<?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 Alireza Sheikholmolouki on Medium]]></title>
        <description><![CDATA[Stories by Alireza Sheikholmolouki on Medium]]></description>
        <link>https://medium.com/@alireza29675?source=rss-9b075e66421e------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*MzbJxdWb5EO5Ih--pbXvZw.jpeg</url>
            <title>Stories by Alireza Sheikholmolouki on Medium</title>
            <link>https://medium.com/@alireza29675?source=rss-9b075e66421e------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 22:54:56 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@alireza29675/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[Build to capture, not to last — Why I am building my custom personal AI agent anyway]]></title>
            <link>https://alireza29675.medium.com/build-to-capture-a05923532393?source=rss-9b075e66421e------2</link>
            <guid isPermaLink="false">https://medium.com/p/a05923532393</guid>
            <category><![CDATA[open-code]]></category>
            <category><![CDATA[openclaw]]></category>
            <category><![CDATA[claude-code]]></category>
            <category><![CDATA[personal-agent]]></category>
            <dc:creator><![CDATA[Alireza Sheikholmolouki]]></dc:creator>
            <pubDate>Tue, 10 Mar 2026 17:40:45 GMT</pubDate>
            <atom:updated>2026-03-10T17:40:45.388Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*D7PVNnsOEMuePAMQ" /><figcaption>Design by my talented human friend <a href="https://unsplash.com/photos/SfgcD0Ui06g">Milad Fakurian</a></figcaption></figure><p>I am spending serious time on a personal AI system I do not expect to last.</p><p>I am still building it anyway.</p><p>Every week it checks my workout logs and body measurements, then updates my next week’s gym plan in <a href="https://hevy.com/">Hevy</a>. Every morning it reads my calendar, <a href="https://todoist.com/">Todoist</a>, and our recent conversations and sends me a recap before I start the day. It has a teacher mode that explains things in the format I learn best, and when I give feedback, it updates that teaching flow so the next session is better.</p><p>This sounds overbuilt. It probably is. But I set up most of it by asking the agent to build the workflows for itself.</p><p>I also expect this exact setup to age quickly. The architectures will change. The tools will change. Even what we call an app will probably change. The current stack will look primitive soon. I still think building it now is a good investment, precisely because there is an opportunity underneath all that churn.</p><p>What I am actually trying to build is not the stack. It is a durable record of my context, my information, and my preferences.</p><p>If much stronger agents are coming, and I think they are, raw intelligence will not be the only bottleneck. The bigger question will be whether those systems have anything real to work with. Will they know what matters to me, how I make decisions, what I keep forgetting, how I learn, what good output looks like for me?</p><p>That is the layer I want to capture now.</p><h3>How this started</h3><p>I started from a simple place. I used Claude Code for coding.</p><p>Then a bigger possibility clicked for me. Instead of treating the agent as a tool for writing code, I could treat it as a layer between me and a repository of my own stuff. Mostly markdown files instead of code. Something a bit like <a href="https://obsidian.md/">Obsidian</a>, except I talk to the agent and it writes, organizes, and updates the notes for itself.</p><p>Those markdown files hold memory, notes, preferences, reusable instructions, and things I want done again.</p><p>So I built a private repo around that idea. It runs on my mini-pc. It is not open source. It is not a startup. It is not meant to be generic. I sometimes call it “Yad”, which means remembrance in Farsi, but really it is just a personal workspace the agent can read, update, and reason over.</p><p>The first version was messy. I mostly talked to the agent and let it write files. Then patterns started repeating. Some prompts kept solving the same kind of problem, so I turned them into skills and sub-agents.</p><p>Then I hit one of the obvious limits of current agents. They do not naturally wake themselves up and do something at 6 p.m. tomorrow or every Sunday morning. So I added a scheduler. That turned the system from a chat interface into something that could actually run recurring workflows. I have a stable solution for this now, and I will probably write about it separately.