Yo Sub Kwon (@yosubkwon) profile picture

I build awesome things. w00t!

Las Vegas, Nevada, USA
Joined January 2009
Photo by Yo Sub Kwon (@yosubkwon)Photo by Yo Sub Kwon (@yosubkwon)Photo by Yo Sub Kwon (@yosubkwon)Photo by Yo Sub Kwon (@yosubkwon)Photo by Yo Sub Kwon (@yosubkwon)Photo by Yo Sub Kwon (@yosubkwon)Photo by Yo Sub Kwon (@yosubkwon)Photo by Yo Sub Kwon (@yosubkwon)Photo by Yo Sub Kwon (@yosubkwon)Photo by Yo Sub Kwon (@yosubkwon)
we're on a crazy moonride right now! #bitcoin
13
I've canceled Calendly, DocuSign, Notion and HubSpot. Not because they're bad. Rebuilding them now costs me less than a month of the subscription. Accounting and task management went too. I think CRM is next. Which one are you canceling first?
1
1
91
A tactic that has worked for me: give an AI agent your to-do list and let it try to do the tasks on its own. Where it gets stuck shows you what knowledge or access it's missing. Fix that once and it helps every task that needs it.
46
I think the bigger AI bottleneck is imagination, not prompt tricks. How many useful things do we never try because we haven't imagined asking? A custom app, tools that work together, maybe your blinds and lights (hardware and access permitting).
1
1
4
166
Sometimes the best response to an AI agent asking for credentials: there's already a service that does that. Agent wants a deploy token? Route it to the deploy service. Finance agent needs bank data? A browser tool fetches it. Fewer credentials handed out.
1
1
67
There's a lot of value in the contacts you already know. The difference between the personal CRM I built and LinkedIn: it doesn't depend on people maintaining a profile. It knows the actual history between us, and there's no algorithm deciding who I see.
52
Running many AI agents? One approach: an auth agent handles their access requests, and the system around it has to enforce the permissions. A convincing argument shouldn't be enough. Say one books a flight: account access and permission to spend are separate decisions.
1
81
I had AI build a personal CRM from my own email and messages, then asked it: who in my network could give useful feedback on my AI videos? It answers with people and the relationship history behind each one. Here's the walkthrough, including the prompt I started with.
2
1
5
315
My Mac Studio doesn't replace my Linux servers. It's one more resource in the pool, next to the models my Claude and Codex subscriptions can use. Work should get delegated to it, like deep research that runs in the gaps. Where the balance lands, I'll have to see.
2
168
I think a lot of what feels like a limitation of AI is a limitation of what we've given it to work with. You carry meeting notes, emails, your calendar and past conversations into your work. Then you give AI a couple of sentences.
1
82
Prompting is dead. The tricks, anyway. Clear instructions still matter. When I had AI build a word processor, the prompt was a test: same doc in Word and our editor, same edits, swap, compare. Describe success the system can check. Hard to learn, and fading with each model.
3
185
Most of what my AI agents do doesn't need to be fast: calendar cleanup, document classification, accounting upkeep, filling out forms. That can run overnight on local open-source models. What still goes to frontier models: anything where misreading intent creates real problems.
4
3
126
21,332
My dog has entered the vibe coding era.
1
3
158
I gave one AI agent two weeks and a one-paragraph prompt to rebuild Microsoft Word for my company. Same .docx, same edits, compare the renders. Still failing the parity gate by 0.001. I shipped it to my engineers anyway. Would you have?
1
4
284
I built ProveMyHome with AI: turn a home inspection report into a usable property record. After a rebuild, I'm still asking who pays. Homeowners, inspectors, or agents? Operator information: operator.yosubkwon.com
1
3
737
Yo Sub Kwon retweeted
Breaking: Anthropic just cut off Max plan access for third-party Claude tools like OpenClaw. If you're a power user running agents across multiple subscriptions, this changes your math immediately. A quick video reaction to what happened, why Anthropic did it, two stopgap options if you need to keep things running today, and why neither of them is a long term answer. The free ride on subscription arbitrage is over. Check out our blog post to read what comes next: permaship.ai/blog/anthropic-… #openclaw #claude #anthropic #opensource
2
6
2
9
1,383
Claude Code source leaks (again) and this time it's got a gacha-buddy system built in where every user deterministically (by user ID) gets a buddy. It gets a procedurally generated sprite and its own personality.
1
1
1
312
In the long, long ago, when you would run code you just wrote and it actually worked on the first attempt, you would be jubilant. That scenario will never happen again. Now, I expect it should work on the first attempt.
1
100
14 of the 17 flights I've booked so far this year had delays. The only thing that has been saving the connections I haven't missed is the connecting flight being delayed as well. Suppose this is the new normal.
114
I know a lot of complaints in the AI world lately are about how poorly Opus, Codex, or some other model is performing in the moment, which can have a lot of variability. For some things, I'm sure it makes a big difference, but generally for software development, it really doesn't make much of a difference at all. Whether people intuitively understand it or they have read research, it's a fact that a workflow outperforms one-shot model usage. The more robust your workflow, the less important the actual quality of an individual model working within it. A really simple workflow might be build with Claude, and critique with Codex. Repeat. Complex workflows are often created by individuals tailored for their own purposes and may involve sub-agents, a bunch of customized prompts, or a series of loops that check all steps of a PRD are completed and that the QA, Security, or any post-review agents say everything is good. The more your trust your evals and and workflow, the less concerned you can be about the day-to-day performance of a model, and the less you have to be worried about model drift or how well your systems will perform when you change to new models. If your output is generally acceptable after it goes through your process, then you can optimize more for cost and speed.
1
110
I have open-sourced a project: Nexus. It's a project I've been working on to make AI proactive, not just responsive. As a CEO, I juggle priorities and the real bottleneck was never just execution, it was judgment. I didn't want AI to wait for me to hand it tasks, but something that could understand my goals, identify and prioritize valuable work on its own, and act accordingly. AI was already very much involved in my processes. Eventually, it reached the point where it could independently surface meaningful opportunities, propose work, execute it, and follow through. That's now Nexus: github.com/PermaShipAI/nexus Essentially, any number of specialized agents continuously look for work they believe would be valuable based on overarching objectives. A commanding agent will evaluate those proposals against values, OKRs, and other strategic priorities, requesting revisions or additional review when needed before approving or rejecting them. Once approved, work is routed to an executor, whether that’s Claude, Codex, Gemini, PermaShip, or another system. From there, Nexus can verify quality, assess outcomes, and determine whether follow-on work should happen next. It has already handled a wide range of work for me: everything from smaller tasks like building and deploying a website with DNS and email, to large, complex multi-repo projects with staged feature releases. Over the past few months, it has been responsible for thousands of PRs. Since then, it has expanded far beyond software development alone. I believe this could be quite useful to a lot of people.
8
5
1
56
201,337