<?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 Luc Closser on Medium]]></title>
        <description><![CDATA[Stories by Luc Closser on Medium]]></description>
        <link>https://medium.com/@closerluc?source=rss-3fa5dbac75af------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*Gm5Da3-LHpSHn3-IbATmYg@2x.jpeg</url>
            <title>Stories by Luc Closser on Medium</title>
            <link>https://medium.com/@closerluc?source=rss-3fa5dbac75af------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 10:27:52 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@closerluc/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 Harness May Be the Problem]]></title>
            <link>https://medium.com/@closerluc/the-harness-may-be-the-problem-70ccccf08f31?source=rss-3fa5dbac75af------2</link>
            <guid isPermaLink="false">https://medium.com/p/70ccccf08f31</guid>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[agent-harness]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Luc Closser]]></dc:creator>
            <pubDate>Tue, 22 Sep 2026 13:17:35 GMT</pubDate>
            <atom:updated>2026-09-22T13:22:08.969Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CCVd6i_w6suIp36epxq7BQ.png" /></figure><p>For the last several months, I have been building and testing agent harnesses, both my own and other people’s: systems that coordinate multiple agents, pass work between them, and constrain their behavior with deterministic tools.</p><h3>Why I built a harness</h3><p>The underlying idea was straightforward. A group of agents should be able to take on work that a single agent cannot. The harness would manage the handoffs without allowing agents to pollute each other’s context, and the tools would keep them from wandering off and doing dumb things.</p><p>I put a lot of work into making that system behave. Eventually, it did.</p><p>I have also spent time using other people’s harnesses. They have reinforced the same concern: the orchestration often becomes a large part of the problem it is meant to solve. The company I work for is moving toward a custom-built harness, and I think that is the wrong direction. I may be proven wrong. I am waiting for the evidence.</p><h3>The single-agent experiment</h3><p>Then I measured it, and tried a simpler experiment: I gave one agent the same kind of work.</p><p><em>Spoiler Alert: The single agent did better, by a fairly large margin.</em></p><p>That was not the result I expected. The harness was using the same model, so the model itself was not the difference. Yet the orchestrated version needed three or four hours to produce something acceptable, and it still was not especially good. A single model instance could do the job in about 40 minutes and produce a stronger result.</p><p>At some point I looked at the output from the individual agent and thought: wait a minute. They did not used to be this good.</p><p>They have improved a lot.</p><h3>Loosening the constraints</h3><p>So I ran more experiments. I gave a single agent a substantial task with a simple prompt, relaxed several of my usual constraints, and let it work for 40 minutes. I asked it to write unit tests, do mutation testing, avoid Gherkin, and keep the CRAP score low.</p><p>It produced a remarkably good outcome.</p><p>It was not perfect. There were a few minor points that could have been improved. But it was far better than what the harness had produced. The agent had broken the work into modules that made sense, and the resulting complexity stayed low.</p><p>That part was interesting in its own right. The more important question was what the tools had contributed.</p><h3>When the tools leave traces</h3><p>Mutation testing did find bugs and omissions. It helped. But it also left traces in the code. When I traced those back, I found that the mutation-testing algorithm had sometimes pushed the agent into doing things that were plainly stupid. It imposed good pressure in some places and bad pressure in others.</p><p>That is a useful reminder: a constraint can improve an agent’s output while still distorting the work. A harness may make an agent more predictable, more measurable, and easier to control. It can also add overhead, fragment the task, and steer a capable model away from the judgment it would have exercised on its own.</p><h4>Guardrails that the agent cannot bypass</h4><p>I have found that orchestration is most useful when it moves the guardrails outside the agent. Hooks and external tools can enforce rules the agent cannot ignore or bypass: linting is the obvious example. Agents are lazy by design. They will take the path that completes the task with the least resistance, which is often helpful, but it also means that <strong>instructions alone are weak guardrails</strong>. Enforcement has to live somewhere the agent cannot negotiate with.</p><h3>The default is changing</h3><p>For a long time, the working assumption was that capability lived in the orchestration. The individual agent was unreliable, so we needed a structure around it: specialized roles, controlled handoffs, deterministic tools, test gates, and carefully designed feedback loops.</p><p>That assumption may now be outdated for some kinds of work.</p><p>The models are getting better fast enough that the harness can become the limiting factor. If the model can hold the task together, decompose it sensibly, write the code, test it, and revise it over a sustained session, then a complicated orchestration layer may be doing more damage than good.