</p><p>That is when it became useful in a deeper way. Daily recap. Weekly gym plan updates. Background learning. Periodic security checks on my mini-pc. Teaching flows that improve over time. More importantly, it started leaving behind a detailed trail of how I live and what works for me.</p><p>I talk to it, it takes notes, I get better reasoning from it, so I talk to it more.</p><h3>Build to capture</h3><p>There are many valid directions in personal-agent work right now.</p><p>Some people want to push the state of the art. Some want to build lasting products or infrastructure. Some want to make open source systems that can survive through community effort. All of that makes sense.</p><p>I am after a different payoff.</p><p>For me, the more durable question is this: can I extract truthful information from myself and make it usable by future systems?</p><p>That includes obvious things like my calendar, tasks, notes, and recurring workflows. But it also includes softer things that are much harder to reconstruct later:</p><ul><li>how I like to learn</li><li>how much detail I want in explanations</li><li>what kind of reminders help me and which ones annoy me</li><li>what I tend to forget</li><li>how I make tradeoffs</li><li>what “good enough” means to me in different parts of life</li></ul><p>Those are not product features. No one can ship them for me. They are personal operating data.</p><p>This is why I am comfortable building on a temporary stack. The temporary part is the scaffold. The valuable part is the extraction process.</p><p>Every time I teach the system something true about me, I get value twice. I get better help now, and I create better raw material for whatever stronger system comes next.</p><p>That is also why I am not optimizing this to be generic. I want it to become more specific to me over time. The more personal it gets, the less replaceable its context becomes.</p><p>If a much better agent arrives later, I do not want to start from a blank chat window and a vague prompt. I want to hand it a well-structured record of my preferences, decisions, relationships, routines, and feedback loops. I want to know what to ask, because I have already spent time learning what matters.</p><h3>A few temporary layers that helped me</h3><p>While building this, I ran into a few real gaps. The solutions I added are simple, but they might give you ideas.</p><p><strong>Scheduler.</strong> This gives the system a clock. That one addition changed a lot for me. It is how I run recurring workflows like daily recaps and weekly gym-plan updates. I have a stable setup for this now and will probably share it in a separate post.</p><p><strong>/learn.</strong> This is not about teaching me something in the moment. It tells the agent to go do deep background research on a topic, store what it learned in its own memory files, and come back stronger for future conversations. I have found this surprisingly useful, and I do not see this pattern discussed enough.</p><p><strong>/interview-me.</strong> This helps the system pull context out of me instead of waiting for me to write it down perfectly. It keeps asking until vague thoughts turn into explicit preferences, memories, or workflows. It often feels like giving my second brain a way to ask follow-up questions.</p><p><strong>/teach.</strong> This one teaches me. I can learn something, react to the format, and feed that back into the system so the next session fits me better.</p><p>None of these are grand inventions. They are basic but effective adapters between my life and today’s agents. Their job is to help the system grasp what I think and produce artifacts: notes, rules, preferences, examples, repeated judgments. That output survives even if the adapter does not.</p><p>I also have many more custom skills than the ones above. Most are too personal or too narrow to be useful as general advice, but I will probably share more of them over time.</p><h3>More modalities, more capture</h3><p>The way I think about the next step is simple: I want better input.</p><p>Right now most of this system depends on typed conversation, transcribed voice through <a href="https://wisprflow.ai/">Wispr</a>, structured data, and scheduled jobs. That is useful, but incomplete. A lot of real life never makes it into a chat window.</p><p>So I want to add more modalities and let the system participate in more of my day. Speech-to-speech is the newest one. I added it a couple of days ago because I wanted to talk to this agent naturally and capture more thoughts with less friction.