<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Michael Margolis on Medium]]></title>
        <description><![CDATA[Stories by Michael Margolis on Medium]]></description>
        <link>https://medium.com/@yipe?source=rss-b211895d1365------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*gyodPPY5c0urBZoQqOwG-w.png</url>
            <title>Stories by Michael Margolis on Medium</title>
            <link>https://medium.com/@yipe?source=rss-b211895d1365------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 01:44:10 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@yipe/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com" rel="hub"/>
        <item>
            <title><![CDATA[The New Long Pole]]></title>
            <link>https://yipe.medium.com/the-new-long-pole-cdfb4348602b?source=rss-b211895d1365------2</link>
            <guid isPermaLink="false">https://medium.com/p/cdfb4348602b</guid>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[leadership]]></category>
            <dc:creator><![CDATA[Michael Margolis]]></dc:creator>
            <pubDate>Mon, 01 Jun 2026 02:57:57 GMT</pubDate>
            <atom:updated>2026-06-01T02:58:35.927Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>TL;DR:</strong> AI shrank the oldest bottleneck in software and exposed every long pole that was hiding behind it. Here’s a model to spot your new ones and shorten them in a way that builds psychological safety instead of blame.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*0HgF1NZwnPqWZao9" /><figcaption>Photo by <a href="https://unsplash.com/@75i?utm_source=medium&amp;utm_medium=referral">Ingo Ellerbusch</a> on <a href="https://unsplash.com?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>The “long pole” (or “<a href="https://writingexplained.org/idiom-dictionary/long-pole-in-the-tent">the long pole in the tent</a>”) is a term I first heard maybe 20 years ago. It described the <em>single thing</em> that takes the longest to complete. Like a tent pole, it’s what defines how large the tent is and is the load bearing element quite literally holding everything else up. It could be a task, a person, a decision, a department, a ritual, a gate, a lack of alignment, or even just the time it takes to get permission to do something. Every project has a long pole, no matter how efficient you are.</p><p>Many projects can feel like a circus at times, so that just makes me love that phrase even more. 🤡</p><p>Since the dawn of software engineering, the long pole has almost always been the same thing: writing the code. And for as long as I can recall, our industry has been on a relentless mission to shorten that specific pole.</p><p>It went something like this (over-simplified, of course):</p><ul><li>Punch cards are slow and a pain. So we started typing assembly on these keyboard things.</li><li>Assembly is slow and hard to write. So we created C, and let the compiler write the assembly for us.</li><li>Memory management proved hard and fragile. So we built Java to handle memory for us.</li><li>Getting updates out by mailing floppy disks and CDs was too slow, so we created software updates and moved more and more towards web development instead of native.</li><li>Standing up a server can take weeks. So we built stuff like AWS and serverless.</li><li>Software development got too fast for traditional manual QA to keep up, so we moved to more automated test suites.</li><li>Coordinating people is messy. So we built agile, scrum, kanban.</li><li>Oops, now shipping multiple times a day too risky. So we built CI/CD, feature flags, and observability.</li></ul><p>At the end of the day, what used to take a hundred engineers to stand up can now be done in an hour by a teenager in an internet café. Amazing.</p><p>And yet, even after every one of those sea changes, writing the code was somehow <em>still</em> the long pole at software companies?!</p><h3>A Crossroads</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*yKWv-2muUG-__LA9" /><figcaption>Photo by <a href="https://unsplash.com/@soymeraki?utm_source=medium&amp;utm_medium=referral">Javier Allegue Barros</a> on <a href="https://unsplash.com?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Until recently.</p><p>Out of seemingly nowhere, AI just shrank that pole dramatically. Writing code is no longer always the bottleneck. The circus tent is collapsing in on itself, the sky is falling, and we’re all scrambling to figure out what to do. We’re running into bottlenecks we’ve never really had to think about, because they were usually hiding behind that code-writing pole.</p><p>So… we find ourselves at a crossroads in this industry. At Medium, we just shipped a new feature that took six weeks to build, end to end. A year ago that would have taken six months. When we talked to the team, many concluded it could have taken just two.</p><p>If that feature could be done in two weeks, the obvious next question is: what would it take to ship <em>any</em> new feature in two weeks? What was the long pole?</p><h3>Guardrails, Not Roadblocks</h3><p>A quick aside, as this model is relevant and has served me well. To get better at understanding blockers, we need to be mindful about the things that slow us down. A lot of what we call “process” can be thought of as three different things mashed together:</p><ul><li><strong>Roadblocks</strong> stop forward motion. Waiting on a decision, a review, data, design, or approval.</li><li><strong>Speedbumps</strong> slow us down on purpose so we don’t break things. These can be good!</li><li><strong>Guardrails</strong> keep us from going off the road while we move at speed.</li></ul><p>We tend to treat all three the same: as something that slows us down. It’s also tempting to just cut process or skip steps because it’s faster, but often those steps were created for a good reason and still have value, and this model helps us differentiate the good from the bad.</p><p>Our job is to figure out which is which on every project, and either eliminate the roadblocks or convert as many into speedbumps or guardrails as we can.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*wqg1a3eUw2vFGTPG" /><figcaption>Photo by <a href="https://unsplash.com/@kaip?utm_source=medium&amp;utm_medium=referral">Kai Pilger</a> on <a href="https://unsplash.com?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>The most effective teams I’ve ever managed (before the AI era) were ones that had truly internalized the idea of “Guardrails, not roadblocks”. It’s important to note that it’s not just the <em>teams</em> that need to interalize it, the company, the leadership, the incentives, and the decision making structures have to get aligned as well.</p><p>Teams need enough context to make these decisions.</p><p>Teams need to feel empowered to make these decisions.</p><p>Teams need to feel safe to take risks.</p><p>And to support all of this, leadership needs to make it possible and safe to do so.</p><p>Everything looks different than before in this strange AI era we’re in, so what does that look like?</p><h3>A New Operating Posture For The AI Era</h3><p>The goal here is ultimately: long pole reduction. If we all align on that being a shared goal, this gets much easier. It gives us <em>shared language</em> and a <em>low-blame way</em> to introspect on every project. It creates the psychological safety to flag questions, without it sounding like finger pointing or frustration. It makes it easier for anyone at any time to ask:</p><ul><li>What’s the long pole <em>right now</em>? How could we mitigate it?</li><li>Does that step need to be blocking, or do we even need to be doing that step <em>at all</em>?