Sam Kennedy (@samrolken) profile picture

Software developer + product builder, launching for e-commerce, local business, LLM/AI experiences. For hire to build web products, fractional CTO, tech advisor

NYC
Joined November 2007
Photo by Sam Kennedy (@samrolken)Photo by Sam Kennedy (@samrolken)Photo by Sam Kennedy (@samrolken)Photo by Sam Kennedy (@samrolken)Photo by Sam Kennedy (@samrolken)Photo by Sam Kennedy (@samrolken)Photo by Sam Kennedy (@samrolken)Photo by Sam Kennedy (@samrolken)Photo by Sam Kennedy (@samrolken)
I’ve always been a claude/anthropic customer and always recommended them and brought them to companies I’m working with. But there’s no way around it: they are full of shit on this one. They always said 20x and that’s it. I never heard that the weekly limits were 10x. I just saw 20x everywhere. And this was backed up by the discount language and more. They lied to us and tricked us about what we were buying and ripped us off, which is a very severe blow to their trust and reputation and will be remembered. And on top of that they make it sound almost like we were too stupid or it was some misunderstanding? Gross.
Don't gaslight us. "I get that some new users assumed 20x applied to both limits. We could have done better here, and I'm sorry." If you didn't want to be misleading, you would have left out the '50% savings' claim on the $200 plan—at 10x the price of the $20 plan ($20 x 10 = $200), a 50% discount clearly implied we were getting 20x the usage.
48
Figma doesn’t support AI tools in 2026.
Replying to @dsp_
Hey David, Gayani here from Figma. You're right that our remote MCP server only accepts clients on our supported list, and Pi isn't on it yet. You can see the current list in our MCP catalog at figma.com/mcp-catalog. If you'd like Pi considered for a future addition, please fill out this form: forms.gle/qSvUawwznWyoj8Go7.
1
44
I never used to hit limits with @Claude Code, so I never felt the need to maximize usage. Now that limits are more apparent, my mindset has shifted to squeezing every last drop out of my subscription. Might backfire on @AnthropicAI... scarcity changes behavior.
128
Paid @sanity_io hundreds of dollars to cover a bandwidth overage and still dead in the water. No support whatsoever is available... just an AI chat bot and discord server where you get ignored.
1
146
Wow my first ever post to hackernews hit the front page. news.ycombinator.com/item?id…
2
1
3
193
I spent a few hours last weekend building a web server with zero application code. Every HTTP request goes straight to an LLM. It decides what to do in an agentic loop using three tools: database, webResponse, updateMemory. Here's what happened →
2
68
The goal: test if AI can replace code entirely, not just write it. Everyone's focused on Cursor, Copilot, AI that generates source files. But what if you skip that step? What if the AI just... does the thing?
1
20
Recent Azure/AWS outages are reminders that every cloud provider is a single point of failure in your architecture.
159
Companies confuse technical debt with architecture debt because the symptoms look the same - delays, outages, rising costs. But the fix is completely different. Technical debt requires building work: coding, refactoring, testing. Architecture debt requires organizing work: governance, standardization, cross-org coordination. One needs more engineers. The other needs executive authority to make decisions across team boundaries, sunset duplicate platforms, and enforce standards. You can't fix architecture debt in a sprint because it's not about code - it's about power, policy, and organizational structure. Sound familiar to anyone?
1
35
Remember when you spent weeks learning Claude Code's hooks, subagents, and custom tools? You thought you were just getting better at automating your dev workflow. Turns out, you were learning something way bigger. 🎯 THE PLOT TWIST: Last month, Anthropic renamed the Claude Code SDK to the Claude Agent SDK. Not just a rebrand - a reframing of what developers have been building and the introduction of a general purpose agentic SDK... one that developers have been going crazy about already. Every hook you wrote to intercept bash commands? Works for any agent task. ✅ Every subagent pattern you learned for parallel code review? Applies to customer support, research, finance workflows. ✅ Every MCP integration you built to connect APIs? General-purpose infrastructure. ✅ 💡 WHY THIS MATTERS: Thousands of developers spent months mastering what they thought was a coding tool. They learned: 🔄 How to structure agent loops (gather context, take action, verify) 🧠 How to manage long-running context with compaction 🔐 How to build safety controls with permissions 🤝 How to compose specialized subagents ⚙️ How to give agents filesystem access and bash tools They were solving problems like "automated code review" and "CI/CD integration." But those same primitives work for ANY workflow where you need an agent to read files, run commands, check results, and iterate. 🚀 🏗️ THE VALIDATION: This isn't speculative. Developers are already shipping production agents for: 💬 Customer support - handling ambiguous requests and routing 💰 Finance operations - analyzing portfolios and generating reports 📋 Compliance auditing - scanning documents and flagging issues 🔍 Research workflows - synthesizing large document collections The infrastructure was there all along. The rename just made it explicit. 🧠 THE REAL LEARNING: Infrastructure knowledge is more durable than framework knowledge. Learning "how this specific framework handles email" is fragile. Changes when the framework changes. ❌ Learning "how to architect agents with primitives" is durable. Transfers across domains. ✅ Developers who went deep on Claude Code customization got the durable knowledge. Now it applies everywhere. ⚡ THE OPPORTUNITY: If you learned Claude Code customization, you already know how to build: 💼 Finance agents that analyze portfolios and run calculations 📚 Research agents that synthesize documents across large collections 🎧 Support agents that handle ambiguous requests and escalate smartly 📅 Personal assistants that manage calendars and draft communications Same tools. Same patterns. Different domain. 🎯 The coding knowledge transferred - you just didn't know it would. What workflow are you going to automate first?
55
He says some dumb stuff sometimes but who doesn’t? I stand with @dhh and let’s build!
25
UI frameworks from Apple, Microsoft, and Google are hilarious. Aqua (25 years ago) was the original liquid glass. But it turned into brushed metal/skeumorphism. Aero glass flattened into Metro tiles, Material Design lost depth overnight. Now we’re back at Liquid Glass, and somewhere Apple's design team is 100% already getting started on ‘groundbreaking’ flat minimalism design system 3.0 due in time for the WWDC 2029 keynote.
180
Who would like to help test out a full-featured MCP server for @37signals’ @basecamp?
1
2
183
Apple should just lean into the “weird haiku” Apple Intelligence style of notification summaries instead of trying to “fix” them.
87
Apple Music for Mac is a buggy afterthought and Apple doesn't care about its users at all.
1
132
How can you call it a pricing page when it has no prices? mariadb.com/pricing/
1
61
I have a lot of luck using 4o to build and iterate on context and then carry that into o1. I’ll ask 4o to break down concepts, make outlines, identify missing information and think of more angles and options. Then at the end, switch on o1 which can use all that context.
this is an amazing way to think about prompting o1 from @benhylak
1
95
Why even try to use @Uber @Uber_Support when the ride you think is on the way actually isn’t.
1
143