Ciprian Grigore

Just wrapped an intense day at the #GenerativeUI Global Hackathon! Built Helios Deal Desk, a B2B exception-approval workflow with no chat box. The agent reads each deal, picks the panels that matter for THIS deal, and renders them live. Click ✨ on any panel to drill deeper, and a follow-up panel spawns nested under it — UI begets UI. Stack: A2UI v0.9 declarative components + @CopilotKit @GoogleDeepMind with native function calling.

@signulll I did an experiment with Claude Code, Gemini, Codex and Opencode. Gemini did not do well. octokraft.com/blog/coding-ag…

not a single person i have ever spoken to uses gemini for coding. this is still very very weird. why is gemini so bad at coding when google has scoured the web full of code for decades?

@GergelyOrosz I think a lot of people won't be using UIs anymore, at least not for things like engineering and data. Unless the UI can give you the answer you need immediately, might as well ask an agent to use the tool.

Prediction: The next 12-24 months, "UX-pilled" builders will be in massive demand. Who can create intuitive interfaces, web+mobile+desktop apps that "feel good," natural, fast, and far better than the competition. THIS will be the difference vs those building "just" with AI.

@SimonHoiberg I think the point is to use those and figure out if anybody wants to use your app.

"Just use Vercel." "Just use Supabase." "Just use Clerk." Cool. Now your auth, database, and deployment are owned by 3 different companies who can change pricing whenever they want. And the rest of your product is wrapping OpenAI. At some point you have to ask yourself: what do I actually own here?

@levelsio Claude code + cloudflare Cli. Cloudflare has become too big for UI

The main thing I miss on Cloudflare is a single [ Block IP ] button It boggles my mind it's not there, you can see a traffic spike and a person doing 50,000 requests per minute, but you have to write a security rule and go through 10 steps to block it Too difficult Just let me [ Block IP ] from the Security Analytics or general Analytics dashboard @daneknecht @eastdakota

@photomatt @garrytan Yes please, that amount of slop does need WordPress to take it to the next level.

I disagree with the @garrytan hate. The cool thing about this era of development is that he could point his agent at this thread and say, " Fix all these problems and it would be solved in 10 minutes. That's amazing! I do think there's an interesting point, though, that his agent is probably dealing with too much context is doesn't need to. If this was built on top of @WordPress his agent could just focus on the content and design and it'd inherit a bunch of best practices, etc. But we need to make it easier for him to point his agent at a repo and say whether WordPress is a good fit for his goals, how he could leverage it most effectively, and deploy easily. But you're missing the point! The SOUL.md of garryslist.org is to change hearts and minds of people to support the California boom loop. Knowing that we can just say: Hey, you'll reach more people to hear your messages if your site loads faster. Stop making fun of LOCs, no one cares. :) And @garrytan I DM'd you but if you want to jam a bit together to put this on WP and fix all these issues I'm happy to help.

@GergelyOrosz The problem is not that he's coding with AI, it's great more people are doing it. It's that he's making some wild statements, dunking on others that actually produce things of value and using his authority position to promote things like LOC as a metric.

Perhaps controversial: I think it is a net good that CEOs and founders who are engineers at heart are back to coding with AI agents. Lots of dunking happening eg on Garry Tan. He's stepping outside of his comfort zone, building net new things + sharing in public.

@gvanrossum And a loop or some kind of runtime

I think I finally understand what an agent is. It's a prompt (or several), skills, and tools. Did I get this right?

@vineerpasam Yeah I'm just too lazy to open another terminal and do it myself

Do you guys allow agents to commit and push your code?

Is the harness really more than system prompts and tools available? I think anthropic proved that less is more and smarter agents perform better when they are not bloated with instructions and tools. Basic tools like sed, grep and awk are much more present in the training data than some custom diff generator.

One interesting dynamic in AI is infra+application co-dependence. An older version is hardware (infra) + LLM (app): - Architectures that work well with current-gen GPUs+TPUs work better because they scale - Hardware makers optimize for the current architectures because that's what everyone is using Now we have a new version LLM (infra) + agent harness (app): - Agent harnesses optimize to work well with current popular language models - Language models optimize to work well with current popular harnesses Arguably, these interdependences are both inevitable (the benefits are obvious), and bad (lock-in prevents innovation). Probably the best that all parties can do is just try to be as flexible as possible within reasonable limits. For LLM trainers, that means training on various agent harnesses, which different companies have done with different levels of success. The issue is that given a fixed number of parameters, it can also hurt performance if you're not careful. For instance, in our agent data protocol (agentdataprotocol.com) paper initial experiments failed to work well when training on multiple harnesses simultaneously. For agent harness creators, this means adjusting to various LLMs. At OpenHands we try to do our best at this, we benchmark on various models (index.openhands.dev), and also do some model-specific adjustments, like modifying tool formats, prompts, and file editing tools depending on the LLM. I'd also love to set up an automatic pipeline to do this as well, although cost is a barrier there. I'd be interested in hearing what others are doing here too!

