Tonner Jackson

Pumped to share @raredayshq was featured in @Inc today for our growing AI practice and Brand work. Emily McCray did an amazing job articulating the throughline of Rare Days from a pandemic born business to the team of +62 that we are today, projected to double our revenue this year. LFG. A massive part of that momentum comes down to how our team works. I have been blown away at the seriousness by which our crew has adopted AI tools to magnify their creativity. We're doing better work, we're doing it way faster, and we're pioneering brand new ways of working. One of my favourite examples of this: just this week, a leading AI lab incredulously asked us how we used a tool to spin up a solution so quickly. A tool they had a part in creating. When the people building the frontier models are surprised by how fast you are deploying them in production, you know your team is onto something special. Proud of the entire Rare Days crew for setting the pace power-partner.inc.com/emilym…

please consider following Madre on social, and supporting if you are in Medellin!

If you have a homie lamenting suburban life, let em know it wasn’t so long ago that the suburb was thought a miracle

The ultimate middle finger to modernity Is to learn a second language in your adult life

Favorite Jobs clip: Everything works backwards from customers. The best of us arrive at this intuitively and make amazing businesses. Some less fortunate go to business school and are distracted from that core truth, and lose the path.

My takeaway from 20 major product launches this month: A lot of companies deploying AI (outside the frontier labs) are making the same mistake factories made with electricity a century ago. Before electricity, a factory ran on one steam engine. It spun a long rod that ran across the ceiling, and every machine drew its power by connecting to that rod with a belt. Each machine had to sit close enough to the belt, making the layout dictated by where the power could reach, not by how the work flowed. When electricity arrived, there were two responses. The first was to pull out the steam engine and drop in one large electric motor to spin the same rod. The factory looked identical, just running on electricity instead of steam. These plants only saw marginal gains. The second was to put a small motor inside each machine. Once every machine had its own power, you could rip out the rod and belts entirely and rearrange the floor around the work itself. This approach is what drove the productivity surge of the 1920s. We're at the swap-the-engine stage with AI. The typical AI deployment today is that first move: - A chatbot where the search bar used to be - An "ask AI" button in the corner - A summary box on the same dashboard. The product underneath is the same, just with a smarter-looking layer on top. That's the big motor spinning the same old rod. The companies who build effectively with AI will start from what the technology now makes possible and ask the first-principles question: How should this product actually work? If an AI agent can do the entire task, you stop designing screens for a person to click through, and the interface moves to the only place human interaction is needed.

A design principle that's become exponentially more important in the age of AI: Hick's Law. It's the idea that the more choices you give someone, the longer it takes them to make a decision. AI makes has made it too easy to ADD more - more options, more features, more ways to interact. Every one of those choices comes with a cognitive cost. Be ruthless about outcome you want for your customers (and what is the path of least resistance to get there).

a GOAT view of NYC Immaculate energy in central park this morning

We built a dashboard to see if our 60-person team ships more with AI. Feel free to copy it. It pulls every project out of Linear and GitHub and tracks what we produce each sprint: - Velocity (how much work we complete per cycle) - Completion rate (how much of the plan we hit) - Escape defects (bugs that slip through to production) - Commit activity per developer, per repo, and on the same screen; it tracks how much we use Claude. The value is in reading those two together. A usage number on its own tells you people are opening the tool, but if you put it next to shipped output, you get a much better idea of its effectiveness. If Claude usage climbs and we ship more at the same defect rate, the AI is producing as it should. If usage climbs and output stays flat, we're paying for activity, and we know exactly where to look. For any team spending serious money on AI, that turns 'we use it a lot' into a measured return.

Look at these 2 designs. The first is how the internet made money in the late 90s. The second is what the internet created. When the internet was first monetized, display ads were the entire model: print ads lifted from newspapers and placed beside text online. The format native to the internet turned out to be the news feed, dynamic content personalized to you. That’s because it wasn’t confined to print. Design has a name for that first stage: skeuomorphism. New platforms imitate old formats because people need familiar metaphors to cross over, and the imitation never turns out to be the end state. AI design is at that stage right now. Agents can already displace interfaces that used to be solved with UI. Yet a lot of what gets built today still borrows its shape from the pre-agent era: forms, dashboards, tickets written field by field. There will be native design to AI, but we don't know what it is yet. We only know that the old paradigms lived with constraints that no longer exist, and the designer's attention shifts to a new kind of UI that could be far more functional than anything before it.

AI is making proof of craft more important than ever. For context, proof of craft is the visible evidence that real skill went into the work: the thinking, the rigor, the process behind the result. Few understood that better than Dieter Rams, the designer behind Braun who shaped how Apple designs today. Rams built his reputation by showing the prototypes and models behind his products, the proof of the work that went in. When anyone can generate a clean-looking result in seconds, the finished piece stops proving much on its own. What separates real craft is the thinking behind it, and the way to show that is to show your process. Irving Penn understood this too. He left the thin black border of the film negative around his prints, proof the shot was composed in the camera and not cropped later in the darkroom. The process is the skill signal.