</p><p>This does not mean harnesses are useless. There are still tasks that need isolation, repeatability, auditability, or access to specialized tools. There are also cases where one agent should not be trusted with the whole job. But the default should change.</p><p>Instead of beginning with a multi-agent architecture, start with one strong agent and a simple, well-defined task. Measure the result. Use hooks and external tools to enforce the rules that matter. Add more agents only when a single agent has demonstrated a real limitation.</p><p>The question is no longer whether agent orchestration can work. It can.</p><p>The question is whether the orchestration is still helping.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=70ccccf08f31" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Nobody’s Body Is On The Line]]></title>
            <link>https://medium.com/@closerluc/nobodys-body-is-on-the-line-ca795ddd816b?source=rss-3fa5dbac75af------2</link>
            <guid isPermaLink="false">https://medium.com/p/ca795ddd816b</guid>
            <category><![CDATA[software]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[accountability]]></category>
            <dc:creator><![CDATA[Luc Closser]]></dc:creator>
            <pubDate>Tue, 07 Apr 2026 11:01:23 GMT</pubDate>
            <atom:updated>2026-04-07T14:03:30.530Z</atom:updated>
            <content:encoded><![CDATA[<h4>The software we’re shipping now behaves like an organism. We’re still pretending it’s a machine.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jd4vdCoVNzloIpI_vYhkHw.png" /></figure><p>In science fiction, alien ships are often organic. No sharp edges, no visible welds, no obvious logic to their structure. They look grown, not built. We always assumed that was just a storytelling shortcut for <em>otherness</em>. It wasn’t. It was an intuition about something we hadn’t built yet: systems that behave without legible rules. For seventy years we built software that was brittle, predictable, and when it failed, it failed the same way every time. That consistency was a bug we accidentally turned into the foundation of an entire trust infrastructure: legal frameworks, audit trails, engineering accountability. We just replaced it with something more capable and far less predictable. We haven’t noticed yet what we lost in the trade.</p><p>We didn’t build trust into software deliberately. We backed into it. When a system failed, it did so consistently and predictably, regardless of the input. This reproducibility made it easier to debug and assign blame, it made failure <em>legible</em>. And legible failure is the only kind you can build accountability around. You can write a contract against it. Audit it. Point to it in court. The entire legal and engineering culture around software assumed, without ever saying so, that the failure mode was knowable in advance. Brittle was a flaw. Predictable was the invisible load-bearing wall.</p><blockquote>“Brittle was a flaw. Predictable was the invisible load-bearing wall.”</blockquote><p>Agentic software removes that wall. Not because it’s less reliable — in many contexts it’s more capable than anything we’ve built before. But its failures are contextual, variable, and difficult to reproduce. The same agent, the same task, a different day — different outcome. You can’t write a contract against that. You can’t point to it in court with confidence. You can’t audit a decision that won’t make itself the same way twice. This isn’t a bug to be fixed in the next release. It’s a category change. We moved from engineering a machine to deploying something closer to a judgment — and we’re only beginning to understand what that requires from us in return.</p><p>We’ve always tolerated organic unreliability when accountability is clear. Surgeons make mistakes, lawyers misread cases, analysts miss signals. We built centuries of infrastructure around exactly that — licensing, malpractice, peer review, traceable decisions, professional identity at stake. It took generations to make human fallibility legible enough to trust. Barry Marshall couldn’t convince the medical establishment that stomach ulcers were caused by bacteria. So he drank the culture himself. Got sick. Proved it. Won the Nobel Prize. His body was the guardrail. Not a committee, not a framework, not a terms of service document. One person absorbing the full consequences of being wrong. We’re deploying systems that will affect millions of decisions and nobody’s body is on the line for any of them.</p><p>The alien ship was never the scary part. The scary part is that we’re flying it before we understand its biology. We need something that makes failure legible, traceable, and contestable. We don’t have the concept yet, let alone the institution. This piece doesn’t have a solution. Neither does anyone else. And until we admit that openly, we’ll keep mistaking deployment for progress.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ca795ddd816b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Puzzle Inside, Plug Outside: What Protocols Like MCP Teach Us About Better Architecture]]></title>
            <link>https://medium.com/@closerluc/puzzle-inside-plug-outside-what-protocols-like-mcp-teach-us-about-better-architecture-1f2a862df262?source=rss-3fa5dbac75af------2</link>
