to find your gift. The purpose of life is to give it away,' I think he meant OSS.

hello and welcome to Sketchy v0.2! It's got at least one Linux user now (thanks for the bug fix, Daniel!), so things are getting hot in here! This time, I've implemented some more widgets suitable for implementing a chat app. Enjoy!

My process when coding with AI agents: Spec: Work w/ the agent to create a plan. Make: Agent writes the tests and the code. Inspect: Check the test results, the static analysis output and the app's features and UX. Tune: Fix the P0 issues. Hope: that your stakeholders appreciate your hard work!

I tried @elevenlabsio streaming text-to-speech API this morning prepared for a big lag between pressing the Play button and hearing anything from my laptop speakers. Boy was I wrong. I couldn't count all the way to "1" before I heard Laura talking to me. Amazing! #Recommended

This is the uncompromising successor to provider that I've been looking for: a pleasant developer experience, a simple place to start and an excellent tutorial. I'll be moving to riverpod 3 for my future Flutter projects. Thanks, Remi! riverpod.dev/docs/tutoria...

Right now I'm sitting on the couch with my phone designing an "Agent Board" to control the set of coding agents on the computer I gave them to play on: an M4 mini on the Internet so I can manage my coding agents when I'm sitting on the couch with my phone.

ChatGPT nerd mode gets me: "Yep, that’s even better: you get to be lazy and principled at the same time. My favorite combination."

Approaching your LLM with a problem to solve instead of a solution to implement files in the face of modern AI coding best practices. Instead of giving your coding agent a tight spec, starting broad and carving it down to specifics works better with creative LLMs like G3 and Sonnet.

I'm please to announce the release of Sketchy, a hand-drawn, xkcd-inspired design language for Flutter on mobile, desktop, and web: pub.dev/packages/ske... It's my first design system, so your feedback is very welcome!

Google has become the Apple of AI with its expensive but end to end high quality hardware to software to data solution. Nvidia is the come one, come all hardware+framework (CUDA) platform; the Android of AI, if you will. Apple is the Apple of computers and phones but not AI.

As an engineer, I ask every customer coming to me with a solution to step back and talk about the problem first. Likewise, I’m finding that I get better AI results starting with the problem. And by “better,” I mean sometimes crazy better ideas I’d have never come up with on my own.

I'm listening to The Last Economy inside Eleven Reader and loving both. #recommended

The only problem I'm having with Gemini 3 is that I want more! The usage is so high that I'm getting outages. G3 pushes me to be more ambitious, ruining me for other models. It's creative like Claude, competent like Codex but faster than hell. Gemini 3 completes me.

Gemini 3 may be more expensive if you're measuring cost/token. But if you're measuring cost/task and G3 can do things in only a few turns that take the others 2-5x (which is what I'm seeing with my coding tasks), then Gemini 3 is faster, higher quality AND cheaper.

OK, after going around and around with gpt-5.1-codex for an hour, gemini 3 one-shotted it. I was saving my limited access for special occasions, but now g3 is all I wanna use.

I've never built a design system before, but I like how this one is coming along for Flutter.

One of my fun programs ever was figuring out the various parts of a typeface, which I could only do by writing the WinForms program (it was another age : ).

After losing Mint and Personal Capital, and now Empower is moving my cheese again, I needed a stable place to track my accounts. Then I found @SimpleFin (www.simplefin.org). And since I wanted to build my own apps, I built simplefin_dart: pub.dev/packages/sim... Enjoy.

Has anyone tried Zed? I resisted it but am really liking it, especially how they do AI coding agents via the Agent Client Protocol. If it supported Flutter the way VSCode did, I think I'd switch.

dartantic_ai v1.3.0 release @ pub.dev/packages/dar...: - Anthropic thinking support exposed like the OpenAI Responses provider, so you can write your code to look for thinking output regardless of the provider - Ollama typed output support - Mistral usage tracking

I shipped dartantic_ai 1.2.0: pub.dev/packages/dar... - Migrated Google provider to the new Google-supported package, fixing some low-level response issues. - Added Google "double agent" orchestrator to support typed output with tools simultaneously. - Added llms.txt for LLM readers. - And more!

Been using AI coding agents for a while? Then watch this video to have Lada Kesseler turn your niggling suspicions into "I knew it!"s: www.youtube.com/watch?v=_LSK... Lada has pulled out the Gamma-style patterns out of coding agents. #recommended

Interested in how dartantic does all of the cool things it does? Check out the wiki: github.com/csells/darta...

