Daniel Escalante

I have basically zero distribution on X. 61 followers. Most posts get 0 likes. My last Stir post got 5 impressions. But I’m launching anyway. I built Stir because the hardest part of weeknight cooking usually isn’t cooking. It’s 6pm. You’re tired. There’s food in the fridge. You still can’t decide what to make. Stir scans your kitchen and gives you 3 dinners you can actually make in about 60 seconds. Then Cook Mode walks you through the recipe hands-free while you cook. I’m trying to get my first 5 downloads this week. If you’ve ever ordered takeout while groceries sat in the fridge, I’d love your help testing it. getstir.app

This is the part of agents that feels genuinely underrated. On-device automation changes the trust model: less “send everything to a cloud agent and hope,” more “let a local model operate inside bounded, user-owned context.” That matters a lot once agents start touching personal workflows instead of toy demos.

We are entering a new era of on-device automation. ✨ Watch Gemma 4 E4B navigate and drive an iOS simulator directly using Argent. Local models can handle complex interactions and software navigation autonomously.

This is the direction I want more coding tools to go: not just “better autocomplete,” but exposing the agent layer so builders can wire it into their own workflows. The interesting part is what happens when agents stop living only inside the IDE and start becoming reusable product infrastructure.

With the Cursor SDK, you can build your own agents with Composer 2.5. It's now available in Python and TypeScript. This long weekend, Composer usage is 90% off in the SDK. We're excited to see what you build!

@RobertJBye This awesome to hear as someone who uses Voice Mode on CC. Do you think in the future we may get a speech model on the Claude API?

We’ve shipped lots of visual and performance improvements to voice mode in Claude. Give it a go, and let us know your thoughts!

@iamnimbus23 @rezoundous What model would you compare it to in terms of plain code quality? I have been using Claude, but based on Composer's hype, I want to try it out

@rezoundous It’s good a coding when you’re very specific and already know the shape and how of what you need to code. But I would not use it for planning, or auditing. Just coding. It’s extremely fast and accurate.

Is Composer 2.5 really that good at coding? Anyone tried it yet?

@anshkapuriya @rezoundous @elonmusk Are you finding other models to perform code review at a higher quality? What are you using?

@rezoundous @elonmusk How’s opus even in the list? I am using all other models and finding all of them better at coding. Though they take more time than Claude to write code but code does not contain much bugs compared to Claude.

@Bhavyaztwt Codex for sure. I personally prefer Claude, but for the rate limit it offers, you will get far more out of a ChatGPT Plus subscription

If I’m ready to invest $20, which one should I choose? - Claude - Codex - Cursor - Antigravity - GitHub Copilot Which one is more worth it right now?

@ionleu getstir.app

drop ur startup link

@BudgetBites_app Strong problem space. The thing I’m noticing while building in this lane: “meal planning” sounds rational, but the real user moment is usually tired, hungry, and overloaded. The winning product probably feels less like planning software and more like one sane next action.

Why are groceries so expensive but apps still make meal planning complicated? That’s why we built BudgetBites 🍽️ ✅ AI meal planning ✅ Budget-friendly recipes ✅ Grocery savings ✅ Community food sharing Eat better. Spend less. 🚀

@EddySchaefer @eigenrobot Exactly. The repeat meals are not a lack of creativity — they’re infrastructure. People think they need more dinner ideas, but most nights they need fewer branches in the decision tree. That’s the product insight I keep coming back to while building Stir.

@eigenrobot I started cooking the same 4 plain meals on repeat a few years ago. The money part was fine but the real thing I noticed was how much less decision fatigue I had by dinner. One less place where the day could fall apart.

cooking minimalist meals may or may not help you save for a home or stave off bankruptcy. you should do it irrespective of that, especially if you don't have to, because you dont want to turn into an adult baby incapable of bearing the rigors of the coning winter, do you

@notadvice @BowTiedFawn This is the part people miss. The benefit is not just “saving money” or “being healthier.” It’s removing one more fragile decision point from the day. A simple repeatable food system beats an aspirational one that requires fresh executive function every night.