            <guid isPermaLink="false">https://medium.com/p/1f2a862df262</guid>
            <category><![CDATA[software-architecture]]></category>
            <category><![CDATA[model-context-protocol]]></category>
            <category><![CDATA[design-systems]]></category>
            <category><![CDATA[agent-systems]]></category>
            <category><![CDATA[api-design]]></category>
            <dc:creator><![CDATA[Luc Closser]]></dc:creator>
            <pubDate>Wed, 09 Apr 2025 14:20:28 GMT</pubDate>
            <atom:updated>2025-04-09T14:20:28.742Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*1P6sURhKyWXMYmK5" /><figcaption>Photo by <a href="https://unsplash.com/@nevenkrcmarek?utm_source=medium&amp;utm_medium=referral">Neven Krcmarek</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>Build for Today, Think for Tomorrow</h3><p>When designing architecture, we often fall into two traps: <strong>over-engineering</strong> for an imagined future, or <strong>under-engineering</strong> with no thought to change. The real challenge is building systems that are <strong>shaped for today’s needs</strong> but can <strong>adapt to tomorrow’s demands</strong>. That balance isn’t accidental — it’s architectural.</p><p>I’ve come to think of it this way:</p><blockquote><strong><em>Inside your system: architecture should be a puzzle.</em></strong><em><br></em><strong><em>At the boundaries: it should behave like a plug.</em></strong></blockquote><p>Why puzzle and plug? Internally, we want <strong>singular fits</strong> — shapes that enforce discipline. Externally, we want <strong>interchangeable connectors</strong> — defined specs, not rigid shapes. The puzzle shapes ensure intentional composition; the plug shapes ensure swappable contracts. Both are about <strong>fit</strong>, but one is about <strong>structure</strong>, the other about <strong>compatibility</strong>.</p><p>Let’s unpack that.</p><h3>The Puzzle: Internal Precision</h3><p>Internally, components should <strong>only fit together in one way</strong>. Like puzzle pieces, each part of the system should have a clear role, well-defined edges, and a specific place in the bigger picture.</p><p>This approach brings several benefits:</p><ul><li><strong>Ownership is enforced</strong> through clear boundaries.</li><li><strong>Debugging and testing</strong> become easier because dependencies are obvious.</li><li><strong>Stability</strong> is achieved by reducing accidental coupling.</li></ul><p>Think of a typical backend service: its auth module should only talk to a user service in one way. The business logic should only reach into the database through a well-defined persistence layer. These relationships are shaped — intentional — like a puzzle.</p><p>This is aligned with ideas in Domain-Driven Design and Clean Architecture: <strong>cohesive internals that guide correct composition</strong>.</p><h3>The Plug: External Flexibility</h3><p>But as soon as you step outside those boundaries — into the realm of <strong>frontends, external services, or other teams</strong> — you need a different mindset. Here, your system must behave like a <strong>socket</strong>: expose a clear contract, support adapters, and welcome changes without breaking downstream consumers.</p><p>Take the classic frontend-backend relationship:</p><ul><li>The frontend doesn’t need to know <em>how</em> a request is processed — only <em>what</em> the response looks like.</li><li>The backend could be rewritten in Go, Python, or Rust, and as long as the <strong>API contract</strong> remains intact, the frontend doesn’t care.</li></ul><p>This is the essence of a <strong>plug interface</strong>: it doesn’t care what’s behind the wall — it just needs the right shape and power.</p><p>This thinking aligns with the Interface Segregation Principle and the idea of semantic versioning: <strong>boundaries that enable evolution without disruption</strong>.</p><h3>Protocol-Driven Plugs: Lessons from the Model Context Protocol (MCP)</h3><p>Protocols like the <strong>Model Context Protocol (MCP)</strong>, pioneered by Anthropic, push this plug concept even further. MCP defines a shared interaction format between language models, tools, and orchestrators — allowing independently developed agents and systems to collaborate in structured, multi-turn workflows.</p><p>MCP is <strong>pure plug architecture</strong>:</p><ul><li>Every participant (agent, tool, client) conforms to a shared contract.</li><li>No system needs to know what’s behind the plug — just the invocation format, observation schema, and allowed actions.</li><li>This enables <strong>interoperability</strong>, <strong>modularity</strong>, and <strong>tool composition</strong> at scale.</li></ul><p>But once you adopt MCP, your implementation must conform. You shape your internals — the logic of your tool or agent — to fit the required protocol. The internals become <strong>puzzle pieces</strong> crafted around the expected interaction pattern.</p><blockquote><em>MCP is a real-world example of this principle at scale: </em><strong><em>plug-shaped boundaries that enable puzzle-shaped internals to interoperate</em></strong><em>.</em></blockquote><p>What makes MCP particularly elegant is its <strong>strict separation of concerns</strong>: it doesn’t just define how components communicate — it defines a model-agnostic protocol for structured, multi-agent tool use. This is architectural discipline applied to the boundaries.