</li><li>Now that we’ve shipped, what would have had to be true for this project to take two weeks?</li><li>Where were we <em>genuinely</em> blocked, and where did we just treat a guardrail like a roadblock?</li></ul><p>As a small fully remote company, Medium has been avoiding big company process and has been very intentional at putting humans first. This means a lot of great face time every week at office hours where we can all see each other’s work, show demos, get updates, and give feedback.</p><p>But in the era of AI, this quickly became a bottleneck, and with increased velocity, there were even more updates. This ritual could not keep up, and the company was moving at the speed of office hours. To address this, we shifted the updates to be async in Linear, which then freed up more of that face time to have more meaningful discussions. We also created per-team leadership channels and are working to normalize and get better at async discussions and decision making.</p><p>This was an obvious long pole, we’re now hitting more, and they seem to be sprouting like weeds. Code review in particular is becoming a bottleneck for us. What are we to do?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*7ZmWqF81AiQSfWBL" /><figcaption>Photo by <a href="https://unsplash.com/@luigy_ghost?utm_source=medium&amp;utm_medium=referral">Luigy Ghost</a> on <a href="https://unsplash.com?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3><strong>Code Review is the New Long Pole</strong></h3><p>LLMs crank out code faster than we can review it. The bottleneck didn’t disappear when AI got good at writing code… it moved to review, and most teams haven’t cracked this yet.</p><p>The data’s pretty stark. Output per engineer jumped significantly from 2025 to 2026 yet reviewer output hasn’t magically scaled at the same speed. <a href="https://linearb.io/resources/engineering-benchmarks">LinearB’s 2026 benchmarks</a> found AI-generated PRs wait 4.6x longer before a reviewer even picks them up. A <a href="https://linearb.io/resources/build-vs-buy">related LinearB writeup</a> shows agentic-AI PRs as sitting idle 5.3x longer at the 75th percentile.</p><p>Sure, they get reviewed faster once someone starts, but that queue wait dominates the whole cycle. Nobody I know has cut humans out of the review loop, and I don’t think we should try quite yet. The goal here is to reduce the long pole. Code review is an important speed bump, it’s just extra painful right now.</p><p>What have we found that actually helps?</p><p><strong>Keep PRs small and focused.</strong> This is your highest-leverage move, and it’s free. AI raises how fast you write code, but does nothing for reviewer bandwidth. One PR, one intent. I analyzed some recent PRs here at Medium and found a very nice (and expected) correlation between PR size and cycle time, with review time dominating.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IFlYNDqdkhZQxGrE2Z0WjA.png" /><figcaption>Source: my own vibe coded tool</figcaption></figure><p><strong>Catch more before you cut the PR.</strong> Make the LLM review itself first. Skills like /simplify, <a href="https://claude.com/plugins/code-simplifier">Code Simplifier</a>, and my new favorite <a href="https://code.claude.com/docs/en/ultrareview">/ultrareview</a> (aka /code-review ultra) spin up reviewer agents to find bugs before a human looks. One heads-up: /ultrareview can cost $5–20 per run, so use it wisely.</p><p><strong>Commit the plan, not just the code.</strong> This is the formal version of a habit worth building: spec-driven development. When you get a human to approve the spec and acceptance criteria before generation, the reviewer compares code against an approved plan instead of reverse-engineering intent from a diff. Whether it comes from Cursor’s Plan mode or something like <a href="https://github.com/obra/superpowers">superpowers</a>, commit it. Bonus move: commit the prompt too, so the reviewer sees the goals and constraints you handed the model(s).</p><p><strong>Experiment with review tools.</strong> There are lots of new tools out there like Greptile and CodeRabbit for automated first-pass coverage, and you can write your own skills tailored to your codebase. They won’t <em>replace</em> the human, but they can help clear the mechanical stuff. It’s like linting but next level.</p><p>At Medium, none of this removes the reviewer’s judgment or accountability, it just spends their attention on what needs a human: logic, architecture, decisions, and “should we even build this” rather than just slogging through volume.</p><p>Over time, this opens up larger ambitions and gets us into a posture of supporting each other and working together to reduce dependencies, instead of feeling blocked by each other.</p><h3><strong>The Next Long Poles?</strong></h3><p>We all know that code review isn’t the only thing slowing us down. And even if you clear the code review bottleneck, you’ll find another pole holding up the tent. The long pole shifts by company, org, team, even project, and it shows up in a lot of forms. The good news: most of them can be shortened.</p><p>A field guide to the usual suspects and some ways to think about them:</p><p><strong>Waiting on a decision?</strong></p><p>Align with leadership on which decisions <em>actually</em> block you, and <em>tell them</em> when you’re blocked. We often don’t realize a question we’ve asked is holding up a team. Outside of Type 1 (irreversible) calls, most leadership decisions are often clarifying speed bumps or guardrails, not roadblocks.</p><p><strong>Too many cross-team dependencies?</strong></p><p>Look at your ownership model. Could your team do the work and have the owning team review it, instead of lobbing tickets back and forth and waiting? That’s been a massive unlock at past jobs, and AI makes it easier than ever.</p><p><strong>Waiting on another team’s service or API?</strong></p><p>If you can’t realistically go build a service or API for another team… don’t just wait on their code. Agree on the interface, mock it, and build against the contract. Integrate for real when they’re ready. Contract-first turns one serial dependency into two parallel tracks and keeps you unblocked.</p><p><strong>Stuck behind a review gate? Security, legal, infra, release?</strong></p><p>Pave the road. Pre-approve the common patterns so the default path needs no sign-off, and save human review for the genuinely risky stuff. Aim for “compliant by default,” not “ask permission every time.”</p><p><strong>Waiting on design?</strong></p><p>What if basic screens didn’t need much designer time? A design system, shared components, design-blessed DESIGN.md files, and skills could let teams stand up standard screens on their own, so designers spend less time designing mundane repetitive screens and more time on the meaningful work that needs their deep domain expertise. Many small designs could just be a Loom instead of a full Figma file, saving even more time.</p><p><strong>Decision isn’t getting made because the meeting keeps slipping?</strong></p><p>Ownership is how you drive impact, and you can’t drive anything while you wait. Lagging decisions kill motivation and erode culture over time, and they teach a team that stalling is the safe move. So move the decision async. Write the decision down, propose the call you’d make, set a “decide by” date, and make silence the default: “shipping X Friday unless someone objects.” Pair it with disagree-and-commit so one holdout can’t stall the room. What’s the worst that can happen? Leadership pulls it back as a Type 1 decision and you wait for the meeting, but that should be the exception, not the rule.