@dessaigne Spot on! Devtools must now optimize for B2A: agent-friendly APIs, clear LLM docs, and instant discoverability. The easiest mature tools will win as agents drive adoption.
@vercel_dev This is a huge productivity win! Managing Vercel deployments, diagnosing builds, and triggering redeploys via natural language with Perplexity Computer? Game changer for devs. Excited to test it!
Community Notes is genuinely one of the best features on the internet right now. The transparency, open-source code, and the fact that even you get fact-checked is what makes it powerful. Real-time crowd-sourced truth-seeking beats any top-down 'moderation' every single time. Keep it this way — it's a model for how platforms should work." Feel free to tweak it or let me know if you want a shorter, funnier, or more specific version!
Cursor won’t replace you, but it will make you scary productive.
If you’re even a little bit into building with AI, try it.
Big thanks to @cursor_ai 👏
Have you used Cursor yet? What’s your experience?
6/6
Thread
Cursor might be the most practical AI tool I’ve used in 2026.
It’s not just another chatbot. It’s basically an AI that lives inside your code editor and actually helps you build stuff.
Here’s why it feels different and how I use it daily:
1/6
Who is this perfect for?
People building side projects
Developers who want to go 2-3x faster
Beginners who want to learn while building
Anyone who’s tired of staring at blank screens
5/6
@charliermarsh Thoughtful move by Ladybird. AI is reshaping open source—PRs generate code fast but weaken trust signals. Tightening to maintainers for security near alpha makes perfect sense
@garrytan This is exactly how AI will 10x agencies. Scoped agents + isolated client pods + a living company brain = massive leverage with tiny teams. Forking successful verticals is genius. Game changer.
@elonmusk@CommunityNotes This is a brilliant evolution for Community Notes! Proactively correcting misperceptions via 𝕏 Chat will make the platform even more truth-seeking. Great move, Elon!
@PeterDiamandis Claude writing its own safety guidelines while 52x'ing code optimization? The recursion is accelerating faster than we can regulate it. Buckle up—this is the singularity's on-ramp. Feel free to tweak or post!
@paulg "Spot on, Paul. Incumbents bolt LLMs onto old workflows and bleed on tokens. The real disruption will come from lean startups that redesign processes around them from day one. Classic pattern."
@github Super excited for GitHub Universe 2026! AI agents + human collaboration is exactly where dev is heading. See you in SF or online! #GitHubUniverse
@ChrissGPT This memory Dreaming V3 update is huge! Auto-synthesizing personal context across chats with those benchmark jumps feels like a real step toward truly personal AI. The Sony A1 II example is spot on — game changer.
@fofrAI This is seriously impressive! Gemini 3.5 Flash + Antigravity CLI crushing long-horizon tasks with almost zero hand-holding. 20 mins for SD 1.5 on CPU and a working LoRA trainer from papers? Agentic coding just leveled up.
This is a really interesting development.
Instead of hard-coding workflows, giving agents the ability to discover and compose their own skills is the next logical step.
The real challenge in 2026 will be balancing this autonomy with reliability and cost control.
Great to see open-source pushing this forward.
OpenSquilla released MetaSkill, a self-organizing skill protocol that teaches an open-source AI agent how to discover, filter, and compose its own atomic skills rather than relying on hand-written workflows.
The framework relies on intelligent model routing, sending simpler tasks to lower-cost models while reserving stronger models for heavier workloads, aimed at reducing costs without sacrificing performance.
#sayitbuildit @OpenSquilla
This is such an important point.
Most AI agents fail not because they’re dumb, but because they don’t understand the structure of the websites they’re on.
Giving them a clear “map” (like web.md) before they start is one of the smartest things you can do in 2026.
The shift from “hope the agent figures it out” to “give the agent proper context first” is happening fast.
This is a big signal.
Meta making AI agents available to millions of businesses means we’re moving from “experimenting with agents” to “agents in daily operations”.
The next big challenge won’t be building them, but making them actually remember customer conversations and preferences across sessions.
That’s where the real value (and money) will be made.