The AI industry has a climate problem it doesn’t love talking about on Earth Day.

The International Energy Agency (IEA) projects global electricity consumption for AI operations will hit 800 terawatt-hours (TWh) in 2026. That’s about what Japan uses in a year. In the U.S., data centers are on track to consume somewhere between 6.7 and 12 percent of national electricity by 2028. In Ireland, the share is already over 20 percent. Training one frontier model (the largest and most capable AI systems, built by labs like OpenAI, Anthropic, and Google) can draw as much power in a single run as hundreds of American households use across an entire year.

None of this is a reason to stop. But it is a reason to ask a sharper question.

Is AI worth the electrons?

Subscribe now

The answer is domain-specific.

For drug discovery, protein folding, climate modeling: yes, obviously. For most consumer AI features bolted onto apps that didn’t need them: probably not. For generating a blog post about your cat: almost certainly not.

For food waste, the numbers are interesting. And the companies are already there.

Afresh announced $34 million in new funding yesterday. They build AI for grocery, specifically the fresh perimeter (produce, meat, dairy, bakery, and other short-shelf-life categories), where most of a store’s waste happens. They’ve prevented more than 200 million pounds of food waste to date, working with Albertsons, Meijer, and Wakefern. Winnow is doing parallel work in commercial kitchens: measuring what chefs actually throw out so they can see their own patterns and adjust. Flashfood has partnered with Kroger, Loblaws, and others to mark down specific items before they spoil, instead of discounting whole categories. Apeel Sciences developed a plant-based coating that extends produce shelf life. Too Good To Go runs one of the largest consumer marketplaces for end-of-day surplus from restaurants and shops. Mill processes home food scraps into chicken feed.

Some use AI; some don’t. All of them share one feature: they’re trying to close a specific information gap in a specific place in the system.

ReFED, which publishes the definitive annual report on U.S. food waste, flagged something important in this year’s edition. AI’s proven wins on food waste so far are on the commercial side. Forecasting. Inventory. Routing. In the home, the tools are still early. And the home is where the waste actually lives. Households account for 30 to 40 percent of the U.S. total, more than restaurants, grocery retail, or farms.

That gap is where we’re working.

At Agile Defense, we built systems for situational awareness (SA) to enable domain visibility. Same principle applies to a refrigerator. You cannot decide about what you cannot see.

The same pattern keeps showing up in consumer tech. Before budgeting apps like YNAB and Monarch, most people had no real picture of where their money went, so they overspent in ways they didn’t intend. Once the picture became legible, behavior changed. A fitness tracker doesn’t make anyone want to walk more; it just finally tells them they aren’t. A smart thermostat doesn’t demand virtue; it shows you the heating bill you thought you were avoiding.

These tools don’t lecture. They don’t rely on willpower. They show you what’s there and let you do the rest.

A kitchen is overdue for the same treatment.

One banana, one recipe

A back-of-the-envelope on what “worth the electrons” actually means.

Consider a single banana going off in the back of the drawer.

Growing, shipping, refrigerating, and packaging it produced roughly 110 grams of CO₂e (carbon dioxide equivalent, a unit that rolls multiple greenhouse gases into one comparable number). If that banana ends up rotting in a landfill instead of being eaten, it generates another ~100 grams of CO₂e as methane, which is about 28 times more potent than CO₂ over a century. Round it down to ~200 grams of CO₂e per banana you failed to eat.

Now the other side of the ledger: an AI recipe query.

A text query to a large language model uses somewhere between 0.3 and 3 watt-hours (a watt-hour is the energy it takes to run a one-watt device for one hour; a typical LED bulb uses around 10 watt-hours per hour). OpenAI’s Sam Altman pegged the average ChatGPT query at 0.34 watt-hours. A recipe generation is a little heavier than average. Call it 1 watt-hour to be conservative. At the U.S. grid average of roughly 400 grams of CO₂ per kilowatt-hour, that works out to about 0.4 grams of CO₂e per query.

So,:

• One wasted banana ≈ 200 grams CO₂e

• One recipe query ≈ 0.4 grams CO₂e

• Ratio: saving one banana is worth about 500 recipe queries

If AI helps you save the banana, the math isn’t close. It’s not even on the same ledger. Five hundred “what can I make with this?” queries pay for themselves in a single piece of fruit you were going to throw out anyway.