</p><p>Even better, MCP components are <strong>self-describing</strong>: agents publish structured metadata about what they can do, what parameters they expect, and how to interact with them. This is like a plug that tells you its voltage, shape, and safe use cases — it’s not just a port, it’s a documented promise.</p><p>Finally, MCP’s design makes usage <strong>traceable and observable</strong>. Every call, observation, and reply follows a standardized pattern. You don’t just build for change — you build for <strong>transparency, testability, and evolution</strong>.</p><p>It also introduces an advanced idea: <strong>sometimes plugs don’t connect directly — they go through a standardized adapter layer</strong>. That layer becomes the universal socket, like MCP, letting otherwise unrelated parts work together.</p><p>This is increasingly relevant in systems where AI, microservices, or serverless tools must coordinate. A shared protocol makes it all possible — but it also demands internal discipline.</p><h3>Case Study: Swapping the Backend</h3><p>A while ago, we had a Node.js backend serving data to a React frontend. The frontend team was fast and independent, so we agreed on OpenAPI specs for our contracts. We documented not just the schema, but also response times, auth flows, and error shapes.</p><p>A year later, we had to replatform. We were moving to a microservice setup using Go, and our first instinct was panic: “How do we rewrite everything in time?”</p><p>But something critical saved us. Because the contracts were <strong>plug-shaped</strong>, not <strong>puzzle-shaped</strong>, we didn’t need to coordinate deeply with the frontend team. We could change everything behind the wall, and as long as the API interface matched the spec, nothing broke.</p><p>We did hit a snag: pagination. The old API returned flat arrays, but the new backend added metadata. A simple contract test flagged the issue — and because we treated the API as a plug, we adjusted the adapter without changing the frontend at all.</p><p>That’s the power of plug-based interfaces. Meanwhile, inside the Go service, we shaped the logic very differently — using goroutines, a caching layer, and a new database model. That part was a puzzle, carefully assembled.</p><h3>Puzzle Inside, Plug Outside: A Pattern for Change</h3><p>Great systems are made of:</p><ul><li><strong>Tightly-coupled internals</strong>: where the puzzle pieces ensure cohesion.</li><li><strong>Loosely-coupled externals</strong>: where sockets allow adaptability.</li></ul><p>This gives us a principle:</p><blockquote><strong><em>Shape your systems like puzzles on the inside, but design your boundaries like electrical outlets.</em></strong></blockquote><p>You get:</p><ul><li><strong>Stability inside</strong>, where internal change is deliberate.</li><li><strong>Flexibility outside</strong>, where evolution is survivable.</li></ul><p>Here are three questions I now ask in every architectural decision:</p><ol><li>Is this an <strong>internal detail</strong> that benefits from tight shape control?</li><li>Is this a <strong>public interface</strong> that others rely on?</li><li>Are we designing the right <strong>adapter points</strong>, so future changes don’t ripple outward?</li></ol><h3>Avoiding Over-Engineering</h3><p>Too much flexibility leads to confusion. Too much rigidity leads to fragility. The art of software design is knowing where to draw those lines.</p><p>Not everything needs an abstraction. Not everything should be swappable. But at the seams — the <strong>customer-facing contracts, the inter-team APIs, the SDKs</strong> — you want plug behavior.</p><blockquote><em>“Premature generalization is just as dangerous as premature optimization.</em></blockquote><p>Design with intent. Not everything should be a plug. But what <em>must</em> be pluggable should be treated with the rigor of an electrical spec.</p><h3>Interfaces Are Promises</h3><p>Ultimately, architecture is about trust. Internal shapes enforce discipline. External interfaces offer freedom.</p><p>So next time you’re designing a system, ask:</p><ul><li>Is this an internal piece that should only fit one way?</li><li>Or is this a boundary that needs to flex over time?</li></ul><p>Design accordingly.</p><blockquote><em>Puzzle inside. Plug outside.</em></blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1f2a862df262" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Let the Job Be the Job]]></title>
            <link>https://medium.com/@closerluc/let-the-job-be-the-job-e0616253f256?source=rss-3fa5dbac75af------2</link>
            <guid isPermaLink="false">https://medium.com/p/e0616253f256</guid>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[work-life-balance]]></category>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[work-culture]]></category>
            <dc:creator><![CDATA[Luc Closser]]></dc:creator>
            <pubDate>Mon, 31 Mar 2025 20:13:28 GMT</pubDate>