</p><p><strong>Blocked on capacity? Just not enough people?</strong></p><p>Sometimes the fastest unblock is… cutting scope. Propose what you’d drop and when you can pick it up, gated on either a milestone, after a set time, or once people free up. Half a long pole is usually work that can wait, not work nobody would miss.</p><p><strong>One person is the bottleneck for everything?</strong></p><p>That’s a <a href="https://en.wikipedia.org/wiki/Bus_factor">lottery-factor</a> problem dressed up as a workload problem. Pair on it, write the runbook, spread the context, so the team isn’t waiting on a single human.</p><p><strong>Do you find your team blocking <em>other</em> teams?</strong></p><p>See if you can stand up a self-service version of what you do. The stuff clogging your queue is usually the same handful of requests over and over. Turn those into something people can do themselves and you delete the dependency on both sides. This can look like IT automation, runbooks, admin tools, or even just permission for someone else to do the work if they are able! Every request you make self-serve is one less thing in your queue, and more importantly, one less team waiting on you.</p><p><strong><em>Still</em> waiting?</strong></p><p>Just ship the thing behind a feature switch. You can bug-bash it, play with it, and get real feedback on the actual thing while the bigger decisions cook. Don’t just sit and wait, keep moving forward. Code is easier than ever to iterate on or even delete if needed.</p><p><strong>Genuinely stuck?</strong></p><p>It happens to everyone eventually. When it does, clarity on goals is what keeps you moving. There’s almost always valuable work that doesn’t depend on whatever’s blocking you. I’m sure you have no shortage of quality and reliability fixes that help users directly, and a steady backlog of tech debt and high-impact cleanup most teams are sitting on: CI/CD speedups, framework upgrades, flaky-test fixes, and the like. The teams that keep momentum are the ones with clear north stars and well-groomed backlogs, so when one project stalls you can hop to an unblocked one without losing a beat.</p><p><strong>The pattern?</strong></p><p>The pattern across all of these: a block is usually a missing decision or missing context. Leadership’s job is to push both down. So after a leadership unblocks you, expect leadership to ask: <em>what would you have needed to do this without asking us?</em> And if you are blocking another team, ask them the same question!</p><p>That answer shortens the <em>next</em> pole before it ever goes up.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OpTFossuD8jcdwODHG78ZA.png" /><figcaption>Source: ChatGPT</figcaption></figure><h3><strong>The One Weird Trick</strong></h3><p>If you forget everything in this post, remember this.</p><blockquote>Make finding the long pole a team sport: surface it together, name it without blame, then ask what would need to be true to shorten it for next time. Then do it again.</blockquote><p>That’s the whole trick. Naming it turns a vague “ugh, this is taking forever” into a specific thing we can actually shorten. And “what would need to be true?” turns a one-time unblock into a permanent one. You don’t just get past the pole, you make sure it’s gone before the next project puts up its tent.</p><p>Do that every time and the poles stop sprouting like weeds. The tent goes up faster. Shipping features in two weeks stops being a happy surprise and starts being the floor.</p><p>So… Find the pole. Name it. Ask what would made it disappear. Go make that true.</p><p>What long poles are you facing right now, and how are you addressing them? Let me know in the comments.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cdfb4348602b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Cursor, Superpowers, and Accidental iOS Tokenmaxxing™]]></title>
            <link>https://yipe.medium.com/cursor-superpowers-and-accidental-ios-tokenmaxxing-3fb206cb65a7?source=rss-b211895d1365------2</link>
            <guid isPermaLink="false">https://medium.com/p/3fb206cb65a7</guid>
            <category><![CDATA[ios]]></category>
            <category><![CDATA[anthropic-claude]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[programming]]></category>
            <dc:creator><![CDATA[Michael Margolis]]></dc:creator>
            <pubDate>Mon, 27 Apr 2026 22:06:09 GMT</pubDate>
            <atom:updated>2026-04-27T22:06:09.808Z</atom:updated>
            <content:encoded><![CDATA[<p>So… I accidentally spent 912 million tokens finding out that AI agents have gotten much better at migrating Objective-C to Swift.</p><p>This is not a story about how AI is bad. This is not even a story about how spending tokens is bad.</p><p>This is a story about going to bed after telling an agent, “please proceed carefully,” and waking up to discover that it confidently spent the entire evening grinding through 50% of massive migration without coming up for air.</p><p>For context, I’d say I’m a maximalist about AI’s leverage but a realist about its limits. These tools are already changing what teams can take on, how quickly we can learn and reason about unfamiliar systems, and what kinds of tedious, expensive, and delicate work is suddenly tractable. However, “AI <em>can</em> do this” is still not the same thing as “we should yolo this into production.” Maybe one day, but today is not that day.</p><p>I’ve been an Objective-C developer for much of my professional career. A year ago, I tried using various models to migrate Objective-C to Swift, and it did not go well. The same models that could generate deeply idiomatic TypeScript would produce bizarrely Pythonic Swift that often would not compile.</p><p>Fast forward a year.</p><p>Models, tooling, and agents have gotten better, so I wanted to see what would happen now. I wanted to understand how far today’s models could get with a constrained, mechanical migration in a real legacy codebase.</p><ul><li>Could they follow a careful plan?</li><li>Could they avoid “while I’m here” refactors?</li><li>Could they preserve behavior at scale?</li><li>Could they get through enough code to be useful?</li><li>Could I have agents coding away and automating the drudgery while I slept?</li></ul><p>The answer was: yes.</p><p>The other answer was: oh no.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vpB36JFZ6kfHfnzuwd4dMA.png" /><figcaption>912M tokens later…</figcaption></figure><h3>The plan</h3><p>I started with /brainstorming from <a href="https://github.com/obra/superpowers">Superpowers</a> and iterated until I had a plan I liked.</p><p>The core rule was simple:</p><ul><li>Migration commits must be semantically identical to the original code.</li><li>No behavior changes. No modernization. No “while I’m here” optimistic fixes.</li><li>No Swift idioms, don’t try to make things Swifty. Any idiomatic Swift cleanup would happen <em>later</em>, in separate commits, after the mechanical migration was done.</li></ul><p>The plan had four phases.</p><p><strong>Phase 0: Mechanical prep</strong></p><p>Before touching code, build an inventory of every .m and .h file, add a CI check to prevent new Objective-C from being added. Gotta make sure you aren’t pouring water into the boat <em>while</em> you’re bailing it out.</p><p>Then, audit all owned headers for nullability. This part mattered most to me. Every missed nullable in Objective-C can become a Swift !, and every Swift ! is a like leaving a small haunted doll in the codebase’s basement, waiting innocently and patiently in the one code path nobody tested.