Earth Day is a good day to be honest about both possibilities.

AI is not automatically good for the climate. Most of what gets branded “AI for climate” is marketing. Some of it will land. Most of it won’t.

The places it will land are places with huge, specific, measurable inefficiency, where better information translates directly into less matter wasted. Food is one of those places. That’s the bet. And a lot of good companies are already making it.

In the meantime: open the fridge. Pull out the back row. Throw away what’s gone. Plan dinner around what’s left.

That part works regardless of what happens to AI.

Happy Earth Day.

—Jay

springhouse.co

On Tuesday, a jury in Santa Fe, New Mexico ordered Meta to pay $375 million after finding the company misled users about the safety of its platforms for children. Six weeks of testimony. Psychiatric experts. Whistleblowers. A recording of Mark Zuckerberg’s deposition played in open court. The prosecution’s closing argument: “The safety issues weren’t mistakes. They were a product of a corporate philosophy that chose growth and engagement over children’s safety.”

One day later, just this week, a jury in Los Angeles found Meta and YouTube negligent in a landmark social media addiction trial, awarding $6 million in damages to a young woman who became addicted to the platforms as a child. The jury found that both companies designed their apps to hook young users and failed to warn them of the dangers. The plaintiff’s lawyers argued that features like infinite scroll, autoplay, and notifications made the platforms a “digital casino” that children couldn’t put down. The lead attorney called it “the engineering of addiction.” Experts are calling it social media’s Big Tobacco moment.

One of the witnesses who took the stand in the New Mexico trial: Aza Raskin, the man who invented infinite scroll. He testified against Meta about a feature he designed twenty years ago for blog posts.

Last month, the European Commission ordered TikTok to disable infinite scroll or face fines of up to $2.1 billion, ruling that its addictive design violates the Digital Services Act — the first time any government has declared social media design architecture illegal. New York, California, and Minnesota now require warning labels on apps that use infinite scroll, autoplay, or algorithmically driven feeds.

Two jury verdicts in two days. An EU enforcement action. State-level legislation across the country. The architecture of the modern internet is being challenged on every front.

But to understand how we got here, you have to go back to where it started.

Subscribe now

The invention no one asked for

In 2006, a young interface designer named Aza Raskin was using Google and getting frustrated. Every time he reached the bottom of a page of search results, he had to click to load the next page. It was clunky. It broke his flow. So he had an idea: what if the page just kept loading? What if, as you scrolled down, more content appeared automatically — no clicking, no waiting, no end?

He called it infinite scroll.

Raskin designed it for a simple, practical purpose. He was thinking about blog posts and search results. If you hadn’t found what you were looking for, why make you click? Just load more. It was, by any reasonable measure, a better interface.

He pitched it to Google and Twitter. They adopted it. And then social media happened.

What Raskin didn’t foresee — what he says he was “completely blind to” — was how infinite scroll would interact with business models built on attention. Social media companies don’t sell products. They sell your time. The longer you stay on the app, the more ads you see, the more money they make. Infinite scroll removed the one natural stopping point — the bottom of the page — and replaced it with nothing. There was no end. There was no signal that said “you’re done.” There was just more.

Raskin has since called infinite scroll “one of the first products designed to not simply help a user, but to deliberately keep them online for as long as possible.” He’s estimated that time equivalent to 200,000 human lifetimes is wasted every day because of it. He has apologized publicly, multiple times, for what he created.

But infinite scroll was just the mechanism. The philosophy behind it was already taking shape.

The News Feed changes everything

On September 5, 2006 — the same year Raskin invented infinite scroll — a Facebook product manager named Ruchi Sanghvi shipped a feature that would reshape how billions of people consume information.

It was called the News Feed.

Before the News Feed, Facebook was essentially a digital phone book. You had a profile. Your friends had profiles. If you wanted to know what someone was up to, you went to their page and looked. It was static, deliberate, and user-directed. You chose what to see and when to see it.

The News Feed flipped that model. Instead of you going to the information, the information came to you — in a live, continuously updating stream. Every time a friend posted a photo, changed their status, liked a page, or friended someone new, it appeared in your feed. You didn’t have to do anything. You just opened the app and content was already there, waiting.