            <atom:updated>2025-03-31T20:13:28.595Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>In tech, burnout isn’t a bug — it’s a feature. We celebrate overwork, reward visibility, and confuse struggle with value. It’s time we killed the hero myth and stopped performing productivity.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*HfBdJVNmjn92jl9P" /><figcaption>Photo by <a href="https://unsplash.com/@derstudi?utm_source=medium&amp;utm_medium=referral">Timon Studler</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>I heard a phrase recently that stuck with me:</p><blockquote><em>“We should be able to let the job be the job.”</em></blockquote><p>Meaning: work should be <em>part</em> of our lives — not the whole thing. It shouldn’t define who we are. It shouldn’t consume us. And yet, in tech — especially in startups — that boundary keeps blurring.</p><p>How often do we introduce ourselves with our job title?<br>How many of us build our identity around what we ship?<br>How many careers were built on burnout, blurred lines, and badge-of-honor exhaustion?</p><h3>Work as Identity</h3><p>At my company, there’s a guy whose Slack name is literally “Brendan Shaw (Mr. Kafka).” It’s a joke — but also a mirror. He’s not just good at Kafka. <em>He is Kafka.</em></p><p>It’s flattering — until you realize what it implies:<br>If you’re not fused with your tool, your title, or your domain… are you even trying?</p><p>In many high-growth environments, people aren’t just celebrated for doing great work — they’re celebrated for <em>becoming</em> the work.</p><p>“Extreme ownership” gets twisted into “extreme availability.”</p><p>“High performance” quietly becomes “no life outside of this.”</p><p>But that’s not sustainable. And worse — it’s not fair.</p><p>We should be able to:</p><ul><li>Say no without guilt.</li><li>Log off without shame.</li><li>Introduce ourselves without leading with our LinkedIn headline.</li></ul><h3>The Hero Myth</h3><p>There’s a common myth: that most of the work is done by a handful of “heroes” going 190%, while the rest carry the leftovers.</p><p>But is that really true?</p><p>Sometimes, those heroes are just overcompensating for systems that expect failure and rely on heroics to patch it — unclear priorities, broken delegation, misaligned incentives.</p><p>Sometimes the people working “regular hours” aren’t underperforming — they’re just doing their <em>actual</em> job, while others burn themselves out trying to patch everything else.</p><blockquote>We need to stop mistaking imbalance for virtue.</blockquote><h3>The Rise of Performance Theater</h3><p>Why does this keep happening?</p><p>Because most systems don’t know how to value quiet consistency.<br>They reward visibility, responsiveness, and sacrifice — traits that often have little correlation with long-term outcomes.</p><p>If you’re answering Slack at midnight or patching systems over the weekend, you’re seen as dedicated — even when the failure was avoidable.</p><p>So we perform.<br>We stay online late, not just to finish something, but to be <em>seen finishing</em>.<br>We play the role. We speak the language. We wear exhaustion like a badge.</p><blockquote>This isn’t <strong>productivity</strong>.<br>It’s <strong>performance theater</strong>.</blockquote><p>And it thrives in cultures that measure presence instead of outcomes.</p><blockquote><em>We don’t just overwork — we do it to be seen, to be valued, to prove we matter.</em></blockquote><h3>Why Work Feels Like Worth</h3><p>It’s cultural. We’re taught to be something, not do something. “What do you want to <em>be</em> when you grow up?” Not: “What impact do you want to have?”</p><p>It’s systemic — Output becomes identity. If you’re available 24/7, you must care. If you say no, you must not.</p><p>It’s generational — Many of us were handed the unspoken expectation that our parents’ sacrifices would be validated through our success. A good job wasn’t just about income. It was about proving you mattered.</p><p>And it’s social — Especially in startups, work offers belonging. Missions feel existential. “All in” becomes the norm. You don’t just work here. You <em>are</em> this company.</p><blockquote>If your value is only visible when you overextend, then you’re never allowed to stop.</blockquote><h3>You Are Not Your Job</h3><p>Your job is what you <em>do</em>, not who you <em>are</em>.</p><p>You’re allowed to <strong>be ambitious</strong> and <strong>still</strong> <strong>value rest.</strong><br>You’re allowed to <strong>work hard</strong> and <strong>not be available 24/7.</strong><br>You’re allowed to <strong>care deeply</strong> without <strong>collapsing your identity into a company logo</strong>.</p><h3>Let It Be Enough</h3><p>Last year, after a round of layoffs, I was working 11+ hours a day — trying to keep an 18-person dev team and four PMs productive. I thought that was leadership. I wore exhaustion like armor.</p><p>Now? After a reorg and support from a new VP and CTO, I work ~8 hours a day. The work is still impactful. My boss holds me and my team up as examples.</p><p>But I’ll be honest — it <em>feels</em> wrong.<br>Like I’m cheating the system. Like if I’m not overwhelmed, I must be underperforming.</p><p>But maybe that’s not my flaw.<br>Maybe that’s the default behavior the system taught me — one where exhaustion equals value, and balance looks like apathy.</p><p>I’m unlearning that now.</p><blockquote>Let the job be the job.<br>Let your life be yours.</blockquote><h3>So What Do We Do?</h3><p>Start small.<br>Set an end-of-day time and hold it sacred.</p><p>Let silence on Slack be proof of balance — not betrayal.<br>Let yourself log off when the work is done — not when you’re depleted.</p><p>And if your workplace only values you when you’re giving 190% — Maybe it’s time to value yourself enough to give them less.