Future-Proof Your Career with AI Newsletter Strategy

View Email

Jensen makes the case for open source AI

khemaridh@substack.com7/26/2026

Substack

View Email

AI is coming for your Slack workspace

khemaridh@substack.com7/23/2026

Substack

View Email

Using ChatGPT 5.6 for knowledge work

khemaridh@substack.com7/19/2026

Substack

View Email

Will Claude Cowork get replaced (by ChatGPT)?

khemaridh@substack.com7/16/2026

Substack

View Email

Chat is dead. Agents are the new default.

khemaridh@substack.com7/12/2026

Substack

View this post on the web at https://khemaridh.substack.com/p/is-fable-worth-the-hype I have a confession. I was pretty excited when the US Government banned Anthropic’s Fable model. Not because I am anti-AI. Nor because I believe in the heavy hand of government. It’s because I felt behind. I didn’t think I had the right projects and “asks” for such a powerful model. And that I’d be wasting my precious Fable credits on some pedestrian tasks. As Ethan Mollick states, I was nervous that I wasn’t ambitious enough. So I took the plunge and burned those tokens. Here’s what I learned. Claude Cowork just launched a BIG update…. And your agents can now run from the cloud — while your laptop is closed! Now, imagine sitting down Monday morning and your meetings are already prepped, your inbox is triaged, and last week’s notes are organized — all before you open your laptop. That’s Claude Cowork running on a schedule you set once. This self-paced course teaches you to build that system. Finish it in a weekend. What makes Fable different Every new model (especially the ones that represent a step-change, like Opus) requires new prompting approaches. Here are some tips from Anthropic on guiding Fable [ https://substack.com/redirect/2e38d2f1-7086-41f6-8c9e-8d8847a2fc14?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] (curated for knowledge workers): Give the reason, not only the request. I’m working on [the larger task] for [who it’s for]. They need [what the output enables]. With that in mind: [request]. Consider all effort levels. Effort is the primary control for the trade-off between intelligence, latency, and cost on Claude Fable 5. Use high as the default for most tasks, with xhigh for the most capability-sensitive workloads and medium or low for routine work. Lower effort settings on Claude Fable 5 still perform well and often exceed xhigh performance on prior models. First-shot correctness on complex, well-specified problems. Early testers reported single-pass implementations of systems that previously took days of iteration. Enterprise workflows. Claude Fable 5 follows instructions, stays in scope, and produces professional-grade output on financial analysis, spreadsheets, slides, and documents. Fable test #1: Massive Context dump I have a folder with over 250 Zoom transcripts covering AI trainings, sales calls, and networking. It also has every presentation, proposal, syllabus, and product page I’ve ever written. Here’s the prompt (via voice dump): Go through both of these repos, particularly the wikis, which have training materials for my AI consulting training business, and then contacts, which has all of the sales calls, so a lot of the marketing and prospect information related to the business. training-wiki and contacts-wiki. Some of the calls have not been indexed. Use your judgment as to whether it makes sense. I want you to do three things. 1. Create a beautifully designed website that appeals to financial services buyers that is focused on conversions and getting people to submit a request for a consultation. Use your judgment on what to include on the website to help it convert. Create a comprehensive and ordered curriculum for this audience that is effectively a menu of what people can choose from. Create that as a dynamic HTML that would live on the website, but it would be behind some kind of access control. We don't need to incorporate the access control now. It could just be a lazy password box. Then I want you to create a slide deck that covers the entire training material for a three-hour session. This would also live on the website behind some paywall. Maybe there's a client side and a public side to the website. Last, create a strategy document in Markdown that helps me grow the business to a $1,000,000-a-year business without hiring any employees. Basically I said “read all this stuff and give me a new website, deck and strategy for making a million dollars a year.” It ran for 30 minutes and here’s what it produced. The website and decks were NICE (especially considering this was one prompt). However, the strategy document was even more impressive. I won’t share the specifics (as it contains a lot of confidential information) but there’s been one part of my business I’ve been sleeping on — creating more B2B training materials. It’s in part because I haven’t had the time, but also due to a bit of imposter syndrome about going full-bore into the EdTech space. I haven’t really talked much about this (even to prospects) but Fable quickly excavated 3 critical conversations and showed me very bluntly what I was missing: This level of judgment — finding a very valuable needle in a haystack — has never been demonstrated by other models. Fable test #2: The one-shot experiment Next up, I wanted to customize math problems for my kids during the summer. I wanted the app to be interactive and something I could easily monitor (especially since we begin our travels soon). Here’s the prompt: I want to create disposable math problems on a website for a fourth grader. I want to map it to the common core curriculum: there should be the different categories of the things that you are expected to know. You can go into each section and tell them how many problems you want to work on. You could do 5 problems on fractions or 20 problems on division or 5 on multi-step. For now, let's have three levels of difficulty, and you can also just ask for a test. Again, where you specify the number of problems and then you specify the level of difficulty. For this first version, there's no saving. You just run it on a one-off basis and get your score. We probably need a data structure for the curriculum and a separate one for the spine. For each question, I don't know exactly how that would work. We need a method to make sure that the answers are actually correct. There should be two hint buttons for each question. The first we provide a visual hint using some kind of visual engine, like Three.js, and then the second one would be just a text-based hint. When you set up your questions, you would just state whether you are using hints or not. The design should be very distraction-free, aimed at a kid who just needs to focus on this. There should be some kind of backend so that me, the creator, understands what is happening, for this version, I don't want any history to be collected. The next step (we're not going to do this yet) would be to expand this to 7th grade. The results were also impressive (with the caveat that I didn’t run any significant tests): Here’s the admin view: Admittedly, I didn’t have the time to go deep here. But at first blush this is pretty cool. Fable test #3: The YOLO exercise I’ve been building out a lot of wikis [ https://substack.com/redirect/d056826d-8b92-4d51-aa3e-a367d9e6057d?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] to help my Cowork and Claude Code agents navigate large amounts of context. I have a wiki for my training materials and one for my phone calls and have wanted to create one for all the content I consume. Since I use Readwise [ https://substack.com/redirect/5f9dd933-49bb-4d22-8346-ea003c4d0492?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] to capture all my podcasts, tweets, blog posts and YouTube videos the data was easily accessible. But converting this high-volume data stream into actionable insights that could be used in my trainings, talks, articles and marketing felt like a heavy lift. So I asked Fable, with the following prompt: I want to create a knowledge graph that takes everything I read related to AI and stitches the information together in a way that is queryable for: - blog posts - blog ideas - curriculums - talks - syllabus for my AI consulting business I want the main engine of intake to be the Readwise MCP, where most of the integrations will be linked. From there, I don't do much tagging, but I want that part to be automated. I want the knowledge graph to track concepts. A concept could be like an agent skill, or a concept could be codecs. I also recognize that a lot of these ideas go stale with time, so there should be some time component. For podcasts, I would want to get the raw transcript and build that in. I usually can get that through YouTube, so this definitely needs some kind of YouTube transcriber tool. I plan on running this all through Claude Code as the harness. I would like some human-in-the-loop processes, maybe not for all the articles, but once there are five articles or once a week, I can guide the model accordingly. You should note that there are other wikis, one for training and one for networking, that sit in separate repos. I don't see any integrations at first, but I want to keep them in separate repos, so don't do anything across them yet. Additionally, it built a YouTube-specific scraper and a tool that ingests Twitter bookmarks (including long-form articles directly). Here’s a list of the concepts it processed and classified: It’s now working well and updating every couple days. I’ve done some light querying on the quality of its “intelligence” but it looks promising. Fable test #4: Building a productized service For my final test, I wanted something that could instantly improve my business and my ability to deliver more value to my clients. And if I’m honest, the harder part of this exercise was identifying the highest impact activity (versus Fable executing on it). Here’s the prompt I used: I want to create a skill identifier for financial services professionals. I want it to interview them with some criteria, like: - What are they in private equity financial services? - What is their role? - Investing, middle office, or fund accounting? I want to give them the option to answer some more questions, so they could always say, "I'm done answering questions." First, it's: - What is your job function? - What industry do you cover? - What are your core activities? Maybe we don't include something for data, but I'll bring this tool into a meeting with a client. I might have them answer a few of the questions beforehand, and then, using Claude Code as a harness, it will populate a list of skills that they should develop. The list of skills I want them to be pre-populated from some of these websites, including the ones below, and then we could probably build up a list of skills as well in advance, or maybe in real time, in response to the prompts. That's what I'm not sure about. The goal is for me to use this as a helper tool while I'm sitting at their desk with Claude Code open, but over time I want to give it to them so they can do it in advance. It will probably evolve to some web-based format. It would be nice for me to have some kind of visual way of showing this, since I will be screen sharing it, and the terminal makes it look less professional. There is an element of salesmanship involved in this as well. I'd like to build a prototype. I'm not sure if we need to store the skills in a database of Markdown files or a series of Markdown files. That will probably be okay. I also have some other use cases that I can import in that could serve as inspiration. https://github.com/anthropics/financial-services/tree/main https://github.com/latour-ai/skills https://github.com/daloopa/daloopa-plugin-claude/tree/main/skills Once again, I’m still in testing mode (it turns out that human judgment is the ultimate bottleneck). But here’s the prototype: And my early read on the outputs is that it’s pretty good. Does Fable pass the hype test? To be honest, it’s too early to tell. While the first attempts were impressive, I did not have enough bandwidth to assess all of the outputs. And more importantly, these tokens were heavily subsidized — so assessing the ROI is even trickier. Future-Proof Your Career with AI is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber. Unsubscribe https://substack.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.XPvHQ6D2nABCjm9xhJlEyBs3Fap-dBLK8eU56yl6Mxg?