The initial reaction was outrage. Users called it creepy, invasive, a violation of privacy. There were protests. Petitions. But Facebook kept the feature, and within weeks, engagement skyrocketed. People couldn’t stop checking it.

In 2007, Facebook added the Like button and began experimenting with algorithms — deciding not just what to show you, but in what order. By 2009, posts with the most engagement were being pushed to the top. The feed was no longer chronological. It was optimized.

Optimized for what? Time spent. The metric that mattered — to Facebook, to advertisers, to investors — was how long you stayed on the platform. Every algorithm update, every design tweak, every A/B test was oriented toward a single question: how do we keep them here longer?

A journalist at Slate later called the News Feed “the inspiration for just about every social-media feature that has come along since.” He wasn’t exaggerating. Instagram, Twitter, Pinterest, TikTok, YouTube, LinkedIn… every major platform adopted the same model. A feed. An algorithm. Infinite content. No end point. Maximum time on app.

The slot machine in your pocket

By the mid-2010s, a former Google design ethicist named Tristan Harris was sounding the alarm.

Harris had spent years inside Google studying how technology affects attention, well-being, and behavior. What he saw disturbed him. In 2013, he wrote a 141-slide presentation titled “A Call to Minimize Distraction & Respect Users’ Attention” and shared it with colleagues. It went viral inside Google, viewed by tens of thousands of employees.

His central argument was simple and devastating: smartphones are slot machines. Every time you pull your phone out of your pocket and check for notifications, you’re pulling a lever. Sometimes there’s a reward — a like, a message, a new post. Sometimes there’s nothing. That unpredictability — what psychologists call ‘intermittent variable rewards’ — is the most addictive reinforcement schedule known to science. It’s the exact mechanism that makes casinos profitable.

Harris pointed out that this wasn’t accidental. App designers knew exactly what they were doing. As he put it: tech companies were “taking behavioral cocaine and sprinkling it all over your interface.” The business model demanded it. If your stock price depends on time-on-app going up, you will find ways to keep people hooked.

In 2018, Harris and Aza Raskin co-founded the Center for Humane Technology. The inventor of the thing and the person who understood what it was doing to us, working together to undo the damage. That tells you something about how far off the rails things had gone.

For years, their warnings were treated as interesting but largely academic. Tech companies made token gestures toward “digital wellbeing” and kept optimizing for engagement. As the verdicts this week make clear, that era is over. The architecture itself is now a legal liability.

But here’s what none of these lawsuits, rulings, or warning labels address: the same design pattern doesn’t just live on social media. It lives everywhere, including the apps you use to figure out what’s for dinner.

The paradox that explains your kitchen

While the tech world was building the attention economy, a psychologist named Barry Schwartz was studying a different but deeply related problem: what happens when people have too many choices.

In 2004, Schwartz published “The Paradox of Choice: Why More Is Less.” His thesis was counterintuitive: more options don’t make us happier. They make us more anxious, more indecisive, and less satisfied with whatever we ultimately choose.

The research that crystallized this was a jam study. Researchers set up a tasting display at a grocery store. When they offered 24 varieties of jam, 60% of shoppers stopped to look, but only 3% bought anything. When they offered just 6 varieties, fewer people stopped, but purchases went up tenfold. More choice attracted attention but paralyzed action.

Schwartz connected this to a broader epidemic of decision fatigue. Every decision you make — from what to wear to what email to answer — depletes the same cognitive resource. By the end of a workday, your capacity to make good decisions is diminished. And then you walk into your kitchen and an app hands you 2,400 options.

But there’s a study that connects these ideas even more viscerally — and it involves food. Researchers gave people soup bowls that secretly refilled from the bottom. The people eating from these bottomless bowls consumed 73% more soup than people with normal bowls. And here’s the remarkable part: they didn’t believe they’d eaten more. Without a signal that the bowl was empty, without a stopping point, they just kept going.

Infinite scroll is a bottomless bowl. And your recipe app is serving you soup you never finish.

Every food app inherited the feed

Here’s what nobody talks about: the same design architecture that powers doomscrolling on Instagram is the foundation of almost every cooking and recipe platform.