</p><p>Yeah, no thanks. Fool me once !…</p><p><strong>Phase 1: Leaves first</strong></p><p>A leaf is a file that nothing else in first-party code imports. Those are the safest files to migrate because they have no owned dependents.</p><p>The plan was to start with the smallest leaf files and migrate one file per commit.</p><p>I asked it to make sure tests passed after each migrated file. In hindsight, I forgot to ask it to write new test coverage as it went. In my defense, this was an experiment to see how well it worked, not something I planned to one-shot and merge.</p><p>As leaves move to Swift, more files become leaves. That creates a cascade. The migration can accelerate over time.</p><p>Nice and controlled. Easy to code review.</p><p>Normally, this is the kind of thing that runs for a few minutes and then stops to ask a question.</p><p>Normally.</p><p><strong>Phase 2: Interiors, bridge-first</strong></p><p>For files other code depends on, the plan was more cautious:</p><ol><li>Extract a Swift protocol that mirrors the existing public surface.</li><li>Migrate callers to use the protocol.</li><li>Rewrite the implementation in Swift behind that protocol.</li></ol><p>The old Objective-C class <em>only</em> disappears in the final step, after being compared to the new one, and confirming the new one works.</p><p>Anything complicated or load-bearing, like singletons, KVO, @synchronized, or NSException, would need a closer look and careful plan before their implementation changed.</p><p><strong>Phase 3: Permanent Objective-C boundaries</strong></p><p>Some things would stay Objective-C for now. Generated models. Mantle. Vendored third-party code. The goal was to keep those behind Swift wrappers, then evaluate later whether to move codegen to Swift or find pure Swift packages for the rest.</p><p>The whole strategy was deliberately light on tooling. No automated .m to .swift translator for this. Those can work well, but this was an <em>experiment</em> in doing it with an LLM, after all.</p><p>So… I told the agent to proceed carefully. I told it to stop if it had concerns or questions. Then I hit go and went to bed.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cG-Z5ytRxn22bqpu5vKs1Q.png" /></figure><h3>Whoops</h3><p>I came back the next morning, manually fixed a few issues that caused a launch crash, answered a few questions from Sonnet, and kicked it off again to crank in the background before making breakfast for my daughter and taking her to school.</p><p>I assumed it would run for 10 or 20 minutes, hit a wall, and stop. That is <em>usually</em> what happens. So I minimized the window, and went on with my day.</p><p>As I wrapped up my wall of meetings, I was delighted to see what it had done. More than 50% of the migration was done. All tests were passing. And, to my surprise, the app not only built in the simulator, but I could log in and click around and everything still seemed to work.</p><p>The bad news? 98 commits. 814 files changed. +29,461 / -16,350.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oYFOjDS9Zv6f_6qA1KA6VQ.png" /></figure><p>I had never gotten an agentic coding task to run for more than an hour without supervision before, and to do that I had to --dangerously-skip-permissions and run it in a container. This was just a regular Cursor agent that somehow just… kept going.</p><p>So, I headed over to <a href="https://cursor.com/dashboard/usage">https://cursor.com/dashboard/usage</a>… and…</p><p>Oh.</p><p>Oh no.</p><p>I had burned through <strong>912 million tokens.</strong></p><p>Thankfully, the vast majority of that was cached input, not uncached input or output tokens.</p><p>But still. Nine. Hundred. Twelve. Million. Tokens.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NEnlWRzjnQJ8VLaODk9yZw.png" /><figcaption>Pages and pages of this. Source: <a href="https://cursor.com/dashboard/usage">https://cursor.com/dashboard/usage</a></figcaption></figure><p>I quit Cursor and went for a little walk.</p><p>The idea that something like this can run for eight hours and be largely correct is… wild. A year ago, I would have said this was <em>years</em> away. Yet here we are.</p><p>Even crazier — the agent had not gone rogue. It did not break production. It did not invent a new architecture. The model followed the plan <em>way </em>better than I expected. It made small commits. It systematically applied mechanical translation. It understood the migration shape. It made meaningful progress through a large, old, real codebase. It just diligently did what I asked, and it did not need to stop because it did not hit any walls it could not reason through.</p><p>Largely thanks to Superpowers (and only a little thanks to my own experience doing migrations) the model had a plan with enough structure to make progress, enough permission to continue, and <em>maybe</em> not enough hard stops.</p><p>It also did not have the human instinct of, “This is becoming a lot. Maybe I should check in before creating the 13th… or 53rd commit.”</p><p>It just kept going.</p><p>As a research spike, this was useful.</p><p>It gave me more confidence that frontier models have crossed a threshold on this kind of work on iOS, and it produced a migration draft large enough to inspect, measure, and learn from.</p><p>That feels… new. And terrifying. Mostly terrifying.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JOaOybTleyCRXFFPrmdN0Q.png" /><figcaption>Now 50% more Swifty!</figcaption></figure><p>So, we’re gonna merge this, right!? Bust out the Mission Accomplished banner!</p><p>No, no we’re not. <em>Just because we can does not mean we should.</em></p><p>I would never take a +29k -16k diff across 814 changed files from an unsupervised agent and drop it into a production mobile app. I also would not ask engineers on my team to waste their time reviewing a diff that large. That way lies ruin, App Store sadness, and mutiny.</p><p>IMO, we are nowhere near the point in the industry where automated code review and test infrastructure can blindly accept a patch of this size. Maybe we will get there. Maybe sooner than I expect. But this is still a research spike, not something we should try to merge.</p><h3>The actual lesson?</h3><p>I fed Cursor and Sonnet a bunch of judgment-shaped instructions, and they followed them better than I expected.</p><p>And yes, while it was more expensive than I expected, it was still worth it as a way to validate the approach. Engineering time is expensive. Slow migrations are expensive. Diving into a huge migration without validations can be even more expensive. Spending $10, $100, or even $1,000 in tokens to investigate something can save weeks of effort, if not more.</p><p>But the lesson is not “tokens are cheap, go wild.” Won’t someone think of the planet!?</p><p>The lesson to me is that agents are becoming capable enough to need clearer boundaries. If an agent stops early and often, the blast radius is small. You address the concern and move on.</p><p>If an agent works, it may keep working. And working. And working.</p><p>Until you have 98 commits, 814 files changed, and a PR so large it would take weeks to review well enough to catch the inevitable numerous subtle, catastrophic regressions.</p><p>For codebase-scale agent work, I would now give much stricter stopping rules.</p><p>A good prompt might say:</p><blockquote>Make one commit per migrated file. Preserve behavior. Stop when you are uncertain.</blockquote><p>A better prompt might instead say:</p><blockquote>… Migrate the 10 safest files and stop for review.