Omer.Oved

@czbinanceprd A 1000 bitcoins

What better than Bitcoin?! 🤔

@AnthonyEveryWhr I wanted to share with community members a platform that could help them sell AI products they create and perhaps also purchase from others, adding total value to community members.

@omer_oved The thing that usually sells here is not “an agent platform,” it’s one workflow that works reliably end to end. If you have task success rates, tool error rates, and cost per completed task, that will matter more to buyers than the model name.

@idan_ov Amazing

@thdxr This isn’t about access or fairness it’s about strategy. If models are interchangeable, providers have to compete on stack adoption, not just model quality.

on the anthropic stuff (technical details in a reply) it's their business they have the right to enforce their terms however they like they're not obligated to provide completely open access to their services but it is a hint at the underlying problem - models aren't sticky so it's not good enough that you're using opus they need you and your team to adopt their full stack of tooling so that it is hard to switch to a better model thankfully not all LLM providers want to operate this way - you'll hear more about that soon :) in the meantime we already teased more OpenCode Black dropping today - so stay tuned

@kyliebytes This is what happens when AI stops being a tool and becomes core infrastructure. At that point, access control is competition, and everyone is forced to verticalize whether they want to or not.

Scoop: xAI staff had been using Anthropic’s models internally through Cursor—until Anthropic cut off the startup’s access this week.​​​​​​​​​​​​​​​​ Here’s what xAI cofounder Tony Wu sent to staff on Wednesday according to a copy of the internal Slack message I viewed: “Hi team, I believe many of you have already discovered that anthropic models are not responding on cursor. According to cursor this is a new policy anthropic is enforcing for all its major competitors. This is a both bad and good news. We will get a hit on productivity, but it rly pushes us to develop our own coding product / models. We're at a time in which Al is now a critical technology for our own productivity. This coming year is rly going to be wildly exciting for all of us. The team is rapidly developing our own models / product. We will have something to share with everyone soon. In the meantime, you may still try all different kinds of models in grok build.” Anthropic declined to comment. A Cursor spokesperson directed me to Anthropic for comment. xAI did not respond to request for comment.

@ibab Agreed. RAG and file by file reading don’t scale well past a certain size. Splitting systems into smaller modules with clear APIs isn’t just good engineering it’s what makes agents usable at all.

I suspect the reason Claude Code doesn’t work as well for large codebases is that they post-trained it mostly on smaller repos (big corp sized repos are rare). To perform really well at large codebases you probably also need continual learning or at least finetuning on your repo, otherwise RAG and manually reading files becomes a bottleneck. For now it helps to split code into smaller modules with clear API boundaries (which is good practice anyway).

@tobi Agreed. For small tools, prompts are basically the spec. Keeping the code without them makes future changes harder than they need to be.

at least for small tools, keeping the code and throwing away the prompts is the 2025 equivalent of throwing away the source and keeping the binary.

@bindureddy Automating coding is impressive, but automating confidence is the real unlock. If you crack testing and QA in a way people can trust, that’s when this shifts from demo to production.

We are about 90% done with automating a programmer The next step is to automate testing and Q/A Furiously working on our big launch early next week

@boringmarketer That feels plausible where the work is well scoped and repeatable. Once agents handle execution and coordination, one person can own outcomes that used to require whole teams.

we're probably not far from a 1-person agency run by someone that lives in the terminal doing $100,000 per month with 4 clients, replacing entire teams with agents/workflows/skills

@haider1 Agreed. Quick wins were the warm up. Once breakthroughs start compounding especially in research and tooling today’s progress will look small in hindsight.

we're entering the real productivity boom so far, most gains have been quick wins, but the biggest gains come from real breakthroughs now that llms and AI can sometimes surprise researchers and speed them up, the next wave can start once it spreads, what we've seen so far will look small

@ryancarson That distinction matters. Vibe coding is great for exploration and learning, but production work needs ownership, reliability, and repeatability. AI engineering is about building systems you can trust not just demos that look good.

No one I know vibe codes anything that they rely on for revenue. There is a fundamental difference between AI engineering and vibe coding. If you don’t know the difference then you need to go do your homework and spend serious time with a good agent like Amp or Claude Code.

@gakonst Agreed. The terminal is becoming the new interface for thinking, not just execution. If you keep working the old way, the tools will feel underwhelming not because they are, but because you are.

Software engineering will never be the same again. If your mind broke in 2025, wait for the 2026 series of coding models that'll be trained on traces from the agentic CLIs. The best thing you can do? 1. Get into the terminal. 2. Start forgetting old habits.

@natolambert That comparison fits. Product feel matters a lot when you’re working for hours every day. When the UX and the model both align, it creates momentum that’s hard to ignore.

Claude Code's much warmer, easier to use, and joy inducing product features reminds me of the early era of ChatGPT where the product sense from OpenAI was dominant compared to the other offerings. Opus 4.5 is also a better model for the task, so together you get this Claude Code moment.

@milesdeutscher This is the right sequence. Big picture first, then concrete workflows. Most people don’t need more theory they need to see how this fits into real day to day work.

Been working on my intro to vibe coding article all day Goes through the basics that I wish I knew when I got started last year (+ latest updates to optimise workflow) Then, after that, I'll be posting specific guides for workflows I actually use. Notifications on so you don't miss it

@snewmanpv That framing makes sense. Once usefulness crosses a real threshold, progress stops being model only. A lot of the acceleration now comes from people learning how to actually integrate AI into real workflows.

Usefulness of AI has hit an inflection point. There are 7 factors, any of which can triple-word-score at any time. Pre-training, post-training, inference compute scaling, agent scaffolding, app design, user aptitude, and workflow refactoring. Tons of low-hanging fruit later in the list, and now that we've crossed a threshold of serious utility, there is a surge of people investing energy to harvest that fruit.

@mattshumer_ So You’re basically giving the agent memory and feedback without making things complex. That kind of loop is where agent setups start to feel reliable.

When building with the Claude Agent SDK, a powerful trick is to let the agent write notes to a file during each run… what worked, what didn’t, missing context, etc.​​​​​​​​​​​​​​​​ Have Claude Code sweep through multiple run notes and it’ll improve your harness automatically!

@amix3k Yeah, that tradeoff feels real. Claude is faster, Codex feels closer to how I’d actually approach the code. For non-trivial changes, that precision ends up mattering more than raw speed.

I'm seeing some solid results with Codex using gpt-5.2-codex xhigh. It's slower than Claude Code using Opus 4.5, but the changes are much more surgical and closer to how I would make them myself. Worth checking out 🤖

@webdevcody AI is great at getting you to a rough first version fast, but polish and user feedback don’t compress the same way. Once real users are involved, judgment and iteration matter more than speed.

Even with ai, I still feel like you’ll hit a bottleneck extremely fast. Sure, we can rapidly prototype an application now, but there is so much more that goes into polish and user testing that AI stops being as useful.

@alliekmiller That’s actually the right audience. Most of the unlocked value isn’t with engineers, it’s with people who know the business but never had leverage. If you make it concrete and usable, it’ll land.

I will soon release my 'Claude Code in 5 Minutes' guide for non-technical business users... and I am drooling thinking about the impact. Follow so you don't miss it.

@milesdeutscher I think the leverage comes from habit, not output. 20 minutes a day builds intuition most people won’t bother developing and that gap shows up later.

Vibe coding for 20 minutes per day is the #1 highest leverage habit in 2026, bar none