It’s 6:17 PM. You’re standing in your kitchen. There’s chicken in the fridge, some vegetables in the drawer, half a bag of rice in the pantry. You open a recipe app. You type “chicken.” 2,400 results. You scroll. Honey garlic chicken. Chicken tikka masala. Chicken piccata. Each one looks good. Each one requires ingredients you’re not sure you have. You tap one, scan the ingredient list, realize you’re missing shallots, go back. Tap another. “Marinate for 4 hours.” It’s 6:17. Back. Tap another. The photos are beautiful. The recipe has 847 five-star reviews. You start reading the comments. You’re now three minutes in and no closer to dinner.

You close the app. You open Uber Eats.

This scene plays out in millions of kitchens every night. Most people assume it’s their fault — they’re indecisive, they should have planned better. But it’s not their fault. It’s the design.

Think about how recipe apps work. You search for something. You get a feed of results. You scroll through them. Each result is designed to catch your eye — beautiful photography, compelling headlines, star ratings. The app wants you to keep browsing, keep saving recipes to collections you’ll never revisit, keep engaging.

The business model is the same as social media: time on app. Recipe platforms make money through ads, sponsored content, and premium subscriptions. The longer you scroll, the more ads you see. A user who finds what they need in 30 seconds and leaves is a failed engagement. A user who scrolls for 10 minutes, saves six recipes, and comes back tomorrow is a success. Even if they never actually cooked anything.

This is absolutely insane when you think about what the user actually needs.

You’re standing in your kitchen. You have specific ingredients. You have a specific amount of time. You have a household with specific preferences. You don’t need 2,400 options. You need one answer.

But one answer is bad for the feed. One answer means you leave. And leaving is the opposite of what the architecture was designed to produce. Vanity metrics flag, founders fret, investors flee: panic ensues.

What would the opposite look like?

If infinite scroll is the problem, then the answer isn’t a better feed. It’s no feed at all.

What if an app gave you one option instead of 2,400? What if the success metric wasn’t time-on-app but whether you actually did the thing you opened the app to do? What if the best possible session was: open, get your answer, close, go live your life? What if intelligence actually gave you what you really needed: one clear, calm answer.

That’s a radical idea in consumer software. Almost every app on your phone is designed to keep you inside it. The entire business model of the modern internet — ads, engagement, retention — depends on you staying. An app that wants you to leave is an app that has to find a completely different reason to exist.

But for certain problems — and dinner is one of them — it’s the only design that makes sense. You’re standing in your kitchen at 6:17 PM. You don’t need to be engaged. You need to be cooking.

I spent years in defense technology building situational awareness systems, tools that helped people make sense of messy, fast-changing environments. One of the core lessons: more data is not the same as better understanding. A feed gives you more data. Situational awareness gives you better decisions. Every recipe app is giving you ten thousand data points. The right answer is the right one.

That’s the philosophy behind what we’re building at Springhouse. Not a feed. Not a scroll. One clear answer based on what’s in your kitchen, and then you put the phone down and cook.

Designing for the kitchen, not the screen

Governments and juries are now confirming what Raskin and Harris have been saying for years: infinite scroll is not a neutral design choice. It’s an architecture that prioritizes engagement over well-being. The EU is legislating against it. Juries are awarding hundreds of millions in damages over it. The inventor himself took the stand to testify against what his creation became.

Aza Raskin, after years of regret, built himself a personal tool to break his own scrolling addiction. It added a small, random delay (just 250 milliseconds), every time he scrolled. The longer he scrolled, the longer the delay got. It gave his brain just enough time to catch up with his impulses. Within days, his scrolling habit was broken. He’d pause and think: “Do I actually want to be doing this?”

That’s a tiny, elegant intervention. A fraction of a second that changed behavior.

The question for anyone building consumer software right now is: are you designing for the screen, or for the life on the other side of it?

The feed was designed for the screen. The kitchen needs something else.

— Jay

Springhouse

springhouse.co

I’ve been putting it off this whole week. Which tells you everything you need to know.

At Agile Defense, the work spoke for itself. Our customers were government agencies and the Department of Defense. Nobody asked us to post on LinkedIn about it. Nobody suggested a newsletter. The idea of “building in public” would have been absurd, and in many cases, literally not allowed.

That was the world I came from. Quiet. Heads down. Ship the work. The work is the marketing. Past performance was the “social proof.”