INSTEAD OF WATCHING THE WORLD CUP THE LAST 4 DAYS. One of our members, Caio, turned our entire engineering team's process into AI tooling. It's a set of skills inside Claude Code, each one carrying a piece of how we operate. With a single command, he can: - Turn a rough idea into a clear build plan, through a structured back-and-forth - Save a batch of changes as a clean, self-contained checkpoint - Work through a reviewer's comments on his code, and write the small automated checks that confirm it works to our standards - Drive a browser with Playwright to test the interface, check the layout, and read the error log - Create and update tickets in Linear, our project tracker, without leaving the editor - Rebuild a full week's timesheet from his calendar, Slack, Granola, and Gmail On top of that, he built Rare Flow, a system that runs an entire project from start to finish. It breaks the work into well-defined phases. Separate AI agents then research, plan, and build in parallel, each in its own clean workspace, so the main conversation stays uncluttered. It also holds state: it remembers where everything stands across days, saves progress after each task, and checks the result against the original goal before marking a phase done. He taught it how our team works, so the output comes back at our standard rather than generic. None of it is novel. It's our own workflow, made repeatable, and that's the whole point of having agents. S/O Caio

Rare Days has fully reconstructed how we build zero-to-one products… The usual way to staff a project is to put a large team on it: several designers, several engineers, and the coordination layer that holds them together. We're testing a smaller unit. One designer, one engineer, one project manager, and Claude. The bet is that Claude takes on enough of the execution to let a tight three-person team deliver what used to take many more: the first drafts, the repetitive build work, the connective code. Less coordination, less overhead, and more of everyone's time spent on the decisions that need human judgement. We're validating it on real projects before we commit to it as how we operate. If it holds, it changes the economics of running a studio.

One of our engineers had code merged into Cloudflare's codebase. We're longtime Cloudflare fans and have been building on it across a bunch of use cases, so shipping into their codebase was a fun moment, lol The best part is, it happened in between karaoke sessions in Victoria. Congrats, Holden.

I LOVE design psychology. We built an AI reading platform that’s served 17,000+ kids by personalising learning for each one. It’s called Royo, and it’s built on 3 principles: 1. The Ikea effect To get a child invested in reading a book, they needed an element of familiarity or personalisation. A platform where they choose the character, the world, and what happens helps them do that. It’s similar to the Ikea model - people place more value on what they help create. A child is more likely to work through a story they made instead of one that’s been simply handed to them. 2. Zone of Proximal development & the science of reading People grow fastest when the material sits just beyond what they can do on their own. This is the zone of proximal development. The second aspect is built on the research consensus that children learn to read by systematically decoding words from their sounds, which is why every Royo story is built on phonics. Together, they let one platform stretch a confident reader and help carry a struggling one. 3. Theory of self-determination built into an AI harness The principle that motivation holds strongest when a person can feel their own competence growing. Seeing their progress add up gives children proof they're improving. Yet none of these principles matter unless a teacher or parent can trust what the AI puts in front of a child. A generative model can invent things, and that has no place in front of a six-year-old. So we built a harness around it - a layer of rules every story has to clear before it ever reaches a child: - Follow the Science of Reading methodology - Include key vocabulary - Maintain a clear narrative - Ensure grammatical accuracy - Prioritize safety for young readers Inside those rules, the model is free to be as imaginative as it likes. The adults stay in control of the rest. Teachers get a dashboard tracking each child's reading progress. Parents get a clear view of how their child is doing and specific ways to support the reading at home. Design drives the aspect of creation, pedagogy shapes the learning, gamification pushes progress. Every layer is built to keep a child engaged.

An underrated way to use AI: Build to remove friction. Most teams reach for AI to add output, more content, more agents. The higher-return move is often the opposite: find the slow, manual work already dragging on your team and delete it. A client needed to migrate a large volume of content into their CMS, the system that stores and manages their site's content, ahead of a launch. Doing that manually is incredibly slow. One of our engineers built a tool that generates spreadsheet templates matched to their content structure, checks every required field, and uploads it all automatically. Non-technical people can run it themselves, and it produces a standalone script they can reuse later without us. A year ago, a tool this small wasn't worth a developer's time. Now it's worth an afternoon, and removing that kind of friction is the most underrated thing AI lets you do.

I run a 60-person design and engineering studio doing 8 figures. These are the 10 best apps in the world, according to them: 1. @figma Where the whole team designs. In an era where people vibe-code entire products from a prompt, Figma stays untouchable. 2. @WisprFlow A voice-to-text app with visual context. James, our VP of Design, sits 4th in the world on its leaderboard and pairs it with Claude Code to prompt by voice. 3. @coderabbitai An AI reviewer that reads every code change before it ships. Cole, one of our developers, keeps it for the bugs and accessibility issues it catches that Claude misses. It even flagged Linear's AI agent as worth using for project management. 4. @Obsidian A markdown editor, built for plain-text format where most of our docs live. Miles, our VP of Partnerships, points it straight at his Cowork folder, so any edit he makes is reflected in Cowork instantly. 5. @raycast A command bar that collapses a dozen Mac utilities into one. Resize windows with a keystroke, search your full clipboard history, run a custom prompt on highlighted text without leaving the app you're in. 6. @getAkiflow A time-blocking tool for running several calendars as one schedule. Its AI scheduling is already strong and still improving. 7. @Maccy A clipboard manager that keeps a searchable history of everything you copy, where the Mac's built-in one falls short. 8. @BetterDisplay Precise control over how external monitors scale and render. For anyone who has fought a second screen's resolution, it's a lifechanger. 9. @BePresent A screen-time tracker and app blocker that uses game mechanics to make the habit changes hold. 10. @jittervideo Where the team builds motion, the animation and transitions for our social media. It connects directly to Figma.