</p><h3>What I’m Learning as a Leader</h3><p>If you’re in a leadership role, you don’t just <em>shape outcomes</em>. You <em>shape culture</em>.</p><p>Ask yourself:</p><ul><li>Do I celebrate late-night heroics more than quiet consistency?</li><li>Do I model boundaries — or reward their erosion?</li><li>Do I create systems where sustainability is possible — or rely on burnout to meet goals?</li></ul><p>As leaders, we must:</p><ul><li>Normalize logging off.</li><li>Redefine performance around clarity, outcomes, and consistency — not just visibility.</li><li>Stop glorifying performance theater and start rewarding actual impact.</li></ul><p>If our teams believe exhaustion is the price of success, then we’re not leading — we’re performing too.</p><p>Let’s stop the show.</p><h3>How Systems Scale What We Signal</h3><p>Culture is what gets reinforced at scale.<br>And systems don’t just reflect values — they <strong>amplify</strong> whatever behaviors they reward.</p><p>If you celebrate the person who pulled an all-nighter, the system will optimize for more all-nighters.<br>If you promote based on visibility, you’ll build a team that prioritizes being <em>seen</em> over being <em>effective</em>.</p><p>Systems scale through:</p><ul><li><strong>Feedback loops</strong> (who gets promoted, who gets praised),</li><li><strong>Defaults</strong> (who gets added to on-call, how planning is scoped),</li><li>And <strong>incentives</strong> (what gets measured, what gets funded, what gets shoutouts).</li></ul><blockquote>Performance theater scales because it has surface area. Quiet, consistent value? That’s harder to measure — unless you design for it.</blockquote><p>If your systems don’t <strong>make space</strong> for sustainable work, they will always <strong>default</strong> to unsustainable work.</p><p>To shift culture, you have to redesign what success <em>looks like</em> at scale:</p><ul><li>Align rewards with outcomes, not optics.</li><li>Build metrics that reflect <em>health</em>, not just speed.</li><li>Track <em>team sustainability</em> as seriously as customer satisfaction.</li></ul><blockquote><em>Let the job be the job. Let the system support that. Let people stay whole.</em></blockquote><p>Or accept the cost of building organizations where no one lasts.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e0616253f256" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Tech Grew Too Fast During COVID — We’re Still Paying For It]]></title>
            <link>https://medium.com/@closerluc/tech-grew-too-fast-during-covid-were-still-paying-for-it-5e21bb778df9?source=rss-3fa5dbac75af------2</link>
            <guid isPermaLink="false">https://medium.com/p/5e21bb778df9</guid>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[post-covid]]></category>
            <category><![CDATA[organizational-culture]]></category>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[startup]]></category>
            <dc:creator><![CDATA[Luc Closser]]></dc:creator>
            <pubDate>Tue, 25 Mar 2025 17:16:58 GMT</pubDate>
            <atom:updated>2025-03-25T17:16:58.053Z</atom:updated>
            <content:encoded><![CDATA[<h3>Tech Grew Too Fast During COVID — We’re Still Paying For It</h3><p><em>The hidden organizational and technical debts from the COVID-era tech boom are still shaping the industry — and burning out the people left to clean it up.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*hadmYsU9VyoztrjT" /><figcaption>Photo by <a href="https://unsplash.com/@jan_genge?utm_source=medium&amp;utm_medium=referral">Jan Genge</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>There’s something I’ve been thinking about lately, and the more conversations I have, the more I realize it’s not just me.</p><p>A lot of companies today are quietly (or not so quietly) dealing with the consequences of decisions made around 2021 — that weird, explosive post-COVID moment when money was cheap, growth was everything, and “just ship it” was the unspoken mantra.</p><p>At the time, it felt like everything could be patched with cash. Need to scale? Hire 30 engineers. Platform unstable? Spin up a new service. Org processes messy? Hire a middle manager, or ten. There was no time to do things properly — only fast. And when there’s enough money flying around, the long-term cost feels like someone else’s problem.</p><p>But now, that bill is coming due.</p><blockquote>“Debt becomes toxic when it’s invisible, unmanaged, or denied.”</blockquote><h3>The Peak Wasn’t a Fluke</h3><p>Quick note: while the post-COVID tech boom started in late 2020 with cheap capital and sudden digitization, it was 2021 when hypergrowth peaked. That’s when most of the organizational and technical shortcuts were taken — and that’s the period I’m focusing on here.</p><blockquote>“The post-COVID tech boom lit the fuse, but 2021 was the explosion — and now we’re all cleaning up the debris.”</blockquote><p>Let’s be clear though: 2021 wasn’t a historical anomaly. Hypergrowth-at-all-costs is a recurring pattern in tech. What made 2021 different was scale and pace — not principle. In many cases, reckless scaling was a rational response to warped incentives. You couldn’t afford not to grow aggressively if your competitors were doing the same. The real question is whether we’ve learned how to structurally correct those incentives.</p><p>Take Shopify, for example. In 2022, they admitted their hiring bets didn’t pay off — and laid off 10% of their workforce. They weren’t alone. Stripe, Meta, and dozens of others followed.