Now we’re building a consumer product, and I’m learning that a lot of the instincts I built over the past decade don’t apply.

“Show your work.”

“Show your work.” That’s the advice I hear from colleagues, investors, advisors, and other founders. Show the process. Share the journey. Let people in before the product is ready. Build an audience that’s rooting for you before there’s anything to buy.

The logic makes sense. I’ve read the playbooks. Founders who share openly build trust, attract early users, and create communities that are invested in the outcome before launch day. Transparency creates loyalty. People root for founders they feel like they know.

I believe all of it. I’m just finding it genuinely hard to do.

It’s not because I’m afraid of failure. I’ve failed plenty. I ran a company for years. I’ve made bad hires, lost deals, worked on strategic duds. Failure doesn’t scare me.

What’s hard is the performance of it.

In defense tech, you build something, you do the “O&M” on it, you get feedback from the people using it at the customer and in the field. The loop is tight and private. You never have to explain yourself to an audience. You never have to turn a setback into content.

Consumer is the opposite. The audience is the product strategy. If nobody knows you exist, it doesn’t matter how good your product is. Attention comes before revenue.

There’s an old line in startups:

First-time founders work on the product. Second-time founders work on distribution.

I knew this going in. Knowing it and doing it are different things.

The part where I tried and couldn’t

Here’s where I’ve landed so far.

People told me I need to be on Instagram and TikTok. Film yourself in the kitchen. Open your fridge on camera. Show the mess. Be relatable. The algorithm rewards vulnerability.

I tried to talk myself into it. I even scripted some videos. But I couldn’t do it. At least not yet. It’s not me. I’m not going to film myself pulling slimy lettuce out of the crisper drawer, hoping to get a viral post. Some founders do that naturally. I am not one of them.

The emerging thinking among builders is that viral growth is actually the wrong strategy for products like Springhouse. You don’t want a million downloads from people who open the app once and never come back. You want a smaller number of people who use it every night, who teach the system their preferences, who build a habit. Viral spikes produce churn.

What we really need is depth. Depth produces personalization. Personalization produces retention.

That doesn’t mean I get to hide. It means I have to find the version of “public” I can actually sustain.

What clicked

There’s an article on buildinpublic.xyz called “Low-Risk Ways to Build in Public (For Introverted Founders Who Hate Oversharing),” and the title alone felt like it was written for me.

The core insight is that building in public doesn’t mean revealing your deepest secrets or forcing an extroverted persona. You don’t have to share revenue numbers or your darkest late-night doubts. You can share what you’re learning about the problem, celebrate the people around you, and document the process without turning your entire life into content.

That framework clicked. Because the truth is, I have a lot I genuinely want to share. I’ve spent months going deep on food waste, kitchen behavior, why every previous attempt at food tracking failed, and what we think is different about our approach. I have opinions about the industry. I have things I’ve learned that surprised me.

The third SA problem

Here’s the thing I didn’t expect.

In my last post, I wrote about how situational awareness — the core of everything we built at Agile Defense — is the same problem Springhouse is solving for the kitchen. See what you have, understand what it means, all to make a better decision.

But there’s a third version of this problem I didn’t anticipate: building the company itself.

As a founder, I’m swimming upstream every day. Product roadmap. Content strategies. Design decisions. When you’re building something new, there are decisions everywhere. You have to orient fast just to keep up.

I don’t have a situational awareness system for that. Nobody does. You just write about it and hope the writing helps you think.

Maybe that’s what building in public actually is. Not a marketing strategy. Not a performance. Just thinking out loud and trusting that the honesty is worth something to the people watching (reading).

What I can do, and what I can’t

So here’s where I’ve landed, a few months into this experiment:

I can write. Substack works for me. LinkedIn works. Long-form thinking is a medium that I can use. I can sit with an idea, turn it over, and put it down clearly. That’s the version of “public” I can sustain.

I can do earned media. I sat down with Mike Wolf at The Spoon and talked about what we’re building. He wrote about it, people read it, and some of them found us because of it. That felt natural — a conversation between people who care about the same problem.

I can share what I’m learning:

• The food waste data that shocked me.

• The graveyard of apps that tried this before.

• The design decisions we’re wrestling with.

Social media and short-form video may come later. For now, writing is the channel.