View Email

Is Fable worth the hype?

khemaridh@substack.com7/9/2026

Substack

View Email

How to get the most out of Fable 5

khemaridh@substack.com7/5/2026

Substack

View this post on the web at https://khemaridh.substack.com/p/meet-my-new-ai-cfo The second quarter just ended, which means it’s time to get my house in order. Specifically, my financial house in order. So that I can evaluate our business results. This is a messy process where the data is scattered across my bank, Stripe, business credit card, and PEO (i.e. Gusto). Plus it requires a bunch of manual adjustments. Yes, I have a bookkeeper (and they’re awesome), but it would be nice to incorporate this analysis natively into my Business Operating System — specifically with Claude Cowork. Turns out I’m able to do this quite easily with an Agent Skill and Living Artifacts. 1. Creating a general ledger For a minimum viable product I needed the data from my bank and credit card statements. While I could have used the Claude Chrome Extension, [ https://substack.com/redirect/5b8ee22d-b97f-4aac-bdba-8739e2a3f100?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] I’m hesitant to point it at any financial websites. So I just did a manual download of 6 months of credit card statements and one export of all cash movements. Build YOUR own AI CFO using Claude Cowork I mean, seriously — it doesn’t get more magical than this. Claude Cowork can proactively brief you for all your meetings, triage your inbox while you sleep and update your CRM with all of the week’s interactions. Set it and forget it — and the results get automatically pushed to you. This self-paced course teaches you to build that system. Finish it in a weekend. 2. Creating an Excel file as the “source of truth” Claude Cowork then took the 6 credit card files and created a tab called “Expenses.” As part of the prompt, I had it categorize each transaction as Software, Travel, Food and Beverage. Next, I did the same with the raw bank export. This was a bit trickier because the descriptions are not always as clean. There is also a reconciliation step since some cash outflows can’t be double-counted. For example, one bank outflow maps to a monthly credit card payment. 3. Creating a Live Artifact in Claude Claude recently released its live artifact feature for Cowork. [ https://substack.com/redirect/7db12c1d-b3ba-4e04-902c-40bcf9dcd23c?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] Here’s the official release note: A live artifact is a persistent, interactive HTML page that Claude creates for you in Cowork — a tracker, a dashboard, a comparison tool, a reference—shaped around your specific work. Every live artifact you create is saved to the “Live artifacts” tab in Cowork, and you can reopen, refresh, and iterate on it from any future session. What’s nice about this feature is that your “back-end” (in this case, my Excel file) refreshes the artifact automatically. Said differently, when you update the spreadsheet — the dashboard updates. This removes the friction of setting up and maintaining a complex data source. The prompt for the initial artifact was ridiculously simple, the key was including the words “Live Artifact”: Create an Excel sheet that appends all the transactions into a "ledger" from the first file (January). This is for my biz spending via a credit card. I want to do analyses for different quarters, perdios, trends, etc andthen turn it into a live artifact Here was the first result: The next task was more ambitious and involved creating the actual financials. The prompt was also very straightforward: don't act, but is there a way to merge the credit card transactions wiht the bank ones to make a more comprhensive picture? It took <5 iterations to get right, but here was the result: You’ll note the little refresh button in the top right — this pulls the latest data from Excel if you need to update the dashboard. With the raw data in place, I could also create little “widgets” to help me with my analyses. For example, this was a widget that helped me track recurring tech subscriptions: 4. A reconciliation skill Now that the dashboard is running, I’ll run the next update at the end of Q3. But I wanted to create a skill that captures some of the nuance of this reconciliation process. This includes: One-off payments that should be excluded Merging clients who pay using different payment methods Creating sub-categories for Stripe (courses [ https://substack.com/redirect/76677a51-970d-496c-893d-2b4ab34a059f?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] versus Substack) This is easily achieved using Anthropic’s skill creator. [ https://substack.com/redirect/341e7213-021b-439c-86da-acf08741f1bf?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] Here’s the output: Your turn This was such a delightful experience, particularly just working natively in a single Excel file. I’m currently replicating it on our personal card — this time, the skill will draft an email for my wife flagging any outliers and subscriptions that might be worth cancelling. Give it a try — this is a process that everyone needs to be doing. And AI makes it painless. Join me for a free conversation on implementing AI inside your organization Unsubscribe https://substack.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.PVy0q8qphuJ32EGlEXlroAfF0aoY04lriZURdvOsmZ0?

