<?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 adam goldstein on Medium]]></title>
        <description><![CDATA[Stories by adam goldstein on Medium]]></description>
        <link>https://medium.com/@adamgoldstein?source=rss-f48139d69ce3------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*gibaPJbKhUmn2kQH1IXm5w@2x.jpeg</url>
            <title>Stories by adam goldstein on Medium</title>
            <link>https://medium.com/@adamgoldstein?source=rss-f48139d69ce3------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 08:30:30 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@adamgoldstein/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[Stoßlüften]]></title>
            <link>https://medium.com/@adamgoldstein/sto%C3%9Fl%C3%BCften-e59cba23f696?source=rss-f48139d69ce3------2</link>
            <guid isPermaLink="false">https://medium.com/p/e59cba23f696</guid>
            <category><![CDATA[decision-making]]></category>
            <category><![CDATA[neurodiversity]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[adam goldstein]]></dc:creator>
            <pubDate>Mon, 02 Feb 2026 16:05:47 GMT</pubDate>
            <atom:updated>2026-02-02T16:05:47.453Z</atom:updated>
            <content:encoded><![CDATA[<h3>A story about how a modern AI tool helped me move from vague curiosity to confident action in minutes, and why that matters for how different people think, plan, and execute in everyday life.</h3><p>The other day I scrolled past a throwaway post on social media about a German habit called Stoßlüften. The idea is simple: even in winter, you open windows wide, front to back, top to bottom, for five to ten minutes. You do not crack them. You do not leave them open all day. You just let the house breathe, then close everything back up.</p><p>The claim was that even ten minutes makes a measurable difference to indoor air quality.</p><p>Normally, that is where the story would end. I would bookmark it mentally, feel vaguely informed, and never actually try it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ga6n_h0yhltKTpCTSfgQxg.png" /></figure><p>Instead, fifteen minutes later, I had cleaner air and, more importantly, I understood why.</p><p>That gap between interesting idea and applied understanding is where large language models are quietly changing everyday, non-technical life.</p><h3>From folklore to experiment</h3><p>I have a long-standing interest in air quality and live in a house with a lot of variables monitored and controlled by air sensors, smart thermostats, purifiers, and humidifiers. My LLM has context on all of this through conversations over the years.</p><p>So, I asked about Stoßlüften.</p><p>Not is this real, but what is actually happening, and how would this apply to my house?</p><p>What I got back was not a generic explanation. It was a translation layer:</p><p>Why heat stays while air exchanges quickly. Why cross-ventilation matters more than duration. Why winter air can be cold but clean. What metrics would change first, and which ones might temporarily get worse.</p><p>This was not advice in the abstract. It was contextualized reasoning applied to a specific environment I had already described over time.</p><p>So I tried it.</p><p>I opened the front door, the back patio door, and an upstairs deck door. Ten to twelve minutes later, I closed everything and looked at my sensors.</p><p>CO₂ dropped across the house. Dramatically. Exactly as predicted.</p><p>Particulate matter spiked briefly from outside air rushing in, then fell back to baseline within ten minutes. Also expected.</p><p>VOCs behaved more strangely. They dropped upstairs and rose downstairs, likely due to turbulence, surface disturbance, and airflow paths that included the garage. Not alarming, but informative.</p><p>Instead of guessing whether this was good or bad, I knew what to watch:</p><p>How fast things decayed. Which floors behaved differently. Which changes were transient versus structural.</p><p>That is the difference.</p><blockquote>The key is not that the model knows the answer. It is that it can meet people where they are cognitively.</blockquote><h3>Executive function, but make it personal</h3><p>For people with strong executive function, this might not sound revolutionary. You read a thing, you synthesize it, you try it, you evaluate results. Done.</p><p>But executive function is not binary. It is a spectrum.</p><p>Some people struggle with translating theory into action, knowing which details matter, holding multiple variables in mind, deciding when something is safe enough to try, or understanding whether a result is signal or noise.</p><p>What LLMs do well, when used properly, is collapse those steps into a conversational loop.