Maybe we fail too

People ask whether Springhouse can finally solve the “what’s in my fridge” problem. It’s a fair question. A lot of companies have tried and failed.

Maybe we fail too. I think about that more than I’d like to admit.

That might be the most “building in public” thing I’ve said. This isn’t part of the content strategy. It’s merely the truth.

So, I think that’s the version of this I can do. Not the polished founder journey. Not the inspirational reel. Just: here’s what we’re building, here’s what we’re learning, here’s what’s hard, and here’s why we think it matters. Here’s what I’m thinking.

If that’s enough, then it’s enough.

And if it’s not, then I’ll figure that out when I get there.

For now, I’m going to keep writing. I’m going to show my work, honestly.

— Jay

Springhouse

Subscribe now

I’d gone to the supermarket, shopping list in hand, following a recipe. Grabbed the sour cream, came home, and opened the fridge to put it away.

There was already sour cream in there. Unopened. About to expire.

This happens to everyone. It’s happened to you. You sigh, shrug off your annoyance, and throw the old one away next week when it’s past its sell-by date. It’s a nothing moment. A three-dollar mistake in a life full of bigger ones.

But that night, for whatever reason, I couldn’t let it go.

I stood there holding both tubs, one in each hand, and thought: why don’t I know what’s in my own refrigerator? In the times we live in, how is it possible that I have no idea what’s inside?

I went looking for something that could fix this. An app. A system. Anything.

I found a graveyard.

Subscribe now

There have been a lot of attempts. I want to be honest about that, because anyone who builds has to understand the history of those who have come before. This space has had its share of misses and failures.

Plenty of apps, and I mean plenty. They all asked you to type in every item by hand. Chicken breast. Sour cream. Condiments (so many condiments). Do that for forty items a week, every week, forever. Nobody did. The apps died quiet deaths.

Then came the recipe angle. Plant Jammer was the most interesting. Tell it what you have, get recipes back. Clever idea. But it still needed you to maintain the inventory manually, and the suggestions weren’t much better than Googling “chicken recipes.” Plant Jammer shut down.

The appliance companies took their shot. Samsung put cameras inside refrigerator doors. You could look at your fridge from the grocery store. You know what a small fish-eye camera inside a cluttered fridge actually shows you? Chaos. Milk blocking the view. Tupperware stacked three deep. No context, no expiration dates, no clue what’s in the crisper drawer. It was a party trick, not a solution.

Barcode scanners. QR readers. Smart tags. All of them required more effort than the problem they were solving.

They all failed for the same reason: they made the tracking part hard and the payoff part small. You did all the work of logging your food, and your reward was... a list. A list you could have written on the back of a napkin.

Nobody solved the actual problem. Which isn’t knowing what you have. It’s actually knowing what to do with it.

I need to explain what I mean by that, because it’s the thing that changed my thinking entirely.

You come home from work. You’re tired. You open the fridge. There’s chicken thighs, half a bag of spinach that’s starting to look sad, a lime, soy sauce, some leftover rice from two nights ago.

A recipe app would need you to search. Type in “chicken thigh spinach lime soy sauce rice” and hope something comes back. It won’t. Or it’ll give you forty-seven results, none of which match what you actually have, and now you’re scrolling through Pinterest at 6:30 PM while your kids ask when dinner is.

But a good cook would look at those same ingredients and think: stir-fry. High heat. Chicken first, get a sear. Spinach goes in last, since it’s delicate and it’ll wilt in thirty seconds. Squeeze the lime at the end. Warm the rice separately.

That’s not a recipe. That’s judgment. It’s pattern recognition built on an understanding of how food actually behaves. Knowing what’s wet, what’s dry, what cooks fast, what needs time.

That’s what we wanted to build. Not another list. Not another search engine. Something that thinks about food the way a cook thinks about food.

A kitchen on a Tuesday night is a complex, changing environment. The data is messy. The decisions happen under time pressure. Nobody has built the intelligence layer.

That’s what Springhouse is. A kitchen intelligence platform.

We track your inventory. We studied every failed attempt at food tracking and designed specifically around the friction that killed them. The goal is full visibility into your kitchen without turning it into homework.

The tracking is the foundation. This is the part I care about, maybe a bit too much. The part that keeps me up at night, driving me a little obsessive.