</blockquote><p>And yes, modern codebases can <em>also </em>be built to make this far safer: functional patterns, strong types, heavy test coverage, clear module boundaries, consistent architecture, fast feedback loops, reliable CI, and fewer spooky side effects. But I’d wager 99.9% of production codebases in the world don’t meet those criteria quite yet.</p><p>I think the takeaway for me is to make sure “continue” does not mean “until the eventual heat death of the universe.” Include explicit limits: stop after N commits or N files. Because “please proceed carefully and stop when you’re stuck” can still mean: carefully do the next thing, then carefully do the next thing, then carefully do the next thing, forever. That is how you wake up to nearly 1 billion tokens used.</p><p>So <em>yes</em>, before you kick off a massive migration before bed, maybe make sure “please proceed carefully” is followed by the one instruction I forgot:</p><blockquote>“Please stop”</blockquote><h3>Come work with us</h3><p>I joked about tokenmaxxing in the title, but Medium is not trying to turn engineering into a token-burning contest… even if I accidentally may have just won (or lost) it.</p><p>We’re trying to learn where AI actually helps, and how to use it responsibly: what work it makes easier, what incentives it creates, where it gets in the way, and how it affects the world, for better and worse.</p><p>That matters here because our mission is to deepen understanding through human stories. As the internet fills up with low-effort content, those stories matter more than ever. Protecting them means we have to move with both speed <em>and</em> care.</p><p>AI is one of many way we increase our leverage, so we can keep building for readers and writers without losing sight of the human voices behind the work.</p><p>If you care about human stories, thoughtful engineering, and the future of online publishing, we’re hiring, and <a href="https://job-boards.greenhouse.io/medium">we’d love to meet you</a>.</p><p><a href="https://medium.com/jobs-at-medium/work-at-medium-959d1a85284e">Work at Medium</a></p><p>I was specifically curious about how far today’s frontier models could get with a constrained, mechanical migration.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3fb206cb65a7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Skill x Will: Turning Delegation into Development]]></title>
            <link>https://yipe.medium.com/skill-x-will-turning-delegation-into-development-c3e7e3a195ee?source=rss-b211895d1365------2</link>
            <guid isPermaLink="false">https://medium.com/p/c3e7e3a195ee</guid>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[work]]></category>
            <category><![CDATA[management]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[career-development]]></category>
            <dc:creator><![CDATA[Michael Margolis]]></dc:creator>
            <pubDate>Wed, 31 Dec 2025 00:30:13 GMT</pubDate>
            <atom:updated>2026-01-05T22:42:25.582Z</atom:updated>
            <content:encoded><![CDATA[<p>This is the first in a series of posts about management, drawn from patterns I’ve seen at work and in coaching.</p><p>The first topic is about a common model: Skill x Will. Particularly, how to use this model to assign work in a way that both <em>delivers today</em> and <em>builds tomorrow</em>. Not by working people harder, but by being more intentional about who gets which opportunities, and why.</p><p>Every team has them. The same few names that always come up when something big, urgent, shiny, or messy lands on the table. The “go-to” people. The reliable ones everyone depends on.</p><p>It feels efficient. It feels responsible. It gets the job done.</p><p>Yet over time, this habit creates quiet fragility and stalls growth. A few people sprinting while everyone else is left waiting. Burnout at the top, boredom in the middle, untapped potential everywhere else.</p><p>That’s not a people problem. It’s a structural problem, a by-product of how we as leaders decide who gets which work.</p><p>If you consistently delegate the most important work to the same high-skill, high-will people, the team may move fast, but it also becomes fragile. Over time, the organization slows as those team members become dependencies and bottlenecks. Worse, they quietly turn into single points of failure when priorities shift or they inevitably leave the company.</p><p>It can feel like you’re putting out fires and moving quickly, but in the process you’re often trading away adaptability, the very thing you’ll need when the next crisis hits.</p><p>And worst of all, it can send a very dangerous message:</p><blockquote>Excellence is a punishment</blockquote><h3>Skill</h3><p>Most of us, especially early in our management careers, default to assigning work based on skill. Who can do it best? Who’s done it before? Who needs the least support?</p><p>It’s a reasonable and often effective instinct, especially under pressure, but it’s only half the equation. Skill optimizes for execution.</p><p>The missing half is will: the motivation, curiosity, and appetite to take something on. When you intentionally factor will alongside skill, staffing stops being triage and starts becoming a tool for development.</p><h3>Will</h3><p>If skill is about <em>capability</em>, will is about <em>energy</em>.</p><p>It’s the spark that makes someone lean in, ask questions, and want to figure things out, even without prior experience. It’s the appetite to learn, grow, add new skills to their repertoire, and take on new challenges.</p><p>We tend to underestimate will, especially in fast-moving organizations. It’s easy to see who can do the work. It takes more intention to notice who wants to, or who should, once the growth opportunity and impact are clear. That’s often where the most meaningful work happens, for both individuals and teams.</p><p>When we give work to people with high will, we create space for:</p><ul><li>Learning: skills and confidence grow</li><li>Mentorship: experienced teammates teach instead of firefight</li><li>Redundancy: fewer single points of failure and bottlenecks</li><li>Engagement: people feel trusted and invested in</li></ul><h3>Archetypes</h3><p>When you factor will into assignments, delegation shifts from heroics and burnout to multipliers and growth. One practical way to apply this is through a set of archetypes.</p><p>These archetypes aren’t about judging or stack-ranking people. They’re a way to notice where someone is <em>right now</em> and what might help them grow. Everyone moves between these boxes depending on the work, the moment, and the context. I’ve certainly been in all of them at different points in my career.</p><p>Used this way, staffing decisions stop being one-off calls and start compounding. Applied consistently, they help teams move faster, grow stronger, and stay engaged.</p><p>This is just one lens I’ve found useful. Feel free to adapt it or create your own version that better fits your style!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*V_iC0qvlw7agjwEHdPnu0w.png" /><figcaption>Note: As a manager, it’s your goal to move people <strong>up and to the right</strong> in this diagram, which is sometimes easier said than done.</figcaption></figure><p>In a well-functioning organization, most of the team will be Consistent Contributors: steady, capable, and dependable. Because they’re reliable, managers naturally turn to them or to Empowered Experts when work is important or urgent.</p><p>From there, different risks emerge depending on how assignments are made and supported.