Dartantic v1.1.0 is out and this is a big release! New OpenAI Responses provider, including session persistence (aka prompt caching), intrinsic server-side tools, and thinking metadata streams. Thanks to @jezell.bsky.social for his openai_core package! Check out dartantic.ai for the deets.

Interested in the core issues in building a real-world AI coding agent? Start with @pvncher's amazing blog post: repoprompt.com/blog/buildin...

I'm busy working on a new dartantic release and as part of that, Sonnet 4.5 runs examples like this: `DARTANTIC_LOG_LEVEL=FINE dart run server_side_vector_search.dart` This was surprising, since dartantic had no environment check for a logging level. But now it does. : )

Been spending a bunch of time with Gemini 2.5 Pro for coding this weekend. It's about 80% as smart for coding as GPT5 and Opus 4, but when I use it, I can't juggle between multiple agents on different projects: 2.5 Pro is too fast! Not a bad tradeoff to be honest.

Holy cow. Somebody thinks that I'm in the top 1% Dart engineers in the world. I do love Dart! algora.io/profile/csells

I used to get up in the morning and deal with my messages. Now I check the progress of my AI agents and direct them to the next step, while my messages go unmanaged. AI just feels more creative and fun.

TIL not to use AI to code a protocol client and server in the same project at the same time, as it will surely route around the protocol restrictions to get things working. Instead, build the new server/client against known working implementations first. This is the way.

Want to know how to get the best from agentic coding agents? Matt's video represents the best explanation for getting solid results that I've seen (aka this is what I do so it must be right : ) www.youtube.com/watch?v=dBWn...

I think I said to an AI coding agent today: "you're fixing the test to not require the thing we're testing for? why would you betray me like that? do you hate america?" the reply: "Ha — I promise I don’t hate America. I hate flaky tests." who says you can't make work fun? : )

Cline is continuing to dig into solving real-world problems with heavy duty AI agents: cline.bot/blog/cline-v...

"We aim to emphasize high-value tokens: where every step in a plan is deliberate, relevant, and positioned where the model can use it. By anchoring the agent on what matters most, we make better use of every token — and focus the model’s attention." cline.bot/blog/focus-a...

Welcome to dartantic_ai 1.0! - 11 providers (Google, OpenAI, Anthropic, ...) - Tool calling with typed I/O - Runs everywhere Dart runs sellsbrothers.com/welcome-to-d...

I've had some early experience with the Dart MCP Server and it's got a ton of potential for attaching AI agents to your running Flutter apps: medium.com/flutter/supe...

I have discovered that I am as picky as any artist with their work. Dartantic 1.0 was functionally complete and passing >1000 tests across 11 chat model providers and 4 embeddings model providers three days ago. But I couldn't ship it because it wasn't *right*.

The 5 Stages of AI Grief: sellsbrothers.com/the-5-stages...

"If something must happen, please don't use a stochastic process to do it. Eventually, due to the nature of sampling from a probability distribution, it will fail to do the thing that must be done." @domesticmouse.com

I've been talking to the Cactus folks about getting their local LLMs for mobile into dartantic. In the meantime, don't let that stop you from bringing them into your mobile Flutter apps: medium.com/@shemet0roma...

I'm happy to announce flutter_ai_toolkit v0.9.2. This release slays the remaining bugs on the road to v1.0 plus: - multi-turn tool call support - support for uploading attachments to the cloud - overriding the speech-to-text conversion pub.dev/packages/flu...

I was on an AI-focused podcast last week about how to think about AI as a new software engineer. Take a look: youtu.be/AgSHvUgxio0?...

Got tools and/or an output schema that your LLM isn't paying attention to? Then you need the System Prompt Optimizer! It's based on dartantic_ai, so you can plug in Google and OpenAI models today, with many more coming soon! github.com/csells/syste...

Does your business bank on Flutter? Check out the Flutter Funding Committee proposal to keep core Dart/Flutter packages thriving! sellsbrothers.com/critical-flu... #Flutter #OSS #Dart

Me: "F!$# dude. When I said all, I meant ALL!" Claude 4 Opus: "You're f*$#ing right! This is some bulls*$#!" I think I may be rubbing off on my AI...

OH from the park: Kid #1: "How do you take someone's brain?" Kid #2: "With my chainsaw! (maniacal laughter)" Some things never change : )

This is my way: www.reddit.com/r/ClaudeAI/s...

I've now built my custom Dart lint rule as a defense against AI-generated code. I feel like this is the start of something... github.com/csells/excep...