Oh, but once we know what you have on hand? That’s where we start to solve the real problem.

Springhouse understands food properties. Not just names. It knows zucchini is high-moisture, quick-cooking, mild. It knows to salt it first if you’re stir-frying, or you’ll end up with soup. It knows zucchini subs for yellow squash. It knows your chicken thighs need a different approach than chicken breast.

With your exact inventory, your preferences, your equipment, and that understanding, it generates one clear set of cooking instructions for tonight. Not suggestions. Not a mood board. Here’s what to make. Here’s exactly how to make it. Let’s go.

And, it learns. Every time you cook, every time you skip something, every time you rate a meal, it gets sharper. It builds a model of your household, your patterns, your taste. Over time, it will know you.

Mike Wolf of The Spoon just published a piece about us. He covers food technology for a living, and the headline he chose was: “Can Springhouse Finally Solve the ‘What’s in My Fridge?’ Problem?”

I like that he said “finally.” It means he knows the history. He knows about the graveyard. He’s watched companies try and fail for a decade.

His question is fair. Maybe we fail too. I think about that more than I’d like to admit.

But here’s what I keep coming back to: 30 to 40 percent of food in American households gets thrown away. Not because people are wasteful or careless. Because they can’t see what to do with what they have. That’s over $2,000 a year per household, going into the garbage.

The problem isn’t information. People know they waste food. They feel guilty about it every time they clean out the fridge. The gap between knowing and doing is enormous.

What actually works is reducing friction. Making the right choice the easy one. Taking the cognitive load off people’s plates, so the default behavior becomes the better behavior.

That’s the bet. Not more recipes. Not more lists. Less thinking. Better decisions. Dinner, solved.

We’re launching on iOS this spring. The waitlist is open. If you’ve ever stood in front of your refrigerator holding two tubs of sour cream, wondering how your life came to this, we have been there. We are building for you.

👉 springhouse.co

Mike’s full piece and our conversation on The Spoon Podcast are here: Can Springhouse Finally Solve the ‘What’s in My Fridge?’ Problem?

— Jay

Claude Code launched as a research preview in February 2025. By the summer, it had become one of Anthropic’s fastest-growing products. By fall, it was generating over $500 million in annualized revenue. This January, Anthropic launched Cowork, bringing the same agentic architecture to non-developers.

The trajectory is remarkable. A command-line tool that lets an AI read your files, write code, run commands, and complete multi-step tasks without constant hand-holding. Developers reported that Claude Code completed in a single pass what would normally take 45 minutes of manual work. Non-developers soon followed, using Claude Cowork for organizing files and generating expense reports.

Then came Clawdbot.

An open-source project built on top of Claude, Clawdbot racked up over 60,000 GitHub stars in weeks. Users were running it locally on their own machines, controlling it through WhatsApp and Telegram, having it book dinner reservations, manage emails, and execute tasks autonomously while they slept. It spread so fast that Anthropic stepped in with a trademark request, and the project rebranded to Moltbot overnight. The community barely blinked. Same architecture, new shell.

This is what an AI inflection point looks like: tools that change how work gets done, spread organically, and spawn entire ecosystems faster than anyone predicted.

I’ve been thinking about what the equivalent moment looks like for food.

Why food hasn’t had its moment yet

Cooking is one of the last daily tasks that AI hasn’t meaningfully touched.

We have recipe apps. Lots and lots of them. We have AI chatbots that can suggest dinner ideas. We have smart appliances with touchscreens and voice assistants. Yet, every night, millions of people still stand in front of a full fridge with no idea what to make.

The tools haven’t worked because they’re solving the wrong problem.

Recipe apps give you more options when what you need is fewer. AI chatbots conjure up dishes that don’t quite work for you. Current smart fridges add complexity and a clunkiness with unnecessary screens and motorized doors, etc. None of them start where you actually start: with the ingredients you already have.

Claude Code worked because it met developers where they were. It operated in the terminal, understood their codebase, and fit their existing workflow. It didn’t ask them to change how they work. It just made the work faster. Clawdbot took it further by meeting people in their messaging apps, letting them delegate tasks through the same interfaces they use to text friends.

Food needs the same approach: intelligence that meets you where you are, in your kitchen, with whatever’s in your fridge.

