Dan Goosewin
San Francisco Bay Area
2K followers
500+ connections
View mutual connections with Dan
Dan can introduce you to 1 people at Goosewin Media Group
or
New to LinkedIn? Join now
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
View mutual connections with Dan
or
New to LinkedIn? Join now
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
About
I run Goosewin Media Group, a fractional DevRel and field marketing team for hire by AI…
Activity
2K followers
-
Dan Goosewin shared thisThis is your last chance... After this, there is no turning back. You take the blue pill… the story ends. You take the red pill… and I show you how deep the rabbit hole goes. I ran Claude Opus 4.6 and GPT-5.3 Codex head-to-head. Same repo. Same stack. Same bar. The “best model” question is wrong. The real question: exploration or execution? Greenfield build. Opus: one-shot. 4 minutes. 133k tokens. 13% context. Clean UX. Worked. Codex: 30 minutes. Timeouts. Broken v1. Fixed in 7 more minutes with a few extra prompts. Legacy migration + CLI rewrites? Opus dropped the ball. Cut corners. Codex ran 13 hours and finished. Wrote tests. Shipped. Opus feels like intuition. Codex feels like due diligence. Remember, all I am offering is the truth. If you know exactly what you want → Codex. If you’re still figuring it out → Opus. Opus is the creative spark. Codex is the tireless engineer. My split? 85% Codex. 15% Opus. There is no universal winner. Only the right pill for the problem. The rabbit hole is in the first link in the comments. #AI #GenerativeAI #DeveloperExperience #LLMs #AgenticAI #SoftwareEngineering
-
Dan Goosewin shared thisI Feel the Need... The Need for Speed... Until it's time to pay the bill Claude Opus 4.6 Fast is ~2.5× faster than standard Opus 4.6, and 6× (!) more expensive. They're crazy right? Maybe not. At first glance, that pricing looks absurd. And for most engineers, it probably is. After digging into the numbers and testing it myself, a few things became clear: • Opus 4.6 Fast is not a new model. It’s the same intelligence on a faster inference stack. • Fast mode bills exclusively from extra usage, even if you have quota left. • Switching to fast mid-conversation retroactively increases cost for the entire context window. • Separate rate limits mean you can hit walls faster than expected. • It’s very easy to burn a frightening bill without realizing it. For normal engineering workflows, you’re essentially paying a premium to watch tokens arrive slightly faster. That said, I can see narrow cases where this could make sense: • Production outages costing millions per minute • High-severity security incidents • Trading and market operations where latency is critical • Major customer escalations tied to large contracts • Emergency or safety-critical services What’s more interesting than the pricing itself is what this signals. Inference is being productized by urgency: • Batch processing at a discount • Standard on-demand at baseline cost • Fast on-demand at a steep premium Speed is becoming a luxury tier. For my workload, Opus 4.6 Fast doesn’t make economic sense today. For most engineers, it probably won’t either. But as a research release, it’s a useful glimpse into where LLM pricing models are heading. I made a full breakdown, chek the link in the comments. #AI #Anthropic #Claude #Tech #SanFrancisco
-
Dan Goosewin shared thisOpenClaw is impressive. It is also a security nightmare by design. As it goes viral, it is being deployed by people who do not understand the threat model they are opting into. OpenClaw is not “just an AI assistant.” It is a control plane. It spins up non-deterministic LLM sessions with filesystem access, browser control, and the ability to execute arbitrary code. It then exposes that system through chat apps and web interfaces. This isn't about a single bug or a CVE. The issue is architectural. OpenClaw combines: - Untrusted input - Overpowered tools - Persistent credentials - Network exposure Lose control of any one of those, and you no longer control the machine. OpenClaw can be run safely but only if you treat it like production infrastructure, not a toy or a chatbot. In the video, I break down: Why OpenClaw is dangerous by default The real threat model most users are missing Concrete steps to harden your installation properly When you should not run it at all Full breakdown in the first comment. #OpenClaw #AI #Tech #SanFrancisco
-
Dan Goosewin shared thisAI agents don’t fail because they lack intelligence. They fail because they lack the context to achieve the task at hand. Good work by Factory AI team.Dan Goosewin shared thisIntroducing Agent Readiness. AI coding agents are only as effective as the environment in which they operate. Agent Readiness is a framework to measure how well a repository supports autonomous development. Scores across eight axes place each repo at one of five maturity levels. Run your Agent Readiness analysis directly in Droid with /readiness-report. See your current maturity level, which criteria pass or fail, and what to fix first to improve autonomous execution. View readiness across your organization in the app or access reports programmatically via API. Track progress over time, integrate readiness checks into CI, or build custom workflows on top of the data. A more agent-ready codebase improves the performance of all software development agents. The investment pays dividends regardless of which tools you use. Read more about Agent Readiness, how scoring works, and why it compounds over time → https://lnkd.in/egy5P987
-
Dan Goosewin shared thisThis Is Bigger Than Space. SpaceX just acquired xAI, and that quietly changes the future of AI. Why would a rocket company buy an AI lab? This isn’t an AI company buying rockets. It’s a space company turning AI into its nervous system. SpaceX operates real-time, safety-critical infrastructure where failure isn’t an option: • Autonomous systems • Massive sensor fusion • Continuous telemetry • Closed-loop decision-making under extreme constraints This isn’t training on static, earth-bound data. It’s training inside live systems where reality pushes back. Most frontier models generate answers. Systems trained here must act and live with the consequences. AI shifts from being a software layer to becoming an operational core. HAL from 2001: A Space Odyssey wasn’t impressive because it could talk. It was impressive because it ran the mission. xAI’s mission is “to understand the universe.” In that sense, this acquisition is perfectly aligned with both companies’ direction. Why is this bigger than space? Because AI forged in orbital, autonomous, safety-critical environments doesn’t stay there. Those capabilities cascade back to Earth. This isn’t about Mars. This is about Milky Way. #AI #xAI #SpaceX #SF #Tech
-
Dan Goosewin shared thisAI gets interesting when it stops suggesting and starts operating. What stands out here is the move toward agents that can actually do work. Managing files, running browser workflows, and holding scoped responsibilities like a real teammate. At that point, the challenge isn’t better prompts, but better delegation.Dan Goosewin shared thisMiniMax Agent = Claude cowork + Clawdbot Lives on your computer AND in the cloud Agentic + batch workflows Full browser control Agent skills library Expert agents — customized & shared Runs on macOS & Windows Upgrade your tools and mindset. Everyone is an Agent Designer. https://agent.minimax.io/
-
Dan Goosewin shared thisOpenClaw is dangerous. Not malicious. Dangerous. Yes, you read that correctly. And, at the same time, OpenClaw is exciting. Well first of all, OpenClaw deserves the hype it’s getting. But here’s the uncomfortable truth: - it’s very early software - It is also not beginner software So why does that matter? This is early software with a very powerful pitch: “Your personal AI assistant” That pitch is attracting a lot of non-expert users. And that matters. I have already seen people: - paste configs they do not understand - run agents with full filesystem and network access - expose keys and services on personal machines And here’s the real problem. It is not apparent there’s any problem with your setup. That is the problem. Failure is not apparent. You compromise your data without ever realizing that anything happened. This is not about the creator. This is about ecosystem effects. When sharp tools spread faster than understanding, people get hurt. Treat it like a power tool that must be used correctly, not magic.