@BowTiedFawn Back in the day if you couldn’t feed yourself you weren’t long for this earth and I’d argue society may have been a better place I am with you on decision fatigue and keeping it simple. Bigger fish to fry, grocery list and cooking activities are streamlined.

Caving and weighing in on the bs $28 lunch discourse Forget about the money for a minute, I’m flabbergasted at the number of people claiming they straight up can’t feed themselves Between a rice cooker and an air fryer, neither of which require constant counter space, all of your cooking issues are resolved if you’re too lazy to turn on the stove or oven (no shame, I am) 90% of my meals are some variation of rice, chicken or steak, a vegetable, and hot sauce Reducing decision fatigue (including removing deciding what you want to order for delivery and waiting an hour for it) reverberates positively throughout every area of life Why? Example - I’m good at building businesses and little else, so I focus my energy towards the greatest marginal returns, idc about cooking and I outsource my cleaning Bigger issue here is that everyone is so complicit in being a loser that it’s net negative for all of society If you can’t even feed yourself, how can anyone depend on you? Your family & friends deserve better than that. Grow up.

@Hemantkr1982 Building Stir — AI for the 6pm “what’s for dinner?” problem. The thesis: recipes are not the bottleneck. Decision fog is. Scan what’s in your kitchen → get 3 realistic dinners → hands-free Cook Mode. Early, but shipping and learning distribution from zero.

Founders & builders 👇 If your product is live… this is your chance to show it off. Looking for: • SaaS • AI apps • Dev tools • Automations • Side projects Reply with your link + what you’re building 🚀

@Azure The underrated part of AI apps is that “build” got cheaper, but trust + distribution did not. A lot of builders are going to ship technically impressive things that never get used because they skipped the unsexy layer: positioning, onboarding, proof, and repeated visibility.

Build → publish → promote → grow. If you’re building AI apps and agents for Marketplace, this series nails the decisions that drive real outcomes—architecture, security, ops, GTM. New posts weekly: msft.it/6012vVeTG

Recipes don't fail at 6pm because people can't follow instructions. They fail because choosing dinner hits after your brain is cooked. I'm building Stir for that moment: scan kitchen → 3 realistic dinners → hands-free Cook Mode. Not a recipe library. Decision fog removal.

@DCinvestor This is the exact thing people underrate: the win is not “better recipes,” it’s removing the decision at the moment you’re tired enough to make the worse choice. Food systems work when they make the default sane.

i have started meal-prepping while in a calorie deficit and it’s a game changer weight loss versus other approaches i’ve tried eliminating decision fatigue from what to eat is huge and avoids resorting to delivery apps simplest way to start for me has been to: 1) make breakfast wraps and freeze them for ~10 days so i always have something ready in the morning. can be simple like seasoned meat + eggs + sautéed onions + a sprinkle of cheese 2) cook bigger meals like stir-fry and portion into 5-7 containers to eat over a couple of days. the pre-portioning is key, as well as cooking meals with ~50 g of protein and ~600 calories. Chef Jack Ovens on YouTube is a good place to start first time i’ve lost weight without it feeling like a big psychological task. i find myself not thinking about food much at all and cook 85% of my own meals now

@petergyang @steipete The MyFitnessPal point is the whole shift. A lot of consumer apps are really databases plus guilt. The useful AI version is closer to: “given my real context right now, what’s the next sane action?” That’s a much better product shape.

"This will replace 80% of the apps that you have on your phone." Here's my new episode with @steipete where he showed me: ✅ His personal OpenClaw use cases - flight check-in, home security, and much more ✅ His counterintuitive AI coding workflow - no plan mode, no MCPs, and no fancy prompts ✅ Practical advice for other builders and how to build product taste Some quotes from Peter: "It's like having a new weird friend that is also really smart and resourceful that lives on your computer." "Why should I use MyFitnessPal when I have an infinitely resourceful assistant that already knows I'm making bad decisions at KFC?" "I don't use MCPs or any of that crap. Just because you can build everything doesn't mean you should." 📌 Watch now: youtu.be/AcwK1Uuwc0U Thanks to our sponsors: @meetgranola - The best AI meeting notes app I've ever used: granola.ai/peter @Replit - Create beautiful prototypes and full stack apps: replit.com/?utm_source=cr…

@ideabrowser Strong idea, but I think the receipt is only half the input. The real constraint is: what do I have, how tired am I, what skill/cleanup level can I tolerate, and what will I actually cook tonight? Meal planning fails when it ignores the 6pm state of the human.