</p><p>Speech-to-speech models are fast but not especially smart. So I paired that fast response layer with slower agents in the background that can reason better and gather context without breaking the flow of conversation. That architecture is probably worth a separate post too.</p><p>That is the direction I care about most. A system that helps me extract more correct information from myself, from my routines, and from the environments I move through. A system that produces better digital artifacts of who I am and how I work.</p><p>I am not trying to make it generic enough to become a trend. I want it to become more grounded in my actual life, not less.</p><p>Technology is changing fast. My preferences are changing much more slowly.</p><h3>Closing</h3><p>I do not expect this exact personal agent stack to survive.</p><p>I do expect the context it captures to survive.</p><p>To me, owning that context in digital form is extremely valuable.</p><p>If something close to superintelligence arrives in the next few years, it will not magically know you. It will still need context. It will still need a record of what matters to you. It will still need examples of what you mean by good.</p><p>And if you already have that, the benefit starts immediately, long before the next leap. Your current tools already get better.</p><p>When much better systems arrive, I want to meet them with years of captured context, tested preferences, and examples of what good looks like for me.</p><p>So the practical move today, at least for me, is not to worship today’s shiny tools or wait for the perfect product. It is to start externalizing what matters. Capture your context. Capture your preferences. Capture the feedback loops that make an output feel right or wrong to you.</p><p><strong>Build to capture, not to last.</strong></p><p>If you are building something similar, I would love to hear your approach. What are you capturing now that your next stack should inherit?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a05923532393" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[From Overwhelming to Organized — Reading Long Slack Threads with ChatGPT ]]></title>
            <link>https://alireza29675.medium.com/from-overwhelming-to-organized-reading-long-slack-threads-with-chatgpt-92a34307a9b4?source=rss-9b075e66421e------2</link>
            <guid isPermaLink="false">https://medium.com/p/92a34307a9b4</guid>
            <category><![CDATA[ai-prompts]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[slack]]></category>
            <dc:creator><![CDATA[Alireza Sheikholmolouki]]></dc:creator>
            <pubDate>Sun, 19 Feb 2023 12:52:14 GMT</pubDate>
            <atom:updated>2023-02-24T21:05:55.445Z</atom:updated>
            <content:encoded><![CDATA[<h3>From Overwhelming to Organized — Reading Long Slack Threads with ChatGPT 🤖</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TxOBMBQJSz6wCTl6YamVqA.png" /><figcaption>The worst selfie that I took in Barcelona — Park Güell, Barcelona, Spain</figcaption></figure><p>After returning from a long vacation, I found myself facing the overwhelming task of catching up on dozens of lengthy, +50 replies Slack threads, which I had to follow to know what was going on in the team and the company. I knew that going through all of the messages would be time-consuming and overwhelming. Am I going to do it by myself one by one in 2023? Hell to the no!</p><blockquote><strong>Important Note:</strong> Before sharing any information with ChatGPT, review the content for any sensitive or confidential information. This includes but is not limited to personal information, financial data, trade secrets, or any other private information.</blockquote><blockquote>Please exercise caution and responsibility when sharing information with any AI language model.</blockquote><p>That’s when I decided to turn to <a href="https://chat.openai.com/">ChatGPT</a>, a powerful AI tool that can summarize entire conversations while retaining important information. I decided to prepare a GPT prompt instead, copy/paste all the conversations there and read the summarised unified version.