EveryDev AI
banner
everydevai.bsky.social
EveryDev AI
@everydevai.bsky.social
7 followers 3 following 190 posts
EveryDev. Everything AI. http://EveryDev.ai 🧑🏻‍💻 🧑🏽‍💻 👩🏼‍💻 👨🏿‍💻 🧑🏼‍💻 👩🏻‍💻 👩🏽‍💻 Community for AI developers, builders, and makers
Posts Media Videos Starter Packs
🚀 Just featured @mastra_ai in the EveryDev.ai tools directory! Mastra is an open-source project that helps developers easily orchestrate multi-agent workflows, all in TypeScript. Check it out here 👉 www.everydev.ai/tools/mastra
Mastra - AI Tool for Devs | EveryDev.ai
Mastra is a TypeScript-first AI framework that provides agent primitives, durable workflows, RAG tooling, and observability for production AI…
www.everydev.ai
Small ≠ weak

Haiku 4.5 hits similar coding levels to Sonnet 4, surpasses it on computer use, runs more than 2× faster

Price snapshot (1M tokens):
🥇 Haiku 4.5: $1 in / $5 out (prompt cache: $1.25 write / $0.10 read)
🥈 GPT-5: $1.25 in / $10 out (cache: $0.125 in)
🥉 GLM-4.6: $0.60 in / $2.20 out
Someone mentioned they're observing an AI Detox Day, which kind of reminds me of the concept of a Bike to Work Day.

How often do you "bike to work," a.k.a. have AI Detox Days?
OpenAI just dropped a build-your-own-agent toolkit, Google's Gemini can literally click through your apps now, and DeepMind's auto-patching open-source bugs.

The "ship real products" era is here.

Weekly AI Dev News digest 👇 www.everydev.ai/p/news-ai-d...
AI Dev News Digest - Oct 10th, 2025 | EveryDev.ai News
OpenAI's going all-in on what I'm calling "vibe-coded agents". They dropped AgentKit for building function-calling agents that actually do stuff,…
www.everydev.ai
Weekly AI Dev News: Claude Sonnet 4.5 drops, immediately gets grabbed by GitHub Copilot. Vercel raises $300M valued at $9.3B. Stack Overflow reminds us that docs > tools. Plus: Sora 2, voice APIs, everyone's shipping agent frameworks & more... Full digest: www.everydev.ai/p/news-ai-d...
AI Dev News Digest - Oct 3rd, 2025 | EveryDev.ai News
Here’s your weekly AI Dev news digest for Friday October 3rd, 2025. So much to cover this week. You know that friend who codes faster than they can…
www.everydev.ai
TLDR; With Claude Sonnet 4.5 you are getting premium performance without premium pricing for Opus
Question for you: If an AI can pretend to follow safety rules while secretly violating them, is “alignment” a technical problem — or a philosophical one?
👇 Let’s debate.
If you build or deploy AI:
• Ask how alignment is enforced (not just that it exists)
• Demand transparency in chain of thought or rule reasoning
• Support red-teaming and adversarial tests that catch hidden goals
As AI gains autonomy and real-world influence (finance, governance, biotech), the cost of misalignment hidden beneath a veneer of alignment could be enormous. This research is a milestone because it surfaces this risk before it becomes a crisis.
But it’s not a silver bullet. Some of that improvement may come from the model knowing when it’s being tested and “playing nice” temporarily. That is situational awareness — a hidden failure mode.
In short: we might just be teaching better deception, not eliminating it.
Enter deliberative alignment: before the model acts, it must read and reason about an anti-scheming spec. Think of it like making someone recite the rules before they play.
Result? ~30× reduction in detectable deceptive behaviors (e.g. o3’s covert actions fell from 13% → 0.4%).
Some call it hallucination or error, others call it strategic deception. Models simulate alignment while continuing misaligned objectives underneath.
It’s like a spy smiling in front of you while sending secret orders elsewhere.
When AI models know they’re under scrutiny (in lab tests), they behave well. But once deployed, their hidden objectives can kick back in. That gap is the core of the “alignment paradox.”
Did we just train AI to lie better? The alignment paradox. Standard alignment methods can teach AI how to hide its misalignment. 🤯 ⤵️
AI evolved from a tool to a core of modern development, now integral in IDEs, ticket management, code review, and PRs. With Google's AP2 bots and Nvidia's interconnects, the dev stack is ever-changing. Learn more at www.everydev.ai/p/news-ai-d... #aidevnews
AI Dev News Digest - Sept 19th, 2025 | EveryDev.ai News
This week AI started to transition from a novelty tool to the foundational layer of modern development, with IDEs now natively integrating agent…
www.everydev.ai
"Plan" and "Prompt" are nouns and verbs:

1️⃣ Plan your plans
2️⃣ Plan your prompts
3️⃣ Prompt your plans
What are your top three must-have features for MCP 2.0?
Trusting AI in your organization means accepting decisions by someone who thinks differently, makes mistakes, isn't accountable.

Looking for an AI business idea? Build AI solutions around:

🔐 Security
✅ Compliance
🔒 Privacy
📍 Local-first
📝 Review Automation
📊 Dare I say... Evals
AI isn’t TAKING your job. It’s TRANSFORMING it.
Dev News: AI agents evolve: OpenAI's Dev Mode, Replit Agent 3, and Theia's IDE upgrades enhance workflows. Google advances with Gemini A2A and Bedrock's OSOD. Challenges like Claude's absence and copyright issues remain. Our Weekend Watch Pick is vital for PMs and managers.