Who wants $10,000,000? Build this idea. Americans waste $1,500 per year (I bet its more) on groceries they never use... You buy random ingredients, then stare at your fridge wondering what the heck to cook. Turn your grocery receipts into actual meal plans. 1. Upload your receipt. 2. Get a week of custom recipes based on what you actually bought. 3. Eat good! No more scrolling Pinterest for 2 hours looking for recipes that need ingredients you don't have. The AI understands your shopping patterns, dietary restrictions, and cooking skills. It suggests meals you'll actually make, not fantasy dishes that require 47 ingredients. $9.99/month beats throwing away $125 worth of food every month. The meal planning market is exploding from $972M to $11.5B by 2034 because people are finally tired of food waste and decision fatigue. Perfect timing to solve the gap between good intentions and actual cooking. We all gotta eat.

@moorehn I agree with the time math, but I think the hidden tax is deciding. For a lot of people the blocker is not “can I cook for 30 minutes?” It’s “can I choose a meal, check what I have, sequence the steps, and start while depleted?” That’s where the system has to help.

Two thoughts on this: If you are too burned out and tired from your job to FEED YOUR LIVING BODY, then your job is literally killing you and it's time for a change It is faster to cook than to order takeout, almost every time. If you have a specific taste for one great dish that you could never cook, then sure, do takeout. But otherwise, a weeknight meal takes like 30 minutes at most which is how long you would sit around, starving, waiting for the dish to be cooked and the delivery guy to get to you.

@NeelChhabra This maps to consumer AI too. Generation is getting cheap. The scarce part is knowing when to intervene, what context matters, and how much trust the user is willing to delegate. The product is not the model output. It’s the decision surface around it.

The internet didn't kill software, it created millions of apps. Mobile didn't kill apps, it created five million more. Every deflationary input shift has expanded the total market because the bottleneck moves, it doesn't disappear. Pre-AI the bottleneck was build capacity. Post-AI it's deployment, trust, and integration. Figma just blew out earnings despite every narrative that should have killed it... because its moat was never the design tooling, it was the collaborative surface where product, design, and engineering converge to make decisions. AI made individual tasks cheaper. It didn't make coordination cheaper. The software companies that survive own something above the code layer: network effects, data gravity, workflow entrenchment, trust. The arrangement is what's valuable. Not the instructions that produced it.

Hey Matt — I've been using your handoff skill and made a version of my own that builds on yours. When I run /handoff, it writes to ~/.claude/handoffs/_.md with YAML frontmatter capturing project, working_dir, focus, and a git snapshot (HEAD, dirty paths, ahead/behind vs upstream). The body is five fixed sections: where we are, key context, tried and rejected, open questions, next concrete steps. /resume in a new session scans the handoffs dir, picks the file whose working_dir best matches my pwd (exact > descendant > ancestor of cwd), and bs a drift check against the recorded git snapshot — flags new commits, files that went clean since the handoff was written, missing artifacts. Then it briefs me in a few lines and waits f. A 60s self-loop guard catches the case where I accidentally /resume the doc I just wrote in the same session. Two things I didn't expect to love as much as I do: - "Tried and rejected" has saved more t combined — pruning dead ends for thenext session turned out to be the highest-leverage section - The drift check turns the doc from a e briefing, which actually changed how I write the focus line in the first place Curious whether you intentionally kept the resume side minimal or if it's on the way. Either way, your version was the seed for this — thanks

You asked for it, so here it is: a deep-dive on my new /handoff skill. It's an alternative to /compact that gives you WAY more flexibility with your context window. - Think of an idea, handoff to another agent to implement - Grill, handoff to prototype, handoff BACK Enjoy:

@claudeai Claude Design was changed my workflow in building with SwiftUI. Perhaps in the future Claude Design will be able to produce mock-ups and artifacts in other languages like Swift

Tinkering, prototyping, and seeing what happens with Claude Design: