Activity
1K followers
-
John Arrow shared thisMy recent side quest. https://lnkd.in/gneHpQtkI'm the VC who created AI Scott Adams. Here's why I'm continuing the project, despite his family's objections.I'm the VC who created AI Scott Adams. Here's why I'm continuing the project, despite his family's objections.
John Arrow commented on a post
2w
Hi Theresa Payton ✪ I'm the creator of AI Scott Adams through my entity Age Of AI. Thanks for sharing. One key piece of the story that you may not be aware of is that Scott Adams, while living asked people to create an AI of him dozens of times after he died. Here a sample where he authorized and actively encouraged what we made:
https://x.com/AIScottAdams/status/2021407446376546775?s=20
-
John Arrow liked thisJohn Arrow liked thisI was wrong about Clawbot. Have now been educated by Thanh Pham I thought it would be a toy. Maybe a neat wrapper. Maybe a glorified command center. Wrong on most counts. I had an extra Mac Mini sitting around, so I moved it there and put Codex 5.4 behind it. That changed the feel immediately. It stopped feeling like a demo and started feeling like infrastructure. My wife uses it in Telegram. We have a date night thread that gives us things to do. I gave it our Hawaii honeymoon itinerary and now it checks flooding, makes sure the plan still works, and surfaces backup options before anything breaks. My team each has an agent. They manage Linear, check in with me, and keep work from disappearing into the usual cracks. My band has an agent in Telegram that handles practice scheduling and keeps track of what we’re working on. I have separate agents for health and personal finance. Slack for work. Telegram for wife and band. Same underlying system, different surfaces, different jobs. It’s not perfect. I have spent an absurd number of hours tweaking it. Context, permissions, routing, handoffs. All the annoying parts matter. Still, this changed how I think about orchestration. The unlock was giving it constrained access to GitHub and Vercel. A new site idea doesn’t go into a backlog anymore. It just happens. I was way too dismissive of this category.
-
John Arrow liked thisJohn Arrow liked thisI've been experimenting with AI agents for the past month — not just chatting with them, but having them pick up tasks and get work done. Made some mistakes, learned a lot. Here's what worked and what didn't.
-
John Arrow liked thisJohn Arrow liked this✨ Celebrating Legal Tech Leaders at Legalweek Great to be at the Legalweek Leaders in Tech Law Awards celebrating innovators across the legal technology community! Katherine Allen 👋 had the honor of presenting several awards throughout the evening, including the Monica Bay Women of Legal Tech Award. Named after the former editor-in-chief of Legaltech News, the Monica Bay Award honors women at law firms, legal departments, and legal technology or legal service companies who have made significant contributions to legal innovation. Congratulations to the 2026 recipient, Katherine E. Charonko, Practice Group Leader in Electronically Stored Information & Technology at Bailey & Glasser, LLP. Always great to connect with clients and friends in the community. Katherine Allen 👋 and Claudia Grillo especially enjoyed continuing the conversation with Caren Ulrich Stacy, CEO of Diversity Lab, a Finalist for Best Tech Training Program.
-
John Arrow liked thisWow! Today was a really remarkable day in that two different senior Gartner analysts at two different Gartner conferences - in Dallas and Orlando - recognized KUNGFU.AI from the main stage as an example of what the future of AI professional services looks like and the type of firm that can help the most ambitious companies change the basis of competition in their industries.John Arrow liked thisNice surprise today at Gartner's Product Leadership Conference Keynote, Mark P. McDonald used KUNGFU.AI as a prime example of what the future of professional services looks like: 1. Strategy experts who deeply understand technology, and understand what the business should do to deliver value 2. PhD-level engineers who can make AI actually deliver on the strategy The result? 10x improvements and a direct challenge to every other service provider still operating with bloated, traditional delivery models. This is exactly why we built KUNGFU.AI the way we did. Lean. Expert. Focused on outcomes. Thank you, Gartner, for the recognition.
-
John Arrow liked thisJohn Arrow liked thisCodex CLI, Spark, and GPT-5.4 I've been using Codex CLI every day for the last few months. The app exists. Some people I know at OpenAI have switched to it. I'm still on the CLI. Old habits, maybe. But the CLI is fast, I know the shortcuts, and I don't need another Electron window. GPT-5.4 shipped this week. OpenAI is calling it their all-rounder: the reasoning depth of 5.2 XHIGH combined with the coding ability of 5.3 Codex. One model that plans and builds in the same context. I've been testing it since Thursday. Spark is still more interesting to me for daily work. GPT-5.3-Codex-Spark runs on Cerebras hardware at 1,000+ tokens per second. For front-end work, it is the best model I've used. The common complaint about Codex models is over-coding. You ask it to change a button color and it refactors your entire component tree. You ask for a simple API call and it rewrites your error handling and adds types you didn't want. This is a real problem. I've seen it with 5.3 Codex and now with 5.4. Spark doesn't do this. You tell it what you want, it does that thing, and it stops. For front-end iteration
-
John Arrow liked thisJohn Arrow liked thisGPT-5.3 Instant dropped this week. No new capabilities. No reasoning upgrades. The whole release note is about the model being less annoying. Fewer refusals. Fewer disclaimers. Less moralizing. OpenAI used the word "cringe" to describe their own model's previous behavior. The fix: stop lecturing people, stop refusing safe questions, stop prefacing every answer with seven sentences of throat-clearing. You won't find "times the model lectured you unprompted" on any leaderboard. But it's the thing that actually determines whether people keep using the product or close the tab. I keep coming back to this pattern. The model companies spent two years competing on math scores and coding benchmarks. Now the most-used AI model in the world just shipped a patch for personality. We measure what's easy to measure. We ship what actually matters. Those are almost never the same thing. What's the thing your users complain about that doesn't show up in any dashboard?
-
John Arrow liked thisJohn Arrow liked thisI have a new vibecoding project that I'd love for you to check out. It's called Fedlock and it's an attempt to measure FOMC sentiment with LLMs. https://lnkd.in/enG7up6H And you can check out my full writeup in today's Odd Lots newlsetter. If you're doing stuff with AI or economics, I'd love to hear your thoughts and complaints.
Experience & Education
-
Mutual Mobile
View John’s full experience
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
Other similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top contentOthers named John Arrow in United States
15 others named John Arrow in United States are on LinkedIn
See others named John Arrow