</p><h3>Risks</h3><p>Sometimes assigning work to your top performers can come with other hidden risks, and the archetypes can help you reason about them.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CV9FXlA_hbTNEFgeYoh66A.png" /><figcaption>Alliterative Assignment Archetypes. Source: <a href="http://yipe.medium.com">yipe.medium.com</a></figcaption></figure><p>Even with good intentions, common assignment patterns can create predictable failure modes:</p><ul><li>Alicia, an Apathetic Achiever, was one mis-scoped project and one unfair decision away from checking out. High skill paired with low morale turned her bitter, and that negative energy began to spread through the team.</li><li>Ernesto, an Empowered Expert, was stretched too thin and under-appreciated. Over time, he quietly slid into disengagement.</li><li>Claudia, a Consistent Contributor, was pushed hard in the name of growth but never understood the why behind the work or the support available to them. Without either, they lost motivation and drifted toward apathy.</li></ul><p>When seemingly content talent is under-leveraged, under-recognized, and under-developed, you’ll find your <em>best employees</em> quickly become <em>ex-employees.</em></p><h3>Designing for Both</h3><p>If you only delegate to your high-skill, high-will people, the team may move fast, but they are likely to burn bright and burn out. The real opportunity is not choosing between speed and growth, but designing for both.</p><p>When you approach delegation this way, it becomes more than a way to get work done. It becomes one of the primary ways you build people and teams that can deliver well, sustainably. Done consistently, it moves people up and to the right over time, rather than out.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uPGKN-557s34o28jpuUEwQ.png" /></figure><h3>Pairing in action</h3><p>One practical way to apply this model is through intentional pairing. The goal is not balance for its own sake, but asymmetry that accelerates learning, redistributes load, and compounds capability.</p><ul><li><strong>Apathetic Achievers + Bright Beginners</strong> create mentorship, accountability, and renewed energy on both sides. Bright Beginners gain momentum and confidence on a path toward becoming Consistent Contributors and, eventually, Empowered Experts.</li><li><strong>Empowered Experts + Weary Workers</strong> can restore context, purpose, and momentum. With the right support, Weary Workers often regain confidence and move back into Bright Beginners or Consistent Contributors.</li><li><strong>Apathetic Achievers + Consistent Contributors</strong> create accountability and recalibration. Consistent Contributors model steady ownership and execution, which can help re-anchor Apathetic Achievers and pull them out of their rut when the work has clear outcomes and tight feedback loops.</li></ul><h3>Mindfulness &amp; Support</h3><p>When you are doing a talent review, assigning a project, or deciding who should take on a new responsibility, don’t just ask who can do it best and move on. Pause and be mindful of the tradeoffs, the support required, and the growth opportunities she is creating with that choice.</p><p>Different situations tend to call for different kinds of support:</p><ul><li><strong>Bright Beginners</strong> often need structure, feedback, and early wins.</li><li><strong>Weary Workers</strong> may need training, direction, clarity, and purpose.</li><li><strong>Consistent Contributors</strong> benefit from recognition and stretch opportunities.</li><li><strong>Apathetic Achievers</strong> often need meaningful ownership and a challenge worth caring about.</li><li><strong>Empowered Experts</strong> need trust, protection from overload, and opportunities to multiply their impact by growing others.</li></ul><h3>Delivering Today While Building Tomorrow</h3><p>The point of this model isn’t to perfectly classify people or optimize every assignment. It’s to be more intentional about who gets which opportunities, and why.</p><p>When you do that, delegation stops being just a way to get work assigned and becomes one of the primary ways you shape your team. You can still deliver today, while also building capacity for tomorrow, spreading ownership, growing new strengths, and reducing fragile dependencies.</p><p>Great managers don’t just ask, “Who can do this fastest?”</p><p>They ask, “Who should do this next, and what support will help them grow as they do it?”</p><p>At the end of the day, the real measure of a manager isn’t just what gets delivered, but whether her decisions leave people and teams stronger than she found them.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c3e7e3a195ee" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[We’re Hiring a Principal Backend Engineer to Shape the Future of Medium]]></title>
            <link>https://medium.engineering/were-hiring-a-principal-backend-engineer-to-shape-the-future-of-medium-a0d7896b3717?source=rss-b211895d1365------2</link>
            <guid isPermaLink="false">https://medium.com/p/a0d7896b3717</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[medium]]></category>
            <category><![CDATA[backend-development]]></category>
            <category><![CDATA[backend]]></category>
            <category><![CDATA[hiring]]></category>
            <dc:creator><![CDATA[Michael Margolis]]></dc:creator>
            <pubDate>Wed, 22 Oct 2025 16:17:27 GMT</pubDate>
            <atom:updated>2025-10-29T18:25:08.051Z</atom:updated>
            <content:encoded><![CDATA[<p>When <a href="https://medium.com/@yipe/why-im-joining-medium-0b0479080e18">I joined Medium</a>, it was because I still believe in the internet as a place for ideas. Not noise, not outrage, but content that connects people through shared curiosity.</p><p>Medium’s mission is to deepen understanding and spread ideas that matter. We’re building the best place for reading and writing online: A space that rewards clarity, authenticity, and craft over clickbait and engagement hacks.</p><p>That vision only works if the technology beneath it is as thoughtful as the writing above it. That’s where <strong>you</strong> come in. <em>Yes, you!</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*v_FWVO8mT1JDGUxb" /><figcaption>Photo by <a href="https://unsplash.com/@yanu?utm_source=medium&amp;utm_medium=referral">Yannick Pulver</a> on <a href="https://unsplash.com?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>The Role</h3><p>We are starting the search for a <a href="https://job-boards.greenhouse.io/medium/jobs/4048233009"><strong>Principal Backend Engineer</strong></a> at Medium, our most senior IC level and a leadership position at the company. In this role, you will lead the evolution of Medium’s core backend, the systems that power reading, writing, and discovery for millions of users every day.</p><p>This is a <strong>deeply</strong> <strong>hands-on</strong>, <strong>high-impact role</strong>. You’ll help <strong>craft the strategy for where we want to be as an engineering organization on the backend</strong>, and be accountable for the <strong>quality, standards, and evolution of the Medium backend platform</strong>.</p><p>You will:</p><ul><li><strong>Partner</strong> with engineering, design, data, and executive leadership to define and deliver the future of Medium.</li><li><strong>Make</strong> our platform not just reliable, but genuinely delightful to build on.</li><li><strong>Modernize</strong> a large, living codebase so our teams can move faster and ship with confidence.</li><li><strong>Lead</strong> through influence by setting standards, reviewing critical code and designs, and mentoring engineers across teams.