View Email

Meet my new AI CFO

khemaridh@substack.com7/2/2026

Substack

View Email

The rebel alliance

khemaridh@substack.com6/28/2026

Substack

View Email

I built a self-improving "company brain"

khemaridh@substack.com6/25/2026

Substack

View Email

Codex takes control of your computer

khemaridh@substack.com6/21/2026

Substack

View Email

The power user's guide to Claude skills

khemaridh@substack.com6/18/2026

Substack

View Email

The AI Shift in Investing

khemaridh@substack.com6/14/2026

Substack

View Email

I built a "headless CRM" in Codex

khemaridh@substack.com6/11/2026

Substack

View Email

Your AI budget is growing. Your returns aren’t.

khemaridh@substack.com6/7/2026

Substack

View Email

11 ready-to-use AI skills to help you work faster

khemaridh@substack.com6/4/2026

Substack

View Email

This agent handles iMessage, Gmail and WhatsApp

khemaridh@substack.com5/31/2026

Substack

View Email

Here's how AI helps me read more

khemaridh@substack.com5/28/2026

Substack

View this post on the web at https://khemaridh.substack.com/p/lessons-in-founder-mode-and-ai Greetings from New Haven! I’m celebrating my 25th college reunion and no one can escape the AI discussion. We’ve had a few late night conversations on: Discoverability of Zoom transcripts Are private equity portcos actually implementing AI Integrating AI into undergrad curriculums Rebuilding your company’s data for AI accessibility If you’re new here, every Sunday I curate my favorite reads/podcasts/tutorials (so that you don’t have to). Let’s jump in. 1. How Brian Chesky Is Redesigning Airbnb for the AI Era Airbnb is trying to turn itself into an AI-first company. For CEO Brian Chesky this involves lean teams, abandoning “people managers” and building a robust context framework around all your data. → Watch the video [ https://substack.com/redirect/36b9472f-aa14-4109-a1c1-b3d5e47b9eea?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] 2. How I Choose Which Cloudflare Employees to Replace With AI Despite record revenue growth, Cloudflare laid off 20% of its workforce. CEO Matthew Prince explains who is safe: builders and sellers. Who’s not safe? Employees whose role involves measuring things. → Read the article [ https://substack.com/redirect/64d9767d-4727-4010-b9ed-b65de8884c12?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] 3. Are AI Data Centers Stealing All Your Water? Does ChatGPT drink a bottle of water every time you ask it a question? Who suffers when a new data center is built? These are thorny questions that are usually discussed via narrative and not data. This piece takes a balanced (but opinionated) view and clearly lays out the issues. → Read the post [ https://substack.com/redirect/134df3e5-4709-4cb0-a50d-0989ff7eb515?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] Stop Chatting With AI. Start Building With It. Imagine sitting down Monday morning and your meetings are already prepped, your inbox is triaged, and last week’s notes are organized — all before you open your laptop. That’s Claude Cowork running on a schedule you set once. This self-paced course teaches you to build that system. Finish it in a weekend. The Best of the Rest Is Notion AI ready for prime time? [ https://substack.com/redirect/f91e8170-c1c6-4d02-a7e4-fcf28fbb48d8?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] — I’ve been re-introducing this cloud-based “everything app” into my workflows The case against the AI jobs apocalypse [ https://substack.com/redirect/a8005c79-b669-410d-aaa2-6c4e8006fd4c?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] — Dig into Jevons Paradox, the “lump of labor fallacy” and human scarcity. 7 ways to “Codex-maxx” [ https://substack.com/redirect/a1e61093-ca06-4431-932e-3e3c197d8be8?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] — Using Codex as an “operating system for knowledge work.” Download our free guide to Claude Cowork Unsubscribe https://substack.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.mhU4Mbjmgo8NJwt2Ma4uUXUpMz9RpIycMR6mEqsdwZQ?