</p><p>I did not need a research paper, a building science background, a checklist written for a different house, or the confidence to wing it.</p><p>I needed an explanation that adapted to my constraints, a framing of risks before I encountered them, clear signals for this is normal versus this needs investigation, and permission to experiment without overcommitting.</p><p>That is executive function scaffolding delivered on demand in a form that works for the person asking.</p><p>Different people will use that scaffolding differently. Some want step by step instructions. Some want mental models. Some want reassurance. Some want optional depth. The key is not that the model knows the answer. It is that it can meet people where they are cognitively.</p><h3>This was not really about air</h3><p>Stoßlüften is just a convenient example. The real shift is this:</p><p>I went from a random social media post to a real-world intervention informed by physics, customized to my home, validated by sensors, and understood well enough to repeat safely, all in about fifteen minutes.</p><p>Not because the model told me what to do, but because it helped me understand why.</p><p>That is not automation. That is amplification.</p><p>When people talk about LLMs replacing thinking, they are usually imagining a worst-case version of outsourcing judgment. What is actually happening, at least in moments like this, is the opposite. Thinking becomes more accessible to more people, in more contexts, with fewer hidden prerequisites.</p><p>The house feels better today. The data makes sense. And I did not have to become an expert to get there.</p><p>That is the part that feels game changing. Not flashy. Not dramatic. Just quietly enabling better decisions in the middle of a normal day, especially for people who do not experience thinking, planning, or follow-through in the same way.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e59cba23f696" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Do PMs use AI more?]]></title>
            <link>https://medium.com/@adamgoldstein/do-pms-use-ai-more-ad5c9351f4e8?source=rss-f48139d69ce3------2</link>
            <guid isPermaLink="false">https://medium.com/p/ad5c9351f4e8</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[adam goldstein]]></dc:creator>
            <pubDate>Fri, 30 Jan 2026 16:38:43 GMT</pubDate>
            <atom:updated>2026-01-30T16:38:43.227Z</atom:updated>
            <content:encoded><![CDATA[<p>I’ve been pondering why product managers, especially those with engineering backgrounds, seem unusually energized by AI and agents.</p><p>As a PM, I rarely have formal power. I don’t tell people what to build. I can’t just assign tasks and expect compliance. Instead, the job is about getting people aligned around a problem. It’s about setting direction, defining constraints, creating shared understanding, and building enough conviction that teams want to move in the same direction.</p><p>Most of the real work happens upstream of execution. Problem definition. Narrowing scope. Surfacing blind spots. Talking to customers. Running experiments. Selling a direction to other teams so resourcing actually happens. None of that relies on delegation in the traditional sense.</p><p>What’s interesting is how closely that maps to working with agents. You don’t “manage” an agent the way you manage a person. You frame a problem. You clarify goals and constraints. You define what success means. Then you let the system explore solutions and bring signal back to you.</p><p>That’s already the PM ethos.</p><p>I think this is why PMs have often felt both influential and underpowered at the same time. We’re accountable for outcomes without direct control over inputs. In a human-only system, that’s stressful. In an agentic system, it suddenly becomes leverage.</p><p>Delegation finally exists, but it’s delegation to process rather than people.</p><p>The deeper shift here is that work stops being about roles and starts being about problems. Roles were a way to bundle expertise when humans were the limiting factor. Agents don’t need titles. They need well-formed intent and good feedback loops.</p><p>In that world, humans matter most where judgment, framing, values, and tradeoffs live. Not because we execute better, but because we decide what is worth doing in the first place.</p><p>Ironically, the role that was often seen as fuzzy or “just glue” turns out to be a preview of how humans and machines will collaborate. Not because PMs are special, but because the job evolved around ambiguity, influence without authority, and problem-first thinking long before the tools caught up.</p><p>This moment feels less like disruption and more like recognition.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ad5c9351f4e8" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>