</p><h3>The Hidden Debts</h3><p>What we accumulated wasn’t just technical debt. We accumulated <em>organizational debt</em>, <em>cultural debt</em> — and, in many cases, <em>leadership debt</em>.</p><ul><li>Tech stacks that don’t scale.</li><li>Processes nobody understands.</li><li>Teams built too fast, with no time to stabilize.</li><li>People promoted without real growth or support structures.</li></ul><p>Debt in itself isn’t always bad. But debt becomes toxic when it’s invisible, unmanaged, or denied. What matters is your debt-to-asset ratio. What did we gain from that era — and was it worth the current cost of cleanup?</p><p>The problem is, you can’t raise your way out of this anymore. The market shifted. Capital is tight. And now the people who stuck around are the ones being asked to clean it up — often without the resources, support, or compensation to do it properly.</p><h3>People Are Tired</h3><p>What I’m seeing everywhere is a growing disconnect. People are leaving not just because of layoffs or better offers — they’re leaving because they’re done cleaning up messes they didn’t make. Or worse: they’re being sidelined as leadership tries to build around the problems rather than address them head-on.</p><p>This is how you end up with the “parallel build” pattern. Instead of fixing foundational issues, companies spin up new initiatives, new platforms, new products — anything to keep the <em>illusion of momentum</em>. But now you’ve multiplied the complexity, not solved it. And it gets even worse when there’s no real plan to consolidate or migrate later — those temporary paths become permanent liabilities.</p><h3>What Now?</h3><p>We can’t go back and undo 2021. But we can decide how we deal with its consequences.</p><p>Some companies are taking the hard path: investing in cleanup, confronting the debt, stabilizing before scaling. Those are the ones I believe will survive and rebuild trust.</p><p>Others are still pretending the party never ended. They’re building fast again — but this time with fewer people, less clarity, and even less tolerance for error. That’s a dangerous combination.</p><p>As individuals, I think we’re all sitting somewhere between three roles:</p><ol><li><strong>The Rebuilders</strong> — those staying to clean things up, reshape teams, and fix what’s broken.</li><li><strong>The Leavers</strong> — those jumping to somewhere that didn’t accumulate as much debt (or already paid it down).</li><li><strong>The Endurers</strong> — those who stay but disengage, adapt to dysfunction, and act as informal memory holders of broken systems.</li></ol><p>Where are you on this triangle today? Where do you want to be?</p><figure><img alt="A hand-drawn triangle diagram labeled with three roles: Rebuilders at the top, Endurers at the bottom left, and Leavers at the bottom right. Arrows point from Rebuilders to “Growth,” Leavers to “Opportunity,” and Endurers to “Burnout.” Each label includes a brief description: Rebuilders face problems and invest in cleanup; Leavers seek healthier environments; Endurers stay disengaged and cope with dysfunction." src="https://cdn-images-1.medium.com/max/805/1*Fh6yEe66MdOYM8_d7D3HJw.png" /><figcaption>The triangle of post-2021 roles. Only two paths are sustainable.</figcaption></figure><p>Each corner leads somewhere different:</p><ul><li>Rebuilders often grow into future leadership.</li><li>Endurers risk burnout, stagnation, or quiet quitting.</li><li>Leavers may find faster momentum elsewhere.</li></ul><p>None of these are inherently wrong. But long-term, only two of them are sustainable: <strong>rebuilding</strong> or <strong>leaving</strong>. The middle state — enduring — is corrosive. It leads to burnout, learned helplessness, and quiet quitting.</p><p>So if you’re feeling stuck: ask yourself which direction you’re choosing — toward rebuilding or leaving. But whatever you do, don’t pretend everything is fine.</p><p>Also, take a moment to reflect:</p><blockquote><em>What debts has your team inherited that you’ve been ignoring? What have you normalized that would’ve been unthinkable three years ago?</em></blockquote><h3>Forward Requires Friction</h3><p>We need to institutionalize long-term thinking:</p><ul><li>Build with migration in mind. If you have to build in parallel, make sure the path to consolidation is real — and funded.</li><li>Normalize saying “no” to features or deadlines that would lead to more debt.</li><li>Reward teams that invest in cleanup, quality, and clarity — not just those who deliver fast.</li><li>Make architectural and organizational reviews part of your process, not a luxury.</li><li>Align incentives across the org to favor thoughtful progress over superficial growth.</li></ul><p>Everyone is tempted to repeat 2021 when things get exciting again. The difference will be whether we’ve learned to build responsibly when the next wave comes.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5e21bb778df9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[La vulnerabilidad de las conexiones WiFi de Fibertel Argentina]]></title>
            <link>https://medium.com/@closerluc/la-vulnerabilidad-de-las-conexiones-wifi-de-fibertel-argentina-5ea5c9c503b5?source=rss-3fa5dbac75af------2</link>
            <guid isPermaLink="false">https://medium.com/p/5ea5c9c503b5</guid>
            <category><![CDATA[python]]></category>
            <category><![CDATA[contraseña]]></category>