What the moment requires

For Claude Code and its derivatives, the breakthrough came from three capabilities working together.

First, agentic capability. The AI doesn’t just answer questions. It takes action. It reads files, writes code, runs tests, commits changes. Clawdbot users were having it book reservations, cancel subscriptions, and rebuild websites through Telegram messages. The AI does the work, not just advises on it.

Second, context awareness. Claude Code understands your codebase. Clawdbot remembers your preferences over time. These tools know what files exist, what you’re trying to accomplish, what you’ve done before. They don’t start from scratch every conversation.

Third, trust through transparency. You can see what the AI is doing. It asks before taking significant actions. It loops you in on its plan. You stay in control, even when the system is working autonomously.

For food to have its moment, we need the same pattern, adapted for the kitchen.

What “agentic” looks like for cooking

AI shouldn’t just suggest recipes. It should know what’s in your kitchen, not because you manually entered every item, but because it tracked your groceries, your receipts, and your inventory over time. It should know what’s about to expire. It should understand how ingredients behave: which ones can substitute for others, which ones need to be used first, which combinations actually work.

Then, it takes action. Not physical action (we’re not talking about robots) but decision action. It makes the call.

Tonight, you’re making a stir-fry with the chicken that needs to be used, the bok choy that’s wilting, and the ginger you bought last week. Here’s how, step by step, timed to your schedule.

That’s what “agentic” means for food: the AI carries the cognitive load of deciding, so you just have to cook.

What “context awareness” looks like for cooking

Claude Code knows your codebase. Clawdbot learns your preferences through every interaction. A kitchen AI needs to know you (your dietary preferences and challenges) and your kitchen the same way.

Not just what’s in your fridge right now, but your patterns, do’s and don’t’s. What you cook on weeknights versus weekends. What your family actually eats versus what you aspirationally buy. Which ingredients you always have on hand. How much time you realistically have on a Tuesday.

It’s the power of memory. The same way a good sous chef learns the rhythm of a kitchen, an intelligent system should learn the rhythm of your household.

Context will be what separates generic suggestions from genuinely useful ones.

What “trust through transparency” looks like for cooking

Claude Code asks before it commits code. Clawdbot confirms before it books a reservation. A kitchen AI should show its reasoning the same way.

This dish was selected because your chicken needs to be used, you have 30 minutes, and your household liked the last stir-fry you made. Here’s the framework it’s using. Here are the substitutions it made. Here’s why the timing works.

Trust comes from understanding. If the AI simply says “make this” with no explanation, you won’t trust it. If it shows you how it’s thinking, you’ll start to rely on it.

When will the moment come?

The honest answer: not tomorrow, but sooner than most people think.

Claude Code took six months to go from research preview to billion-dollar product. Clawdbot went from launch to 60,000 GitHub stars in weeks. When the underlying capabilities are ready and the user experience is right, adoption happens fast.

For food, the underlying capabilities are getting there. Computer vision can recognize groceries. Natural language models can understand cooking. The food science (what substitutes for what, how ingredients behave, what timing matters) is knowable and encodable.

What’s missing is the integration. A system that tracks inventory without being annoying. A system that understands food deeply enough to generate real cooking instructions, not hallucinated ones. A system that learns your household without feeling invasive.

That’s what we’re building at Springhouse.

The kitchen’s Claude Code moment

I don’t think the kitchen’s moment will look like a robot cooking for you. That’s decades away, if it happens at all. (more on this in a future post)

I think it will look more like Claude Code and Clawdbot: an intelligence layer that carries the cognitive burden, makes decisions on your behalf, and lets you focus on the part you actually want to do.

For developers, that meant less boilerplate, less context-switching, less time on tedious tasks. The code still got written, but the friction disappeared.

For home cooks, it will mean less decision fatigue, less waste, less nightly stress over what to make, intelligent planning laid out for you. The cooking still happens, but the hardest part gets handled for you.

Claude Code proved that when AI genuinely reduces friction in a daily workflow, adoption takes care of itself. Clawdbot proved that when the capability is real, people will build on it, extend it, and make it their own. The ecosystem emerges organically.

The kitchen is the next daily workflow waiting for that moment.

We’re building toward it.

— Jay, Springhouse