</li><li><strong>Shape</strong> how we build, ensuring our architecture, practices, and systems are strong, scalable, and aligned with Medium’s long-term goals.</li><li><strong>Balance </strong>vision with execution by building, debugging, and refining systems yourself to model technical excellence and thoughtful engineering craft.</li></ul><p>And you can do this work wherever you do your best thinking, whether that’s a home office, a beachside Airbnb, a rainy coastal cabin in the Pacific Northwest, or your favorite café with great Wi-Fi and even better croissants. This role is <strong>fully remote </strong>in the US<strong>.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*N2DjaEJ0Ffyr5Ysz" /><figcaption>Photo by <a href="https://unsplash.com/@proskurovskiy?utm_source=medium&amp;utm_medium=referral">Volodymyr Proskurovskyi</a> on <a href="https://unsplash.com?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>Why This Role Matters</h3><p>I’ve been lucky enough to work with some truly great Principal Engineers in my career, the kind of people who bend the trajectory of a company. They don’t just write brilliant code, they <strong>change how everyone around them thinks about building</strong>. They spot patterns before they calcify into problems, simplify where others complicate, and bring calm, clear reasoning to complex systems and high-stakes decisions. When they get it right, everyone moves faster, builds better, and understands the system (and the mission) more deeply.</p><p>That’s what makes this role special. It’s not just about scale, it’s about <strong>stewardship</strong>.</p><h3>About You</h3><p>You find joy in making <strong>complex systems feel simple</strong>, in building things that <strong>help others move faster and think deeper</strong>. You leave codebases better than you found them, teams more confident than before you joined them, and products more human in how they behave.</p><p>You’ve <strong>built and scaled</strong> systems that millions rely on, and you’ve learned that reliability isn’t just uptime, it’s <strong>trust</strong>. You’ve wrestled with technical debt, fan-outs, and edge cases galore, and found ways to turn them into leverage, not legacy.</p><p>You lead with curiosity, not control. You ask great questions, create shared understanding, and help others see the system as clearly as you do.</p><p>And through it all, you never forget who it’s for: the writers and readers who come to Medium to make sense of the world.</p><h3>Why Medium</h3><p>To me, Medium has always been a rare and wonderful corner of the internet. A place where people come to slow down, think deeply, reflect, and share real human experiences.</p><p>In an era where misinformation, AI, and bots are reshaping public discourse, it’s <strong>more important than ever to elevate real human voices</strong>. That’s only possible if the technology underneath is <strong>fast, stable, and built for longevity</strong> so we can keep Medium human-centered, sustainable, and ready for what’s next.</p><p>If you’re a builder who loves the elegance of distributed systems and the beauty of human stories, <a href="https://job-boards.greenhouse.io/medium/jobs/4048233009">we’d love to meet you</a>.</p><p>Let’s build a better internet together. One that rewards ideas, not division.</p><p><strong>👉 </strong><a href="https://job-boards.greenhouse.io/medium/jobs/4048233009"><strong>Apply here</strong></a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*93eNUm5UN7tYJayn8iXfeA.jpeg" /><figcaption>Source: <a href="https://www.linkedin.com/posts/medium-com_some-publishing-platforms-refuse-to-moderate-activity-7349463524403757056-QMfB/">Medium</a></figcaption></figure><ul><li><a href="https://medium.com/jobs-at-medium/work-at-medium-959d1a85284e">Work at Medium</a></li><li><a href="https://medium.com/me/following-feed/publications/2817475205d3">Following - Medium Engineering</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a0d7896b3717" width="1" height="1" alt=""><hr><p><a href="https://medium.engineering/were-hiring-a-principal-backend-engineer-to-shape-the-future-of-medium-a0d7896b3717">We’re Hiring a Principal Backend Engineer to Shape the Future of Medium</a> was originally published in <a href="https://medium.engineering">Medium Engineering</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why I’m Joining Medium]]></title>
            <link>https://yipe.medium.com/why-im-joining-medium-0b0479080e18?source=rss-b211895d1365------2</link>
            <guid isPermaLink="false">https://medium.com/p/0b0479080e18</guid>
            <category><![CDATA[careers]]></category>
            <category><![CDATA[ethics]]></category>
            <category><![CDATA[medium]]></category>
            <category><![CDATA[authenticity]]></category>
            <category><![CDATA[internet]]></category>
            <dc:creator><![CDATA[Michael Margolis]]></dc:creator>
            <pubDate>Tue, 16 Sep 2025 16:15:56 GMT</pubDate>
            <atom:updated>2025-09-16T16:15:56.630Z</atom:updated>
            <content:encoded><![CDATA[<h4>Building a Better Internet Together</h4><p>Growing up in the early days of the internet, I fell in love with technology because of what it made possible. The idea that <strong>anyone, anywhere, could share their ideas and connect with people around the world</strong> felt like magic. Sure, back then the internet was messy, slow, and chaotic, but it was undeniably <em>human</em>.</p><p>Early in my career, I believed deeply in that promise. If we built the right products, we could make the world <strong>more connected, more open, and more humane</strong>. That belief has been a driving force throughout my career, leading me to create products that help people <strong>do more, reach further, learn faster, and become better</strong>.</p><p>Over the last 25 years at companies like Apple, Twitter, and Dropbox, I’ve seen how even <em>small</em> design choices can ripple out to hundreds of millions of people. Shaping how we connect, what we believe, even who we become.</p><p>And while my optimism hasn’t faded <em>that</em> much, I’ve seen how easily those same tools can be twisted and how difficult it is to <strong>steer them back toward good</strong> once the underlying incentives go wrong.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*FALyZ_AF_lGH2EOB" /><figcaption>Photo by <a href="https://unsplash.com/@bradyn?utm_source=medium&amp;utm_medium=referral">Bradyn Trollip</a> on <a href="https://unsplash.com?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h4>The Road Paved with Good Intentions</h4><p>At Twitter, I led various engineering teams that crafted features like quote tweets, threaded conversations, and a more real-time ranked timeline. All of these were designed help people find and engage with what mattered most.</p><p>The people I worked with were brilliant and <strong>deeply committed to making the world better</strong>, yet we watched time and time again how features meant to foster connection were used to spread disinformation, deepen divisions, and amplify hate <em>despite</em> our best efforts to stop it.</p><p>It starts innocently enough. A much-anticipated feature to boost engagement here, an algorithm tweak to surface trending topics there. But over time, through a<strong> </strong>thousand small decisions, a pattern emerges.</p><blockquote>Outrage grabs attention → attention attracts eyeballs → eyeballs drive ad revenue</blockquote><p>Like it or not, <strong>Enragement = Engagement</strong>, and once something becomes <em>profitable</em>, companies rarely stop doing it.