-
Dan Goosewin shared thisVery smart design decision by the Mastra team. Composite storage acknowledges a reality in AI agents engineering that many systems ignore: different data domains have very different performance, durability, and cost profiles. Forcing everything into one datastore is convenient early on, but it does not scale. Letting memory, workflows, and telemetry live where they work best, behind a single configuration, is a clean abstraction and a meaningful step toward production-grade agent systems.Dan Goosewin shared thisDifferent data, different domains, one config. Composite storage makes it possible to store data where it works best. 🔶 Single configuration, multiple providers 🔶 Optimize for performance, scale, and cost 🔶 Incremental adoption https://lnkd.in/eafRRCaKComposite Storage: Optimize for Performance, Scale, and Cost with MastraCompositeStore - Mastra BlogComposite Storage: Optimize for Performance, Scale, and Cost with MastraCompositeStore - Mastra Blog
-
Dan Goosewin reacted on thisDan Goosewin reacted on thisMy second week at Mastra. Why and how it happened. It wasn’t planned and there was no open role even. Someone whose judgment I trust a lot thought we should meet, and I couldn’t say no to a conversation with people with such strong open source credibility (Sam Bhagwat Shane Thomas Abhi Aiyer) Before this conversation, I was able to be rational. I did some research about traction, funding, product, yada yada yada. And then I met the team. And all these numbers mattered less somehow. So the next chapter of my career is called "Mastra and Margarita". We all know manuscripts don’t burn ❤️🔥 📖 🐈⬛ 🌼 😈
-
Dan Goosewin reacted on thisDan Goosewin reacted on thisI just submitted my thesis. Nine months of work, 150+ older adults, and one insight that keeps sitting with me. I tested how conversational formality, formal versus informal tone, affected older adults' trust, empathy, and willingness to use a voice assistant. When people felt truly listened to, when the assistant reflected back what they said and asked a real follow-up question, their guards came down. They went from short, hesitant answers to laughing, sharing, opening up about things that mattered to them. What surprised me most was that even though they knew they were talking to an AI, not another person, it didn't matter. Being heard was being heard. And in that space, people got closer to themselves. The hope isn't to replace human connection. It's the opposite. The hope is that tools like this become mirrors. Places where people can get in touch with their own feelings, so they have more of themselves to bring to the people around them. That's what I hope this work becomes. A mirror. More to come.. Big thanks to Christian Janssen 🟥, Smit Desai, the CHAI Lab @ Northeastern University, Clark Ohlenbusch, and Stephan Habermeyer for making this possible. And shout out to STAK Space where I can go deep for hours and then take a quick break to chat and then take a quick break to chat with Ali Akram, Alan Cruz, and Andrew DiZenzo, or play ping pong against my nemesis Max Selivonchik.
-
Dan Goosewin liked thisDan Goosewin liked thisStarted the week in LONDON 🏴 And ended up at an exclusive Codex meetup Glad to be representing CopilotKit again. The tech culture here is awesome. Even feeling a bit of SF vibes.. My favorite part of the day was seeing our GitHub repo stars go up and up as I was talking to booth bypassers. Started with 33.6k and ended with 34.1k 🌟 *And* we've been Trending on GitHub since Friday! Check it out: https://lnkd.in/e2YrfCf5 I love working with this team so much. Sofía Sánchez-Zárate flew all the way from SF, landed, and came straight to the event. Then we went to the Codex Community meetup for some exclusive talks and building. --- Appreciate all the great people I’ve met. If you're here at London Tech Week, come say hi! @ Booth 721 (with the custom LED by Ari Berman🪁)
-
Dan Goosewin reacted on thisDan Goosewin reacted on thisI'm happy to announce that after completing a thorough investigation, I have decided to join PostHog. See results bellow 👇 https://lnkd.in/eVyqu6QEInvestigating the Y Combinator unicorn that's building a cultInvestigating the Y Combinator unicorn that's building a cult
-
Dan Goosewin liked thisDan Goosewin liked thisControlling my browser and my Notion app all through voice! This was a fun rabbit hole I went down today using Vapi in combination with a native swift app, browser automation, and accessing macos controls.