View Email

Lessons in "Founder Mode" and AI

khemaridh@substack.com5/24/2026

Substack

View Email

Is Notion AI ready for prime time?

khemaridh@substack.com5/21/2026

Substack

View Email

How to design agent skills (like a pro)

khemaridh@substack.com5/17/2026

Substack

View Email

AI now integrates seamlessly with Excel

khemaridh@substack.com5/14/2026

Substack

View Email

How to use the new Codex Super-App

khemaridh@substack.com5/10/2026

Substack

View this post on the web at https://khemaridh.substack.com/p/ai-is-coming-for-finance As the market rides all-time highs, we can turn back the clock to Q1 and the depths of the SaaSpocalypse. With a series of open source plug-ins, Anthropic put the LegalTech industry on blast. Well, they’re at it again. This time setting their sights on the Financial Services industry. (Not to mention, earlier in the week Blackstone and Goldman created a joint venture with Anthropic [ https://substack.com/redirect/92b05035-45e6-4f5d-9b1f-5b83f29db7d6?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] to go after LaTour AI. [ https://substack.com/redirect/8e1671df-9a58-4800-9398-5e087e979231?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ]) And in response to the news, while the broader market was up ~1% according to Bloomberg: “FactSet shares fell as much 8.1% after the announcement while Morningstar Inc. erased earlier gains to fall more than 3%.” So are these new plug-ins going to wipe out analyst jobs in the coming months? Introducing “Agents for financial services” On Tuesday, Anthropic released “ten ready-to-run agent templates for the most time-consuming work in financial services.” Here’s how the official release [ https://substack.com/redirect/e134b43f-3e62-4bbc-96d3-12904910d561?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] describes these agent templates: Each agent template is a reference architecture that packages three things: skills (instructions and domain knowledge for the task), connectors (governed access to the data the task runs on), and subagents (additional Claude models that are called upon by the main agent, for specific sub-tasks such as comparables selection or methodology checks). So to recap, each template is: skill(s) files connector access sub-agents And here are the 5 of the templates they released for research and client coverage: Pitch builder creates target lists, runs comparables, and drafts pitchbooks for client meetings; Meeting preparer assembles client and counterparty briefs ahead of calls; Earnings reviewer reads transcripts and filings, updates models, and flags thesis-relevant changes; Model builder creates and maintains financial models from filings, data feeds, and analyst inputs; Market researcher tracks sector and issuer developments, synthesizes news, filings, and broker research, and flags items for credit and risk review. There are two ways to run each of these templates: as a plug-in for Claude Code and Cowork and a cookbook for Claude Managed Agents. Agent 1: Meeting Prep Now Anthropic isn’t doing any gatekeeping here, all these plug-ins are open source. You can go inspect the contents over at their financial services GitHub repo. [ https://substack.com/redirect/aa0770cf-0117-4a14-a9ae-a70ec9bacb6d?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] Are you able to find the plug-ins? Do you find GitHub scary? As a non-coder, I did at first. But it turns out you need to understand this basic technical literacy if you want to build, manage and customize agents. We’ve taught thousands of non-coders how to develop these critical AI skills in Unfair Advantage — Master Claude Code (in a week). Here you’ll find meeting-prep-agent: When you click into meeting-prep-agent you’ll have the option of viewing the skills and the agents. As a refresher, skills are simple markdown files [ https://substack.com/redirect/0f569e86-9591-44e5-bd6a-21d33f25202d?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] (i.e. text). This agent has 4 skills: I looked at the Client Review skill [ https://substack.com/redirect/bfe393bf-d455-446a-8b99-3a7fb51bebdf?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] and here’s the first half of the file: It’s a 400 word text file. Yes, that’s it. Now it interacts with the agents folder, which includes a 220 word agents.md file: A few of the key lines: Pull the relationship information from CRM MCP Pull context from CapIQ MCP Finally, a simple JSON file enables you to access this plug-in from Claude Cowork or Claude Code. Installing the plug-in The Anthropic repo provides the installation instructions. Claude Cowork To set it up in Claude Cowork: Paste this repo URL — https://github.com/anthropics/claude-for-financial-services — then pick the agents and verticals you want from the marketplace list, or Upload a zip — zip any directory under plugins/ (e.g. plugins/agent-plugins/pitch-agent/) and drop it in. Claude Code # Add the marketplace claude plugin marketplace add anthropics/claude-for-financial-services # Core skills + connectors (install first) claude plugin install financial-analysis@claude-for-financial-services # Named agents — pick the ones you want claude plugin install pitch-agent@claude-for-financial-services claude plugin install gl-reconciler@claude-for-financial-services claude plugin install market-researcher@claude-for-financial-services # Vertical skill bundles claude plugin install investment-banking@claude-for-financial-services claude plugin install equity-research@claude-for-financial-services Once installed, the agents will start operating in the background and can fire from slash commands. Per Anthropic: Once installed, agents appear in Cowork dispatch, skills fire automatically when relevant, and slash commands are available in your session (/comps, /dcf, /earnings, /ic-memo, …). Your turn to try it As you can see, it’s text files all the way down. My early read on these files is that they can activate newbies — but are a far cry from codifying your true domain expertise. But that makes sense. Even in an industry as developed as financial services, activation is still the missing ingredient. Download our free guide to Claude Cowork Unsubscribe https://substack.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.UAvuhetB874AZmMlqdJOMiP7YctomyOFKZvGoEKH_K4?