</p><p>Leaders often invoke ideals like <em>“the best solution to bad speech is more speech” </em>which is noble, but <strong>sidesteps accountability for the systems they’ve built</strong>. More importantly, it ignores the real problem: <strong>the incentives themselves</strong>. Systems designed to maximize engagement don’t care whether they amplify truth or lies, joy or rage, real voices or bots.</p><p>Unless platforms are willing to <strong>intentionally sacrifice profits</strong> to curb that content, this cycle may never end.</p><p>My time at Twitter taught me a painful truth. <strong>When systems reward the wrong outcomes, even the best intentions can produce results you never wanted</strong>. There were moments when it felt like the better we were at our jobs, the worse the world seemed to get. That realization has shaped how I think about software, systems, and incentives ever since.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*gpPECqc4V_yLHotH" /><figcaption>Photo by <a href="https://unsplash.com/@jasonrosewell?utm_source=medium&amp;utm_medium=referral">Jason Rosewell</a> on <a href="https://unsplash.com?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>The Flood and the Slide</h3><p>Fast forward to 2025.</p><p>AI is accelerating the problem, but not because AI <em>itself</em> is bad. Used responsibly, it can be a powerful tool that helps writers grow, learn, and express their ideas more clearly.</p><p>But <strong>the same tools that can be used for good can also be misused</strong>. AI now makes it easier than ever for bad actors to flood feeds with low-quality synthetic content that blends seamlessly with real voices. Each day, it becomes harder to tell what’s authentic, or even to know who is behind the narratives being spread.</p><p>Even Sam Altman recently posted <a href="https://x.com/sama/status/1963366714684707120">this</a> on X</p><blockquote>“I never took the dead internet theory that seriously but it seems like there are really a lot of LLM-run Twitter accounts now.”</blockquote><p>This flood is only part of the problem. Today’s media landscape is also deeply troubled. The <strong>Overton Window</strong> (the range of ideas considered acceptable in public discourse) has turned into more of an <strong>Overton Slide</strong>, tilting ever downward, as sensationalism and outrage dominate the media we consume.</p><p>In this environment, it’s more important than ever to have a safe haven for the <a href="https://en.wikipedia.org/wiki/Fifth_Estate">fifth estate</a>: independent writers, thinkers, and creators who <strong>spark</strong> <strong>meaningful dialogue </strong>and hold power accountable. These are the voices that help society make sense of the world and they are being drowned out.</p><p><strong>The stakes are high</strong>. We’re already living in a world where it’s harder and harder to find what’s meaningful, or even to know what’s real. But it’s <strong><em>not too late to change course</em></strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*CY6ZY0cFRzxkrZXA" /><figcaption>Photo by <a href="https://unsplash.com/@wcwarren?utm_source=medium&amp;utm_medium=referral">Wes Warren</a> on <a href="https://unsplash.com?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>Why Medium</h3><p>After seeing how engagement-driven incentives can distort conversation, I wanted to work somewhere that was <strong>actively building a different kind of internet</strong>. Somewhere that put <strong>principles</strong><em> </em><strong>ahead of growth at all costs</strong>.</p><p>That’s why I’m excited to be joining <a href="https://medium.com/"><strong>Medium</strong></a> next week as VP of Engineering. To help protect and amplify real human voices at a moment when they matter most.</p><p>Medium works differently. Readers pay directly for a medium subscription and writers succeed by delivering meaningful content. This creates a natural balance and <strong>pure incentives</strong>. When readers thrive, writers thrive too, and growth only happens when both sides win.</p><p>Instead of feeds driven purely by algorithms that reward the loudest or most incendiary voices, Medium blends technology with <strong>human judgment</strong>. Editors elevate thoughtful, nuanced writing that might otherwise get buried, helping quality rises above the noise.</p><p>Many companies have values written on a wall or in a slide deck, but Medium backs them up with real action to <strong>protect both its content and its community</strong>.</p><ul><li>Last year alone, the team removed <a href="https://medium.com/blog/state-of-medium-c54d1706a9b4">over ten million spammy or AI-generated posts</a>, keeping the platform cleaner and more focused on authentic voices and real stories.</li><li>Their <a href="https://www.linkedin.com/posts/medium-com_some-publishing-platforms-refuse-to-moderate-activity-7349463524403757056-QMfB/">recent marketing campaign</a> made a clear, public statement. Hate speech and harassment will not<strong> </strong>be tolerated or monetized here.</li><li>Medium invests directly in creators through their writer partnership program that supports independent writers, essayists, and thinkers, giving them both the <a href="https://medium.com/medium-handbook/editor-newsletter-a-deep-dive-into-your-new-editor-inbox-1759778ee492">tools</a> and the incentives they need to keep sharing their unique stories and perspectives.</li></ul><p>Most recently, Medium stepped in to support <a href="https://rslstandard.org/">Really Simple Licensing</a>, an open standard that lets writers tag their work with clear permissions about how it can be used by AI companies. At a time when so much content is scraped without consent, this promises to give creators a voice in how their words are used. It’s a <strong>tangible example of Medium taking a principled stand </strong>to defend human creativity and build a fairer, more transparent internet.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*zRNjHlsCFZ25j22p" /><figcaption>Source: <a href="https://www.linkedin.com/posts/medium-com_some-publishing-platforms-refuse-to-moderate-activity-7349463524403757056-QMfB/">Medium on LinkedIn</a>.</figcaption></figure><h3>The Antidote</h3><p>This isn’t just some career move for me. It’s a deeply personal drive to push back against algorithms that reward outrage, the AI slop that floods our feeds with noise and disinformation, and the platforms that profit from hate and division.</p><p><strong>The internet doesn’t have to be this way,</strong> and I see Medium as an antidote. A rare corner of the internet where <strong>thoughtful conversation is rewarded</strong>. Where <strong>substance matters more than sensationalism</strong>. Where <strong>quality beats outrage</strong>. Where writers can reach readers who want <strong>more than hot takes</strong>. Where people <strong>enrich one another rather than divide</strong>.</p><p>In a world of stormy seas and treacherous currents, <strong>Medium is a lighthouse </strong>guiding readers toward quality, authenticity, and human connection. When Medium succeeds, so do the values it stands for. That growth isn’t just about numbers, it’s proof that <strong>a better internet is possible</strong>.</p><p><em>That’s</em> the future I want to build at Medium: a home for ideas, connection, and authentic human voices.</p><p>And I can’t wait to get started.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*N1aUeHlpVedKoK1lr5cy7A.png" /><figcaption>A photo outside Cafe Bambino in Seattle, WA</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0b0479080e18" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>