            <category><![CDATA[seguridad-informatica]]></category>
            <category><![CDATA[hacking]]></category>
            <dc:creator><![CDATA[Luc Closser]]></dc:creator>
            <pubDate>Wed, 12 Sep 2018 22:37:27 GMT</pubDate>
            <atom:updated>2022-03-02T20:20:09.754Z</atom:updated>
            <cc:license>https://creativecommons.org/licenses/by-sa/4.0/</cc:license>
            <content:encoded><![CDATA[<h4>En menos de 30 segundos la información de nuestra red, puede quedar expuesta por una negligencia</h4><p>Cuando los técnicos instalan un nuevo servicio de WiFi, <strong>Fibertel</strong> por defecto provee una contraseña de <strong>10/11</strong> <strong>números </strong>que casualmente, no solo es el número de cliente que figura en cualquier factura, si no que también comparte <strong>siete u ocho </strong>dígitos con los del <strong>DNI del titular.</strong></p><p>Si el atacante tiene acceso a la factura, solo con leer el encabezado, obtiene el número de cliente y por ende, el de la contraseña y si no es así, hay muchas formas de conocer el DNI de la persona con solo saber el nombre y apellido de la víctima <em>(ej. </em><a href="https://www.dateas.com/es/consulta_cuit_cuil"><em>www.dateas.com</em></a><em>)</em>. Además, es de público conocimiento que los números de cliente de la empresa, para personas físicas, comienzan con 004, 014 y 044.</p><p>Con lo cual teniendo esa información, restaría probar <strong>tan solo 3 contraseñas </strong>para ingresar a la red privada de un domicilio.</p><pre>(004 ó 014 ó 044) + (DNI sin el ultimo digito ó DNI completo)</pre><h4>Ataques de fuerza bruta</h4><p>Aún si el atacante no conociera ninguno de los datos antes mencionados, existen otras maneras de generar contraseñas y así intentar vulnerar la conexión.</p><p>Estos ataques se llaman de <strong>Fuerza Bruta</strong> y consisten en generar contraseñas para luego probarlas en contra del objetivo hasta dar con la correcta.</p><p>El “problema” de este tipo de ataques es que generalmente el atacante no se sabe que caracteres la componen (si tiene mayúsculas, minúsculas, símbolos, o números) como tampoco conoce su longitud.</p><p>Este caso no es así, hay mucha información sobre la contraseña:</p><ul><li>Tiene <strong>10 </strong>caracteres.</li><li>Esos caracteres son <strong>solo números.</strong></li><li>Comienza con <strong>004</strong> ó <strong>014</strong> O <strong>044.</strong></li></ul><p>Con esta información se podría utilizar una herramienta que genere números del 0040000000 al 0440000000</p><p>Si bien se estarían generando un total de <strong>400 millones de opciones</strong>, y realmente pareciera ser un número muy grande, la realidad es que, aun exagerando los tiempos, un generador de fuerza bruta en una computadora puede llegar a probar <strong>17000 palabras por segundo </strong>en consecuencia, a lo sumo en <strong>7 horas </strong>la conexión de un domicilio puede ser vulnerada sin mucho conocimiento y esfuerzo.</p><h4>Optimizando el ataque</h4><p>Aún hay más información que un atacante puede tener en cuenta para agilizar el proceso.</p><p>El final de la contraseña <strong>son dígitos de un DNI. </strong>Con lo cual, se puede asumir que esos últimos 7 números están entre el <strong>2000000</strong> y el <strong>50000000.</strong></p><pre>#Arreglo con los posibles comienzos de la contraseña<br>st = [&#39;004&#39;,&#39;014&#39;,&#39;044&#39;]</pre><pre>for i in range(0, len(st)):<br> for x in range(2000000, 50000000):<br>  print (&#39;{0}{1:07d}&#39;.format(st[i], x))</pre><p>Y así, con ese <em>script, </em>en la misma computadora que llevaba 7 horas probar todas las posibilidades, <strong>en tan solo 23 segundos</strong> se generan todas las contraseñas posibles para una cuenta hogareña.</p><h3>FiberCorp no está exento</h3><p>La misma vulnerabilidad se encuentra en las conexiones de los autodenominados <em>líderes en servicios de telecomunicaciones para empresas.</em></p><p>La única pequeña diferencia es que para el caso de este tipo de conexiones, el inicio de la contraseña es <strong>010</strong> y los restantes dígitos son el comienzo de el número de CUIT de la empresa, por lo cual puede comenzar con:</p><ul><li><em>20, 23, 24 y 27 </em>para Personas Físicas</li><li><em>30, 33 y 34</em> para Empresas.</li></ul><p>Se podría hacer un generador de posibles CUIT o simplemente entender que es un número de 7 cifras que empieza desde el 2000000 y termina en 3400000.</p><pre>st = [&#39;010&#39;]</pre><pre>for i in range(0, len(st)):<br> for x in range(2000000, 3400000):<br>  print (&#39;{0}{1:07d}&#39;.format(st[i], x))</pre><p>Al tener menos posibilidades, el código anterior se ejecuta en<strong> 4 segundos</strong>, en otras palabras, si no se cambió la contraseña que generalmente queda por defecto la red de una empresa puede ser <strong>vulnerada</strong> muy fácilmente.</p><h4>Aclaraciones</h4><p>Si bien es cierto que para poder realizar cualquiera de la tareas referidas anteriormente hay una serie de pasos a seguir, y sin siquiera tener en cuenta que llevarlas a cabo es ilegal; el objetivo de este documento es alertar a los usuarios de alguno de estos servicios sobre el riesgo grave que existe si dejamos la contraseña que puede llegar a quedar por defecto luego de la instalación.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5ea5c9c503b5" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>