View Email

AI is coming for finance

khemaridh@substack.com5/7/2026

Substack

View Email

AI is coming for your mind

khemaridh@substack.com5/3/2026

Substack

View this post on the web at https://khemaridh.substack.com/p/i-tried-chatgpts-new-image-model I’m a words guy, not a meme guy. I like Twitter and hate Instagram. And much of my interest in AI comes from its impact on knowledge work. So when a new image model comes out, I usually give it a shrug. Yes, the AI images of 50 Cent trolling Diddy [ https://substack.com/redirect/507cd485-d61c-45e2-8f5a-0ce44acdb5c0?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] will catch my attention — but that’s about it. But ChatGPT’s Images 2.0 model [ https://substack.com/redirect/eb232604-63ea-4d16-8554-c852a8edc36b?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] got such rave reviews that I had to give it a spin. What makes ChatGPT Images 2.0 so unique? We all remember the last time OpenAI released an image model and we were inundated with Studio Ghibli memes. But it was still janky AF. Whether it was misplaced text, weird finger placements or just that strange AI veneer — we hadn’t crossed the uncanny valley. Google’s Nano Banana model came along and delivered images that were definitely more usable. For me, well-designed text made a huge difference. Like this image for our Claude Cowork guide: So what’s the deal with this new model? The AI Daily Brief Podcast [ https://substack.com/redirect/0f87626b-455f-4460-83e9-b62172d2b675?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] provided a great round-up: Extreme precision It excels at detailed instruction following, including the rendering of dense, small text, tiny UI elements, and complex iconography. Reasoning capabilities Unlike previous models, this is a reasoning model that can search the web for real-time information, generate multiple distinct images from a single prompt, and double-check its own outputs to ensure accuracy. Real-world intelligence The model demonstrates a strong grasp of world knowledge, allowing it to produce useful educational graphics, maps, and even functional design elements like readable barcodes. Visual sophistication The model produces outputs that feel less “AI-generated” and more like authentic photography or deliberate design, often incorporating subtle, realistic flaws that enhance the final aesthetic. Putting Images 2.0 to the test Improving a slide graphic Improve this info graphic Creating a visual timeline for a bio Create a timeline image of Khe Hy's career using info in the public domain and https://www.linkedin.com/in/khehy/ . Make sure you capture the logos and that I was promoted to Managing director at 31 in 2013 ;) Use these colors EC3E72 3293B5 Creating a new screenshot Create a screenshot of Claude Cowork that shows it on a Mac desktop, standard background. Then around the screen shot show representative outputs of powerpoint, excel, and word. It should look like an image that would be on the webpage of an AI product Telling a story via an illustration Create a professional illustration for a cartoon that shows that when the ATM was invented, people were terrified of what it would do to teller jobs. However the number of bank branches and employees increased. Work moved up the value chain to financial advice. Incorporate real data and charts to make the story more convincing. Make it in the blue color of chase. Improving my logo Improve our logo. Make it more future-forward, but also imbue a sense of tradition in it. Below put LaTour AI, you pick the font Create a visual summary Organize these ideas about AI deployment in the enterprise in a meaningful way. Make it clean and professional looking. The tweet is attached Create a training worksheet Turn this post into an interactive Q&A to help folks identify their job to be done. Use known techniques. The worksheet should help professionals think about how to integrate AI https://khemaridh.substack.com/p/what-milkshakes-can-teach-us-about Creating a WSJ advertisement Create a sample 1 page WSJ ad for my consulting business. The big headline should read: Stop using AI like a Google Search. Then below in much smaller, a CTA to hire LaTour AI (make it punchy) + include our logo (attached) Shiny new toy or part of the toolkit? TBH, I’m not sure. I remember when Gemini released its infographic tool which led to a surge of those types of posts. And then everyone stopped caring. I could see this being helpful for presentation visuals and social media assets. But even then, there’s still a huge element of human taste, creativity and aesthetics that will be hard to replace with AI. Stop prompting. Start building. Most professionals are still chatting with AI — asking questions, copying answers, starting from scratch every session. That’s sooooooo 2025. Claude Code lets you build tools that keep working when you’re not: meeting analyzers, prospect research pipelines, dashboards that surface what matters before you ask. Unfair Advantage teaches you how in 7 days — no coding required, just the reps that separate builders from prompters. $249, lifetime access, 14-day money-back guarantee. Unsubscribe https://substack.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.orTmq6v1432mWMkvyytP00vo5yOcqFJw2kX3Px91B6c?

View Email

I tried ChatGPT's new image model

khemaridh@substack.com4/30/2026

Substack

View Email

Why your firm is struggling with AI adoption

khemaridh@substack.com4/26/2026

Substack

View Email

I tried OpenAI's new "Super App"

khemaridh@substack.com4/23/2026

Substack

View Email

Claude Cowork gets some competition

khemaridh@substack.com4/19/2026

Substack

View this post on the web at https://khemaridh.substack.com/p/what-milkshakes-can-teach-us-about I think about this comic strip all the time. Let’s be honest. Using AI to summarize and create emails is one of the most popular use cases for this powerful technology. And with the rise of Claude CoWork, [ https://substack.com/redirect/40f62545-7dac-41a6-8519-ce73e928b47b?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] it’s become even easier to create elaborate presentations, briefing documents and LinkedIn posts. But when you zoom out and take stock of all these deliverables (not to mention token utilization [ https://open.substack.com/pub/khemaridh/p/ai-just-got-way-less-fun?r=bf55&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true ]), what is really happening? And are we actually moving the needle? Claude Cowork is the most powerful AI tool for knowledge work Imagine sitting down Monday morning and your meetings are already prepped, your inbox is triaged, and last week’s notes are organized — all before you opened your laptop. That’s Claude Cowork running on a schedule you set once. This self-paced course teaches you to build that system. Finish it in a weekend. The answer can be found in milkshakes I know this sounds silly, but bear with me. Here’s a 4 minute video [ https://substack.com/redirect/c5bdcb55-006a-406f-a3d4-32c723ef1fac?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] from the late marketing professor Clayton Christensen called Understanding the Job: In this video, he tells the story about milkshake sales. The majority of the milkshakes were being purchased before 8am, which is very counter-intuitive for a product that is a “dessert.” So Christensen set out to answer the question: What job is a milkshake hired to do? There’s an old-school framework in marketing called Jobs to be Done [ https://substack.com/redirect/9170744d-1867-4d49-87b5-c2d5ff1523d3?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] which states that consumers buy products to fulfill a certain job. So the “job” of Lululemon shorts is to help you get fit. And the “job” of a MacBook Pro is to help you generate income. So back to the milkshake. It turns out that the majority of milkshake buyers were morning commuters. They didn’t want something sweet after a Big Mac. Heck, they might not even have been trying to satisfy a sugar craving. Instead, they had “a long boring drive” and needed something to keep them occupied. They weren’t looking for an outright meal, but needed something to hold them over until a mid-morning breakfast. And here’s how a milkshake stacked up relative to the alternatives: Bananas were eaten too fast Donuts were too messy Bagels were hard to eat while driving Snickers bars felt too indulgent The reason this short video always blows my mind is that it shows that it’s not always clear why we do a certain activity. Which has a lot of implications in an AI world. It’s vibes all the way down I’ve worked with many companies filled with extremely hard-working professionals. They are talented and their organizations are highly profitable. But when you ask them to specify what their actual work consists of — they often find it hard to define. In Daniel Miessler’s post the most important ideas in AI right now [ https://substack.com/redirect/c5b608a9-b9c7-487b-aa8f-720022f0df3a?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] he doesn’t hold back (emphasis mine): Most organizations run on vibes and spreadsheets. AI makes all of that visible. The actual work, the actual costs, the actual quality—all of it becomes measurable in ways that were never practical before. Miessler also argues that there’s an “articulation” gap between what experts know and what’s written down: Most expertise lives in people’s heads. Cliff, the 62-year-old who knows how everything works but never documented any of it. When Cliff retires, that knowledge dies with him. And in an ideal world, every leader would have a map of both the job to be done and how it gets executed (by both humans and agents). From vibes to jobs I challenge everyone to really think deeply about their “job” and what value they are ultimately delivering to their stakeholders — whether it’s their clients, shareholders or the public good. AI is already eliminating a massive amount of tedious work — the emails and PowerPoints are already writing themselves. I know that in my business, I’ve been able to create massive amounts of decks, newsletters and training materials. If my “job” is merely to help people learn AI then I am passing with flying colors. But that’s like saying a milkshake is dessert. Upon deeper reflection, I believe that my job is to help people achieve professional excellence. And AI is just a vehicle to help them with that job. Download our free guide to Claude Cowork Unsubscribe https://substack.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.zdDL_b8A1n7FTidgtq-cig1ootvRDTl_Srb6KeToKfw?

