🙋‍♂️ SWE @ServiceNow 👨‍💻 Talks to software & cats 🐈 I read Research papers so you don’t have to ✍️ PhotoJournal & Doodles are my thing 🧑‍🎨

Building cool shit with AI and posting about it ✍️ Currently: 🤖 Going all out on Model Context Protocol 🤓 Diving into Deep Learning & research & sharing fundamental insights 🚢 Creator of MCP Maker & Adam Ascension (game) 📑 Published research paper on Recommendation Systems 🧡 Learning in public & sharing the journey 🛣️ ☀️ Day: SWE @ServiceNow 🌙 Night: Everything else Follow along if you want to see what’s possible when you just start building 🛠️
3
1
25
1,918
RT @ghumare64: MCP has its own certification exam now. This curriculum has a study path for it.
Rohit Ghumare
4
1
Any sufficiently advanced tech is indistinguishable from magic. @Shazam is Magic 🪄
"explain how Shazam works" I had no idea Karpathy is so right about this custom explainer videos will change education
1
47
Check this thread for fledge alpha👇
Fledge Alpha is the new free stealth model in OpenCode. No lab, no specs, no benchmarks on the data page. It went from 1 user to 1,818 in two days anyway. Reasoning modes are low, high and max. I'll test it on a messy repo before saying anything else.
4
275
codewithP 🔳 retweeted
deleting the product it's funny how you can sometimes tell if a team culture is dysfunctional just by looking at how their app is laid out. behind every feature, or tab in a sidebar likely lies a PM or someone responsible for an OKR. the problem is when individual performance is measured by these metrics. it becomes the singular focus of the responsible person to make those numbers look good, since it determines whether they get rewarded or axed, even if it comes at the detriment of the overall quality of the product. this is why you mostly see features get added, but rarely removed. if you own a feature, then it being removed feels like an aura loss. this is even worse in the agentic era when the cost to add features has become so low. you can simply will a new feature into existence overnight and add another tab. but users will notice that you ship your org chart. teams that make people play silly games under the guise of performance management rarely build great products. to build a great product you need to first fix your culture
163
199
3,135
160,022
codewithP 🔳 retweeted
Genius move
🚨 side project alert 🚨 Announcing Muse Gadgets, an open source ESP32 firmware and Linux sdk so that you can make hardware devices that work with Muse. Grab an API token from gadgets.muse.ai and point your favorite coding agent at the github repo to build your own peripherals for Muse.
6
4
268
40,617
Been testing video generation of various models. Made a video gen skill having instructions via opus 5.5 and gave to various models. harness: @opencode here's the result:
3
3
154
Each of the model made atleast 1 verification round at the end, while LongCat made another iteration to improve the output. each video was 1 shotted but pre-built skill had a major role in it (maybe can't call it 1 shot if we already had clear steps)
1
33
TL;DR GLM was fastest 🏎️ DeepSeek was most creative 🎨 Mimo was a thinker 🧘‍♂️ LongCat was most reliable 😼
23
for the first time a model has told me "what you're trying to do won't work" 🤝 Fledge Alpha - took the request and identified my intent - did all the ground work research - identified gaps based on facts - gave an honest response due to capability limits of my environment
Fledge Alpha points toward Thinking Machines Lab 👀 OpenCode’s PR references Tinker. Its reasoning settings match Inkling’s, and all 16 image tests followed a 40px patch grid. No exact text-tokenizer match yet, so the owner and underlying model remain unconfirmed.
1
6
414
codewithP 🔳 retweeted
Matt Pocock 刚分享了他心中的 Top 3 Skill Makers,之前我们推荐过Matt 的 Grill Me,到现在,它仍然是我全局安装的两个 Skills 之一。 这次他推荐的三个人,有一个很鲜明的共同点:都在自己的领域有多年实践,然后把工作方法和判断标准写成了 Agent 可以执行的 Skills。 三个人的方向也各不相同👇 1️) @poteto — Lauren Tan 工程流程、代码质量与并行协作 Lauren 是 React 核心团队成员,参与 React Compiler 的开发,先后在 Netflix、Meta 和 Cursor 工作。 她的代表作 pstack,就是自己日常在 Cursor 使用的一套工程 Skills。 它关注的是:如何让 AI Agent 按照严谨的工程流程工作,能协作、能审查,也能验证自己交付的结果。 先把单个 Agent 的工作质量做扎实,再放心地并行推进任务,这是 pstack 很值得学习的思路。 github.com/cursor/plugins/tr… 2️) @dexhorthy — Dexter “Dex” Horthy Agent 架构、上下文工程与软件生产流程 Dex 是 HumanLayer 创始人兼 CEO,也是《12-Factor Agents》的作者。 他长期从事 DevOps、Kubernetes 和基础设施工作,在 Replicated 工作了约 7 年,经历过工程、产品和管理等角色。 他的工作围绕一个很实际的问题展开:怎样让基于 LLM 的软件达到可以交给真实用户使用的水平? 放到 AI 编程里,就是如何管理上下文、组织研究和规划,让 Agent 能在复杂代码库中可靠地推进工作。 如果你正在把 Agent 用到真实工程项目里,这个方向值得重点看。 github.com/humanlayer/skills 3️) @emilkowalski — Emil Kowalski 界面设计、动画与交互细节 Emil 目前在 Linear Web Team 做 Design Engineer,之前在 Vercel Design Team。 Sonner、Vaul、animations.dev,都是他的作品。 他的 Skills 把设计工程中的具体判断写了出来:动画用什么曲线、持续多久、哪些地方值得加动效,以及如何处理那些影响界面质感的小细节。 这些原本依赖经验的判断,现在可以成为 Agent 设计和检查界面时的参考。 Emil 这套 Skills,我们之前也专门推荐过。做前端和产品界面的朋友,可以重点看看。 github.com/emilkowalski/skil… 我觉得:挑 Skills 时,值得先看看作者长期在解决什么问题。 专业经验越扎实,写进 Skills 里的判断和方法,就越值得拿来学习。
My top 3 skill makers: - @poteto - @dexhorthy - @emilkowalski Always learn a ton from reading their skills.
15
94
742
109,457
Today was a setback at work. Prakhar: 0 Politics: 1 Ah! The sweet taste of defeat.
2
55
Dots Bots Harnesses Skills & plugins There was a time we used to fight over for-loops vs recursions you know.
Introducing dots, powered by GPT-6 Astra. Remarkably capable, always-on agents built to handle everything.
1
56
codewithP 🔳 retweeted
AHAHAHAHAHAHAHAHAHAHAHA NO WAY. > OpenAI just announced Dots. > SpaceXAI bought Dot.com it redirects straight to the Grok Bot download page. based lmao.
1,519
2,333
38,280
9,239,973
codewithP 🔳 retweeted
OpenAI DevDay (things that matter) ✅ New $500/month plan ✅ New Dots personal AI assistant on Astra 6 ✅ New banked reset for everyone ✅ Unlimited usage of Dots (not billed against your regular usage) ✅ Sign in with ChatGPT across other apps ✅ New Codex CLI core ✅ GPT-6.1 Sol (close to Astra 6 in some benchmarks but cheaper/faster) ✅ Ultrafast GPT-6 Astra (300 TPS) for 6x the cost of the base API ✅ ChatGPT Spaces (something like Notion) ❌ Nothing better than Opus 5.5 yet (still #1) ❌ $200 plan limit reduced by 50% now
27
15
455
24,704
To those of you who uptake other's work and dont attribute it to them or call out any credits, FUCK YOU.
3
55
Nope Sorry Can’t use pstack Too token hungry
2
57