</p><p>Here’s the prompt that I made:</p><pre>ChatGPT, I’m going to send you a raw copy of a Slack thread, they are going to be pasted in the following format:<br><br>”””<br>[Username]<br>[Date]<br>[Message]<br>[Emoji Reaction 1]<br>[count of reaction 1]<br>[Emoji Reaction 2]<br>[count of reaction 2]<br>…<br><br>[Username]<br>[Date]<br>[Message]<br>[Emoji Reaction 1]<br>[count of reaction 1]<br>[Emoji Reaction 2]<br>[count of reaction 2]<br>…<br>”””<br><br>I need you to provide me with a text, with the following format:<br><br>”””<br><br>Subject: [Subject of the conversation]<br><br>Summary: [A brief summary of the main points discussed in the conversation]<br><br>Highlights: [Highlight key takeaways from the conversation in the summary section]<br><br>Opinions:<br><br>- [Name of Person 1]: One-liner summarizing their opinion or contribution to the conversation<br>- [Name of Person 2]: One-liner summarizing their opinion or contribution to the conversation<br>- ...<br><br>“””<br><br>Ensure that the one-liners accurately capture each person&#39;s idea or contribution to the conversation.<br><br>Here is the pasted Slack conversation:<br><br>&quot;&quot;&quot;<br><strong>&lt;PASTE YOUR SLACK THREAD HERE&gt;<br></strong>&quot;&quot;&quot;</pre><p>Here is one of the very important results I could get from a discussion among company members about carpet bugs (names are changed):</p><pre><strong>Subject:</strong> Black Carpet Beetles Infestation<br><br><strong>Summary:</strong> Christa discovered a couple of Black Carpet Beetle larvae in their apartment and this has caused concern about a potential infestation. They plan to clean, freeze and go through all the closets as a precautionary measure. They believe that the beetles may have come from another apartment since they have not been on vacation.<br><br><strong>Highlights:</strong><br><br>- The larvae grow to 7mm in length and can cause significant damage to natural fibres.<br>- The group discusses potential solutions to get rid of the infestation, ranging from burning the entire apartment complex to getting an exterminator.<br>- They express mixed feelings about the severity of the problem and whether or not it can be resolved.<br><br><strong>Opinions:</strong><br><br>- Viima: &quot;Just burn the entire apartment complex.&quot;<br>- Antti: &quot;Best I can do for now is a lighter. This sucks.&quot;<br>- Mikko: &quot;Just let them eat some carpets or useless t-shirts. Kidding though :stuck_out_tongue:&quot;<br>- Kaija: &quot;Some exterminator will come and throw all your stuff away and kill all of those bastards. Hopefully, insurances are good.&quot;<br>- Christa and the others in the group did not provide any additional opinions or contributions to the conversation.</pre><p>I decided to avoid extras and just share the prompt. If you’ve had any similar experiences using ChatGPT or GPT API, I’d love to hear about them, and any prompts that you found useful. Additionally, if you have any suggestions for optimizing this prompt, please don’t hesitate to share them. Let’s keep the conversation going! 🙌</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*_TqTdjEaJPC4sWR-" /><figcaption>Photo by <a href="https://unsplash.com/@skraidantisdrambliukas?utm_source=medium&amp;utm_medium=referral">Gertrūda Valasevičiūtė</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=92a34307a9b4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The easy way to install Homebrew on your M1 Macbook]]></title>
            <link>https://alireza29675.medium.com/the-easy-way-to-install-homebrew-on-your-m1-macbook-884fcde172f2?source=rss-9b075e66421e------2</link>
            <guid isPermaLink="false">https://medium.com/p/884fcde172f2</guid>
            <category><![CDATA[m1]]></category>
            <category><![CDATA[macbook-m1]]></category>
            <category><![CDATA[macos]]></category>
            <category><![CDATA[homebrew]]></category>
            <category><![CDATA[m1-chip]]></category>
            <dc:creator><![CDATA[Alireza Sheikholmolouki]]></dc:creator>
            <pubDate>Mon, 26 Jul 2021 15:20:26 GMT</pubDate>
            <atom:updated>2021-07-26T15:20:26.491Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WC7nlwMJD4D5X9rkUx1XbQ.jpeg" /></figure><p>Let’s skip the introduction and keep the solution short and quick. This is probably the easiest solution until Homebrew releases their M1 compatible version.</p><p>Install <a href="https://brew.sh/">Homebrew</a> using the following command:</p><pre>arch -x86_64 /bin/bash -c &quot;$(curl -fsSL <a href="https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh">https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh</a>)&quot;</pre><p>To use Homebrew as you used to do, run the command below (or <a href="https://askubuntu.com/questions/758496/how-to-make-a-permanent-alias-in-oh-my-zsh">add it in your .zshrc config</a>)</p><pre>alias brew=&quot;arch -x86_64 brew&quot;</pre><p>Enjoy.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=884fcde172f2" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>