View Email

What milkshakes can teach us about AI

khemaridh@substack.com4/16/2026

Substack

View Email

How to get your company AI-pilled

khemaridh@substack.com4/12/2026

Substack

View Email

AI just got way less fun

khemaridh@substack.com4/9/2026

Substack

View Email

The Ultimate AI Catch-Up Guide

khemaridh@substack.com4/5/2026

Substack

View Email

Claude now multi-tasks with Excel and PowerPoint

khemaridh@substack.com4/2/2026

Substack

View Email

A complete list of Claude's new features

khemaridh@substack.com3/29/2026

Substack

View this post on the web at https://khemaridh.substack.com/p/claude-cowork-how-to-use-the-new 2025 was all about prompts. [ https://substack.com/redirect/699e12a4-89fb-4f67-a2a1-b6344d654e78?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] This year is all about context. [ https://substack.com/redirect/3a19b746-2202-451d-a82c-7e99e884d148?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] Yes, the models are getting way more powerful at long-running tasks. But more importantly, they now can process much more context. Cowork just added a feature to make it much easier to manage all of this additional context. An introduction to Cowork Projects Throughout 2025, I was on the record saying that Claude Projects outperformed ChatGPT’s version. It seemed like the former used a more intelligent RAG approach to understanding and synthesizing your data. Then Claude Cowork came along [ https://substack.com/redirect/c515e146-4ebc-42d8-9510-95978b9d9252?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] and let users run sessions from their local machines. This immediately fixed the “last mile” problem — but created a new one. How was context managed when it sat on countless files and folders? This past week, the Cowork team delivered a solution. Hit your goals faster using Claude Code (like Meek Mill) Meek Mill discovered Claude Code and he’s been automating like crazy. While he received a helping hand from some “tech youngbull,” he still had some questions. (He also probably struggled with the IDE, Terminal, Skills and API endpoints.) Nonetheless, he found use cases across his family office and business processes. Claude Code users (like Meek) are watching their skills go “parabolic.” It lets you build tools that keep working when you’re not: meeting analyzers, prospect research pipelines, dashboards that surface what matters before you ask. No coding required — just the reps that separate builders from prompters. Creating your first project From Claude’s official feature announcement: [ https://substack.com/redirect/c786dd11-48bf-4d54-b132-93f39cd8eb59?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] Projects keep your tasks and context in one place, focused on one area of work. Files and instructions stay on your computer. To get started, click on the new Projects tab in the sidebar. You’ll be presented with 3 options: If you’ve spent a long time curating a project in Claude “chat,” you can easily import it into your new Cowork project. Additionally, you can bring multiple projects into your new project. Alternatively, you can Start a project from Scratch. You’ll have the option to give it: Create Custom Instructions Add files and folders as your context Specify where your outputs should be saved This is an important distinction — the folder that holds your context can be different from the folder where your outputs are saved. This will be very impactful for teams looking to share common folders for context (but creating files in a different location). Here you can see how the project pulls in my master context files. (Cowork also creates background memory, [ https://substack.com/redirect/05f2ac0d-95d8-4b25-873e-3d79f0a8e905?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] periodically refreshing the context from chats and files when it sees fit.) I ran a lazy prompt to see if it was sourcing its context from the appropriate place. Create a powerpoint presentation of the current curriculum. Add visuals when necessary. Use the style guide Here’s the PowerPoint result. Next up, I wanted to create a landing page for my upcoming course on Claude Cowork. (It’s going to be dope.) Create a sales page for a course called Unfair Advantage using Claude Cowork. The materials will be similar to the way I teach Claude Cowork during training sessions. Use this outline below. In the right panel, you could see Cowork’s plan and which files it was choosing to pull into context. There was another context panel at the bottom right, but I couldn’t figure out what exactly it did. Final thoughts on Cowork Projects If I’m honest, these behave exactly how I would expect them to behave: You can create custom instructions You can easily add and remove context There’s a natural home for scheduled tasks [ https://substack.com/redirect/6888b12f-5afe-4519-bce4-16918d7c1446?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] As for future improvements — I hope all the folder selection options get consolidated and that you can easily see when you’re running into your context limits. And finally, I’d want to better understand how data from external connectors works with this context. Unsubscribe https://substack.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.x_XOG2e2q4y10iLubiOhF9zExwX2Ru_2AN6BXaEDg8M?