@josesaezmerino I would have agreed until I got a ultra wide monitor, now vertical space is what I'm lacking

I don't understand vertical tabs. Why would I want less space for the website content I actually want to look at?

If you're not building an agents orchestrator are you even coding anymore?

@denisyurchak I have a similar setup but ssh through my phone sucks, so I built this, github.com/amurg-ai/amurg You can just spin up runtime on each server and connect them to one hub, then just talk into your phone.

I have 6 Hetzner servers – separate for OpenClaw, one with Claude to code on my projects, and several for my websites. I am running Claude on each one of them, and I noticed that each time I spin up a new server, I have to do the same things for security, Tailscale setup, etc. I created a .md file with the prompt for Claude that gives you a secure VPS you can use with Termius in one prompt. It sets up SSH, fail2ban, UFW firewall, and optionally Tailscale. You literally just: 1. Buy a Hetzner VPS 2. Install Claude and give it the prompt Enjoy! The link to the file below Feel free to suggest improvements and raise PRs if you notice how it can be made better and more secure! And huge shout-out to the main source of my VPS education @levelsio

@Dominus_Kelvin And now replace all that with hetzner bare metal and k8s

Replace Vercel with Docker + VPS. Replace Netlify with Docker + VPS. Replace PlanetScale with PostgreSQL on a VPS. Replace Neon with PostgreSQL on a VPS. Replace Upstash with Redis on a VPS. Replace Meilisearch Cloud with Meilisearch on a VPS. Replace Algolia with Meilisearch on a VPS. Replace Supabase with Postgres + Auth + Storage on a VPS. Replace Railway with Docker + VPS. Replace Render with Docker + VPS. Replace Fly.io with Docker + VPS. All you really need: • Docker • A cheap VPS • PostgreSQL • Redis • SMTP

@camsoft2000 What if you could save the project without a rewrite?

I’m getting to the point with one of the projects I work on where the complexity of AI slop is becoming a real issue. While I can still happily prompt the agent to add x feature and it will do so and it will likely work perfectly, the code is just getting too complex and fragmented. Agents love to copy and paste and keeping patterns DRY is a real challenge. The agent will start diverging all those copy and pastes until you’ve got loads of similar but slightly different blocks of logic. Again it all still works and solves the problem I’m after. But I just can’t get any kind of consistency anymore, the code is a mess and I just don’t have a handle on it. I want a clean unified architecture but agents just code with tunnel vision. The project is now too big and complex for an agent to fully reason with and too big and complex for me to reason with. The only real solution is a complete rewrite. Maybe this is the way things will go. Code will just become disposable. I don’t really want to care about the code and to be honest I don’t but I do care about consistency and maintainability and the AI slop is hurting those very things I do care about. I know some will say “I’m holding it wrong”, use x,y,z skill, tool whatever and already use tools and anti slop skills, plans, docs, etc but the outcome is the same. Vibe coding something into existence is truly magical. But turning it into a mature product with months of iterations is painful. I can’t even hand code this thing because I don’t understand the code anymore and I’m too lazy to try and code myself because I’m addicted to AI. So what’s the solution, either start again and accept that’s just the way we have to roll, or just carry on fighting the slop and accept each new feature will take longer to implement than the last. I’m tired. I’m addicted.

@thdxr That's absolutely fine on your own, and most likely you want to do both, but if you're building in a team in a complex system that has multiple organizations and teams that depend on a repo, just a devs local review is not enough.

i still don't get why we need to push code up to get an LLM review via awkward github ui hacks opencode has /review which can also do things like run your code to check things but a full time team focused on this would do it better, i just don't like the workflow they offer

@TheGeodexes @tannerlinsley But why, when I can just start a session in a tmux pane?

@ciprian_cgr @tannerlinsley OpenCode SDK, Claude Agent SDK, Codex probably has something too. Pi has an sdk too.

Ghostty was fun, but time for something else. I still love opencode, too but with CC plans dead on it… I’m feeling lost. Full GUI? T3 Code? Opencode GUI? Warp? Back to cursor? Try CC again? Raw Codex? My 🧠 hurts and I just need to keep shipping.

@TheGeodexes @tannerlinsley I'm probably complicating things for fun, but which sdk do you mean?

@ciprian_cgr @tannerlinsley Why not just use the SDKs?

@ns123abc Obligatory meme