Niall O'Higgins
@niallohiggins
🔨Eliminating toil with AI @stepchangelabs. Ex-@thegp, @coinbase, nextdoor, metaweb, openbsd.
Can AI Beat a Human Dev? Google Jules vs OpenAI Codex on Live Django/Postgres Issues with @StepChangeLabs APX ⏱ Codex 3 min, Jules 4 ⚠️ Both drop indexes in django migrations but leave db_index=True in code - something a human Django engineer would never do.
So cool how you can just LoRA things & share.
As promised, I’ve open-sourced the Tattoo LoRA! If you’re a dev who wants to integrate it into your app, or you just wanna see how a tattoo would look on yourself, feel free to use it however you want. Would appreciate it if you give credit too ofc. 🖤
Docker desktop on mac has been so enshittified over the years, it's a bit of a sad sausage & resource hog. I mean, just let me run containers on my laptop pls. Switched to orbstack.dev and liking it so far.
OSS coding models are having a good couple of weeks.
Open source models 📈 qwen3-coder is available in Cline
This is incredibly cool.
Seedance 1.0 is so great at making images alive. it's 1080p and ~$0.20 per video and generates in a couple minutes on @FAL Here are some image to video examples + how you can use it for free 🧵
This is cool but of course we want the new Qwen3Coder model!
> 2500 tokens per second > 40x faster than Sonnet-4 > 4 trillion transistor wafers (19x Nvidia B200) These are impressive numbers, but nothing beats a live demo. 1 million free tokens per day of lightning-fast qwen3-32b and qwen3-235b from @CerebrasSystems in Cline. Link below.
There’s basically unlimited demand for AI generated video techcrunch.com/2025/07/23/you…
There is so much AI generated content already in the mainstream. Just look at the Apple Stocks news articles for $CPRT. Society is not ready for where this is headed. All text and video content is going to be 100% personalized to you.
MSFT just launched a Lovable clone built into Github. Instant distribution and integration with existing customer workflows.
Today we’re releasing GitHub Spark — a new tool in Copilot that turns your ideas into full-stack apps, entirely in natural language.
I think @charliebholtz's conductor.build is a glimpse at what a re-imagined, post-IDE developer UI might look like. Directly editing code is no longer a primary activity, but rather managing todos, tests and a lot of AI coder bot histories.
It was nice of Google to OSS gemini-cli and so now there’s a proliferation of Claude Code clones running on open models like Qwen Code github.com/QwenLM/qwen-co…
Nice to see Copilot really trying to improve. More competition (Cursor, Windsurf, Cline, Copilot) = good for devs.
A week and half ago we asked you for feedback on Copilot and what we could improve in @code. Hundreds of you responded. Here is an update on what we've done since then regarding your feedback...
"App that builds itself" feels like a natural paradigm for CRMs.
the notes app that builds itself just got a sidebar you can hack make it yours, change it however you’d like now everyone gets their dream notes app
Solved problem at any engineering organization with > 1 engineer.
Lots of folks talking about running multiple code agents in parallel. But there's a tension between the desire for parallelism and most kinds of feature work, which are sequential. How are you leveraging parallel agents WITHOUT an absolute ton of merge conflicts?
I believe the UX around managing AI tools (even just long-running LLM chat sessions) is in its infancy. A lot of the difficulty is simply due to trying to context switch between huge text buffers in rather janky tools.
I’m slowly beginning to accept that my productivity, when working with AI coding agents, is limited by my human brain. AI can do many tasks in parallel, but I can only track the context of a few, so I only run a few tasks at a time. I am the bottleneck.
Cloud Coderbot APIs are going to be huge. It is surprising there aren't really easy to use ones yet.
Asynchronous @julesagent tasks done synchronously Concurrency looks cool Orchestration is cool #soon #buildinpublic #agenticsdlc
Claude Squad: Nice tmux wrapper application for managing multiple coder bots (supports @claude_code, gemini-cli, @OpenAI Codex) running in parallel on a single repo. github.com/smtg-ai/claude… Uses git worktrees. Wishlist: - Support multiple repos at once. - Support auto-running…
I agree IDEs can get in the way. This is why CLI layers like @claude_code are so nice. They bring the full power of the LLM to your UNIX environment (where everything is a command/file/pipeline), rather than a heavyweight IDE which lacks the composability of UNIX.
Prediction: I think a large number of leading-edge devs will stop using IDEs in the next 12 months. I've always used an IDE because I like the UI, but I find that for agentic coding, there's now a huge amount of cruft in the IDE UI that just slows me down and slows the agent…
Another reason why a transparent OSS coder agent like @cline which isn't coupled to an inference provider is an interesting option. They don't charge you for inference so not going to bait-and-switch you on usage with a subscription.
Anthropic tightens usage limits for Claude Code – without telling users | TechCrunch techcrunch.com/2025/07/17/ant…
TIL you can take a screenshot on mac to your clipboard, paste it into your terminal (CTRL+V) and @claude_code can read it.
Evaluating @cline + MoonshotAI Kimi K2 Instruct on @GroqInc. Interesting to compare with @cursor_ai and @claude_code. Firstly, I'm impressed by Cline. It has a very transparent UX in terms of how it is using the LLM and the costs. I like the design. Secondly, inference is…