View Email

Claude Cowork: How to use the new Projects feature

khemaridh@substack.com3/26/2026

Substack

View Email

Claude is coming for your messaging apps

khemaridh@substack.com3/22/2026

Substack

View Email

The Complete Guide to Claude Skills

khemaridh@substack.com3/19/2026

Substack

View Email

The Rise of the AI Chief of Staff

khemaridh@substack.com3/15/2026

Substack

View this post on the web at https://khemaridh.substack.com/p/claude-can-now-run-scheduled-tasks Chat is dead! Chat is dead! For the better part of 2025, I predicted that chat wouldn’t be how we ultimately use AI. Claude Code [ https://substack.com/redirect/ad2b8d91-30bc-4f96-afd6-366d5a5b0fe0?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ], Cowork [ https://substack.com/redirect/017a6624-e746-466c-ab4c-94bf36bd65bb?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] and now Microsoft Cowork [ https://substack.com/redirect/52bd8c37-7a33-4b94-9536-c7a50570527d?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] proved that I was right. [ https://substack.com/redirect/4a983453-d1dc-46be-b6f1-8f5bbb523a96?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] But one problem still remained: automation. Coders figured out how to solve this challenge with an archaic workflow called CRON jobs. But the normies — myself included — needed something simpler. A way to automate that you could click and organize. And, of course, Claude came through and delivered. Introducing Scheduled Tasks for Claude Cowork This powerful feature lets you run actions including creating and updating files on a schedule. Like anything in Cowork, it can access files/folders as well as any skills [ https://substack.com/redirect/6a221470-10fb-4a74-91b1-e07fae8ff811?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] and tools (like the Chrome browser extension). [ https://substack.com/redirect/8e80b490-747c-42f1-bb19-11049edfe2b2?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] To schedule a task you have to be using the desktop app and be in the Cowork tab. Once you click Scheduled you’ll be taken to the following screen where you’ll select New Task. Next, you’ll get to build out your task: It’s pretty straightforward but some additional bits: Since you only get one shot at this prompt, it better be extremely detailed. Always click the more options button. Since Cowork runs locally on your machine, you need to give it a folder, especially if it’s going to create or modify files. If the task requires some heavy lifting, such as Excel or PowerPoint or browsing the web, I strongly recommend using Opus 4.6. This is where you set the frequency. Remember that your computer has to be on and awake, or else it won’t run. Once you’ve saved the task you’ll see it on your “dashboard.” When you click into it you’ll get to modify it. Here are a few pointers: You can always edit the task (including the prompt) You can run a trial to see how it works (which I highly recommend) These are tools that you’ve given the task “persistent access” to History of runs Now let’s go through 2 examples. Build our own army of agents Everyone’s talking about AI agents. And yes, scheduled tasks are a great first start. But they’re still not that powerful. Claude Code lets non-technical professionals build agents that work autonomously — analyzing transcripts after every Zoom call, surfacing your highest-leverage emails, monitoring deal announcements while you sleep. No coding required. Just clear thinking and our step-by-step course. Unfair Advantage costs $249 for lifetime access and a 14-day money back guarantee. Example 1: Meeting Prep Meeting prep is finally automatable thanks to scheduled tasks and connectors. The workflow is really simple. Every night, Claude looks at tomorrow’s calendar to look up meetings. It then prepares me with the following workflow: Grabs a bio on the attendees (LinkedIn, company website and personal blog) Looks at prior email interactions using the Gmail Connector Looks at prior calls using the Granola Connector [ https://substack.com/redirect/cdecfc6a-9850-4daf-8d25-7662fc0e992e?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] Sends each summary to a Slack channel Here’s what the prompt looks like: Go through my calendar for the next day. Look at all the meetings. Ignore the ones that don't have any invites. Then, for each meeting, send me a separate Slack message that covers the following: - Name - Quick background on them (via LinkedIn and web search) - If they have a blog or social presence, two links to the last things they've posted. - Look in Granola to see if we've had any conversations in the past month and give a two-sentence summary. - Look in my email to see if there's anything relevant from the last month to provide context Since there will be many meetings, separate each message on Slack with "-----" So that there is visual separation Here’s the output in Slack: Example 2: Automated Market Research For the next example, I wanted to create a very customized research agent. I wanted this agent to search the web each day to look for articles and research papers that could help me further develop my training programs. I used the same exact process to set it up. Here’s the prompt that I used: Conduct a series of web searches to look for articles that can help knowledge workers get better at AI. This includes : - use cases - examples - frameworks - organizational design/hiring - mindset shifts - upskilling The avatar reading these articles are knowledge workers who work in financial services, consulting, law management, marketing who know that AI is important but are looking to get deeper. They are smart and driven, but not software engineers. We want to prioritize articles from reputable mainstream sources and prominent bloggers: - FT, WSJ, NYT, Economist, Bloomberg - Harvard Business Review, VC websites, Management Consultants (McKinsey, BCG) - Prominent Blogs: Every.to, Ethan Mollick, CRA Admired Leadership, Conor Grennan, and Allie K. Miller Create a new Excel files with: - Article Name - Source - Author - Link - Summary - Category (Use case, framework, etc) - Ranking (1-5, 5=best) Grab the top 10 articles and make sure the links aren't paywalled And here’s the result — it’s damn good. Your turn So how should you go about using scheduled tasks? Here’s a simple process to set up your first scheduled tasks: Identify a repeatable task Think about a specific desired output (update a list, send a specific message, download a file) Create a detailed prompt that captures all the information in one-shot Identify the connectors that either add context (Granola, Google Drive) or that facilitate delivery (Outlook, Slack) Test it with “run now” Go and watch A Knight of the Seven Kingdoms with all your newfound free time Keep building more and more agents. It’s THAT easy Everyone’s talking about AI agents. And yes, scheduled tasks are a great first start. But they’re still not that powerful. Claude Code lets non-technical professionals build agents that work autonomously — analyzing transcripts after every Zoom call, surfacing your highest-leverage emails, monitoring deal announcements while you sleep. No coding required. Just clear thinking and our step-by-step course. Unfair Advantage costs $249 for lifetime access and a 14-day money back guarantee. Unsubscribe https://substack.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.tbf3N7PffY0Uso7PgdJcmKg6oB8kUBpx_4fvwGzVDKs?