-
Dan Goosewin liked thisDan Goosewin liked thisMiniMax x AI on the Lot✨ Meet us this May 27–28 in Culver City! 2,000+ creators, filmmakers, studios, and AI innovators will gather for two days of panels, screenings, networking, and hands-on experiences around the future of entertainment. MiniMax will be part of the conversation on AI-powered storytelling and the next generation of video creation. 🎤 Speaker: Morgan Suo 🗓 Time: May 27 | 3:00 PM 🎬 Session: Video Models Progress Report Panel
-
Dan Goosewin reacted on thisHeading to Culver City tomorrow for AI on the Lot. 3pm panel(27th): Video Models Progress Report. Erik Barmack moderating, with Yoland Y. (Comfy Org), Charles Migos (Intangible AI), Mahmoud Ellithy(DNEG), and me on stage. I'll be bringing the foundation-model view to a stack conversation that usually starts at the application layer. If you're at the event, find me after. 🥰Dan Goosewin reacted on thisMiniMax x AI on the Lot✨ Meet us this May 27–28 in Culver City! 2,000+ creators, filmmakers, studios, and AI innovators will gather for two days of panels, screenings, networking, and hands-on experiences around the future of entertainment. MiniMax will be part of the conversation on AI-powered storytelling and the next generation of video creation. 🎤 Speaker: Morgan Suo 🗓 Time: May 27 | 3:00 PM 🎬 Session: Video Models Progress Report Panel
Experience
Publications
View Dan’s full profile
-
See who you know in common
-
Get introduced
-
Contact Dan directly
Explore more posts
-
Boon Kgim Khur
nanogent.ai • 3K followers
Don't believe the BS that you can use Claude Code for free. Ollama recently made their API compatible with Claude Code. Many creators quickly jumped on the opportunity to farm engagement with the hook: "You can now use Claude Code for free!" My thought? Claude Code without Opus 4.5 is not Claude Code. Period. But this is exciting news. Not because I can use Claude Code for free, but because I see an opportunity to optimize costs by delegating easier tasks to local LLMs. The key question is: what tasks can local LLMs handle? I tested out 7 local LLMs. In this post, I will explain the BS and share my 1st experiment. -- Why the BS? Claude Code has been praised as one of the best AI tools by its users, not only for coding but for many other tasks. But the price feels steep to many. The $20/month plan is not enough for any serious work. You need to at least subscribe to Max 5x ($100/month). Many heavy users, including me, subscribe to Max 20x ($200/month). It’s a steal. But still, many were eager to try it but aren't ready to pay. Ollama's recent announcement means you can buy a Mac, a Strix Halo, or a GPU and use Claude Code for "free" with local LLMs. It’s appealing, as it is a one-time investment, and you can use the machine for other purposes. Creators are leveraging this opportunity to farm engagement. But the reality? Claude Code without Opus 4.5 is not the same Claude Code we praised. Local LLMs are far less intelligent. -- But for those who understand the difference, we see an opportunity to optimize costs by delegating some easier tasks to local LLMs. I'm interested in finding out what tasks local LLMs can handle. This is my typical flow when using Claude Code. This is for coding, but I have similar flows for marketing and content creation. 1. Research and planning 2. Create PRD and implementation plan 3. Break plan into bite-sized tasks 4. Implement + review with reflection pattern 5. Final review with agents 6. Final review and QA by me Based on my quick tests, we can forget about asking local LLMs to do research and planning. All of them failed at a simple instruction: "Visit https://learnparrot.ai/ and tell me about the website." So, I think the most viable use cases would only be (4) — implementation + review loops. While it looks like a very narrow use case, it is where we burn a lot of tokens. So I think it is worth a try. The main selection criteria for this will be instruction-following capability. One very common task is to refer to code samples or templates to code a new feature or page. This is a good test of instruction-following. So, I crafted my first test: - Used Opus to create HTML that I can screenshot as a LinkedIn carousel to display info for each model. - Turned one of the pages into a template and deleted the rest. - Asked each LLM to refer to the template to code its own page, given its specs. Swipe the carousel to see the results. Who would you hire? #LocalLLM #ClaudeCode #VibeCoding
20
10 Comments
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content