Dagens Vibes — 3. juli 2026

Mitchell Hashimoto@mitchellh

I'm having a lot success using Fable xhigh as a planner/architect, using GPT 5.5 xhigh (subscription) as a coder, then Fable xhigh again as a judge. At API pricing, planning+judge costs are in the ~few dollar range compared to typical $50+ full round trips. I've seen some others using dumber/cheaper coders, but GPT 5.5 even at xhigh compared to Fable 5 is very cheap and very fast. And GPT 5.5 is just... really good. Still been less than 24hrs since the re-release so the longevity of this approach is unclear, but its been working really well.

Peter Yang@petergyang

I think the days of companies relying only on frontier models are coming to an end. 1. Tokenmaxxing at frontier-model API prices makes no sense. Uber burned through its entire 2026 AI budget in 4 months, Microsoft moved engineers off Claude Code due to cost, and companies are realizing that running everything on frontier models can get expensive fast. Tokenmaxxing makes sense when you’re on a subsidized $200/month plan, but it’s unsustainable at API rates. 2. Companies will rely on a portfolio of models. Coinbase recently cut its AI spend nearly in half by switching engineers to Chinese open-source models like GLM and Kimi. Airbnb and Pinterest have done the same with Alibaba’s Qwen models. I believe this will be the default path forward: using frontier models for high-stakes work and cheaper models for everything else. 3. China’s open-source strategy is working. Chinese models are taking market share from frontier models at US companies. China is also building the full AI stack, from energy, like solar and nuclear, to data centers to domestic chips. The Chinese government is planning a $295B investment in AI data centers, with at least 80% of chips built domestically. 4. Frontier labs are in a Catch-22 situation. If they release great open-source models, they might undercut their own frontier API revenue. If they gate the best models behind a trusted list, companies will just lean on open alternatives more. The last major US open-source model was OpenAI’s gpt-oss series back in August 2025, which already feels like decades ago in the AI space. 5. The US needs to think about its AI strategy holistically. I believe restricting access to frontier models will only hurt American innovation. Banning US companies from using Chinese models won’t work either. Just look at how China took over the global electric vehicle market. To maintain our edge, we need the best closed and open models while scaling our energy grid and data centers much faster. 📌 More in my recent essay: https://t.co/jj568Tx1NU

Citeret tweet

Peter Yang@petergyang

Here's my new post with 18 hot takes on where AI is headed next: 1. The frontier-only AI stack is collapsing Companies will use a portfolio of models to save costs, with everyday tasks defaulting to low-cost open models from China. 2. The AI super app era is here Codex, Claude, Cursor, and others are competing to disrupt all knowledge work by using agents for everything you can do on a computer. 3. Traditional software risks becoming a dumb pipe for agents Agents will become the default user. They’ll use APIs and browsers to access your website and app without humans ever seeing them. 4. Cloud agents and collaboration are the next wave The future is agents living in the cloud, accessible from any device, and collaborating closely with both you and your team. This will happen very soon. 📌 Read all 18 takes now for free: https://t.co/jj568Tx1NU

♥ 53↻ 4💬 16🔖 45

Machina@EXM7777

this new AI research just dropped and it's kind of insane if you use AI agents... a tiny model that can't answer a single question just beat ChatGPT, Gemini and Claude at the same time... let me break down what this means for you: right now we all do the same thing... we pick the "best" model and throw everything at it the team behind this research did the opposite they built a tiny model, small enough to run on your laptop, and gave it one job: be the manager it never answers anything itself, it just reads your task and decides which big model should handle each piece picture a conductor in front of an orchestra, they can't play a single instrument, but they know exactly who should play when that's the whole trick this is how it runs one task: > picks the best model to plan the work > picks the best model to do the work > picks the best model to check the work > loops until the checker signs off and this is where it gets stupid the little manager beat every one of those frontier models and set a new record on a hard coding benchmark then they tried the obvious thing and put a top-tier model in the manager seat instead... and it got worse so the takeaway is almost annoying in how simple it is: being the smartest in the room doesn't matter, knowing who to ask does now the part that changes how you work if you're in agents every day: - stop betting everything on one "best" model, a team of models beats any single one of them - never let a model check its own work, it has the same blind spots it had when it wrote it, so it waves its own bugs right through, use a different one - the highest-leverage habit you can build is a "check" step before you ship anything, it quietly kills most bad outputs - the future isn't one giant brain, it's a small manager pointing the right models at the right jobs

Citeret tweet

Machina@EXM7777

https://t.co/CDlzD64NSe

♥ 402↻ 38💬 19🔖 976

Thorsten Ball@thorstenball

I wrote down how we made our codebase work for agents in remote machines. Orb setup, custom dev server tooling, /__dev endpoints for auth, log tricks, 41 AGENTS.md files. It's all in here. https://t.co/dIKhR5Bx8t

Rhys@RhysSullivan

launching https://t.co/5F8Xyzg1Hw today! it's an open source catalog of every products MCP / API / CLI / GraphQL server and how to authenticate to them deep links to generate api keys, 1 click copy spec urls, it's still early but i've been loving having it https://t.co/bfVcPwXAyX

David Soria Parra@dsp_

Super glad to see more browsers exposing MCP servers: Safari is introducing an MCP server in their Technology Preview 247. https://t.co/hnnuWFJNQv

eric provencher@pvncher

Pro tip if you're using the codex app. When you're deep into implementation and want to steer a change, open a side chat, and ask codex about it, then ask it to write itself a prompt to turn this into the preferred design / architecture, and then paste that prompt back into main

The Pragmatic Engineer@Pragmatic_Eng

Anyone who thinks software engineering is ‘going away’ doesn’t understand the job. @KentBeck, creator of XP and TDD, on why Dario has it wrong: [Gergely: Dario said, I quote, ‘coding is going away first, then all of software engineering’. ] “That's a statement by someone who doesn't understand software engineering. Coding is part of what you're doing, but it's only a small part of what you're doing, even if it takes up a fair amount of time. You're building confidence, you're building connections with other people, you’re building your own understanding. All those things are happening while you're coding. And coding's actually a great way to cement understanding. The more you program, the more you understand the domain that you're working in. And so to say, well, we're just going to pass all that off to a machine. Well, that's not all there is to it. A couple of days ago I saw a phrase, and it really hit me, that we're accumulating code faster than we're accumulating trust now. And that sense of trust comes from me struggling to understand some domain concept, ah, I get it! I represented it in the code. I write tests that demonstrate that I really did understand it and now, I trust my program. If we're programming together, that act of programming together means that we trust each other more. And none of that can be automated. None of that occurs. If we prompt, we get the finger guns, the genie goes, yeah, it's all finished, boss. And it is like, well, hang on, finished. What's finished?”