View Email

Claude can now run scheduled tasks

khemaridh@substack.com3/12/2026

Substack

View Email

The "Cursor Moment" for Finance

khemaridh@substack.com3/8/2026

Substack

View this post on the web at https://khemaridh.substack.com/p/i-paid-my-daughter-7-to-learn-claude I pride myself on being a thoughtful and principled parent. And I’ve read enough parenting blog posts to deeply understand that you should never use cash as a prize for intrinsic motivation. But here’s the deal. My 11 year old daughter wants to earn more money beyond her weekly chores. I wanted her to naturally come to AI to experiment and tinker — especially since the message she’s constantly getting at school is that “AI is BAD.” But she wasn’t interested, so I upped the ante. I asked Claude if it’s ok to motivate kids with cash payments. Here’s the bull case: Good idea if: it increases curiosity, she starts choosing projects, and you can gradually reduce payment without a meltdown. And the bear case: Bad idea if: she becomes more transactional about learning, refuses anything unpaid, or it undermines your family’s chore culture. Nevertheless, I bit the bullet. I offered her $7 for every hour she spent learning AI. My plan is to give her projects that would ultimately lead to her building her own AI portfolio of websites, games, agents, chatbots and products. With one caveat. I’d never ask her to do this. Instead she’d have to come to me proactively for new projects. Here’s how the first $7 played out. Stop prompting. Start building. Most professionals are still chatting with AI — asking questions, copying answers, starting from scratch every session. That’s soooooo 2025. Claude Code lets you build tools that keep working when you’re chilling: meeting analyzers, prospect research pipelines and dashboards that surface what matters before you ask. All it takes is 7 days of focus. Step 1: Download Claude To kick things off, I created her own Claude login on my MacBook (since I want her to eventually use Claude Cowork and Claude Code [ https://substack.com/redirect/dd18c2b5-7e8b-438b-9d1c-b239c3daed8c?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ]). I had her sign up for the $20 plan and download the Claude desktop app. Throughout the process, I observed how she attempted to figure out issues like installing the app and linking her Gmail account. I felt like I was giving her some “mini life-lessons” by showing her how to use WisprFlow [ https://substack.com/redirect/3bea2759-a260-4499-9947-5c1690c6f011?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] and to store login info in 1Password. Step 2: Configure Claude and Select Models Once she was in the app, I had her immediately go into settings to ensure that memory [ https://substack.com/redirect/412c3848-18e3-4ced-8103-e3fabc08feda?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] and search old chats were activated. Admittedly, the importance of these settings wasn’t obvious to her. Can you guess what we did next? Model selection. I made it very clear that there’s a hierarchy of models and that she should always start with Opus 4.6 (“trust me, bro”) and then move over to Sonnet 4.6 when she hit her limits. A brief security pause Next up was “the talk.” Not the birds and the bees, but a quick rundown of AI security: For now, we’re using it primarily to build our projects When you’re working on it, I have to be around (after all, I gotta track the hours) Do not share personal information about yourself and our family without asking Do not use this on your homework without consulting me If I’m honest, my biggest fear with AI for kids is not the job loss or privacy — it’s that it will be so tempting for them to lean on AI for emotional companionship. So that’s probably what I’m vigilantly monitoring. Step 3: Have Claude Interview You Here I used the “blank slate metaphor” to explain that AI knew nothing about her. [ https://substack.com/redirect/c6dcd7d0-f91a-4e17-a92b-514570580cd6?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] So I suggested that we have AI “interview” her about her classes, hobbies and tastes. I explained that this would help create what AI folks call “context.” (BTW, I also recommend this interview method [ https://substack.com/redirect/8df4d01e-6383-4108-a244-6bc99f497d0a?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] in all my corporate trainings.) If you’re curious about her mood to this point, the general vibe was “How much more time until I get my $7?” Step 4: Build a game Her first project was to build a game. She used a simple prompt: Build me a game with jumping rats inspired by Jellycats, Pusheen and kawaii art. Use lots of pink. Claude defaulted to building the following artifact: I then asked her to come up with 3 of her own improvements (to spur her critical thinking). And then I asked her to ask Claude for 3 more (to show her how to use AI as a co-pilot). Step 5: Set up GitHub It took me 44 years to finally understand what GitHub was and how I could use it. And I had a proud dad moment when my daughter created her first GitHub account (which we referred to as “a place to save your projects.”) Admittedly GitHub is a pretty scary website for non-coders, but with some help from Claude she was able to create her first “repo” (aka project). TBH, pushing code to GitHub to be deployed live is pretty advanced and we hit our hour limit and she ran out of steam. (We also burned through all her Opus credits.) I wonder if she comes back for round 2 this weekend… Stop prompting. Start building. Most professionals are still chatting with AI — asking questions, copying answers, starting from scratch every session. That’s soooooo 2025. Claude Code lets you build tools that keep working when you’re chilling: meeting analyzers, prospect research pipelines and dashboards that surface what matters before you ask. All it takes is 7 days of focus. Unsubscribe https://substack.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.YcRyql6mIpmMRjycP6jsEThWQL5anpFEd-4PVOnbrqY?

View Email

I paid my daughter $7 to learn Claude

khemaridh@substack.com3/5/2026

Substack

View Email

Another viral doomsday post?

khemaridh@substack.com3/1/2026

Substack

View Email

Stop using AI like it's 2025

khemaridh@substack.com2/26/2026

Substack

View Email

The $1 Trillion SaaS Reckoning

khemaridh@substack.com2/22/2026

Substack

View this post on the web at https://khemaridh.substack.com/p/have-we-reached-powerpoint-agi Friends, something big is happening. [ https://substack.com/redirect/459dce2e-70a7-4484-aa87-b14114177282?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] In the world of PowerPoint. This week, we got two announcements that bring us closer to PowerPoint AGI. Claude launched a native PowerPoint plug-in NotebookLM (by Google) finally allows you to edit and export to PPT Can we finally reclaim the thousands of hours lost each year to building slide decks? Prompting Was the Training Wheels. It’s Time to Take Them Off. I know you spent all of 2025 learning how to be a fantastic prompter. Sure, that’s super important. But if you’re still copy-pasting into a chat window and re-introducing yourself during every chat session, you haven’t created any personal leverage. Claude Code (which inspired Cowork) is a completely different game. It lives permanently inside your file ecosystem and builds tools that keep running far beyond your chat session. A great prompt gets you a good answer once. Claude Code gets you a tool that generates good answers forever — transcript analyzers that profile prospects after every Zoom call, email agents that surface your highest-leverage replies, a second brain that interacts directly with your knowledge. Unfair Advantage takes you from basic prompter to AI builder in 7 days. $249, lifetime access, zero coding required. The Claude PowerPoint Plug-in Claude recently launched a native plug-in that you run directly from the PowerPoint app. According to the product launch page [ https://substack.com/redirect/0dfa6495-2191-490e-ae07-db1329319244?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ], this plug-in can: Build from your brand templates Modify an existing deck Generate a full deck from a prompt Create native charts and diagrams Once you install the plug-in, you’ll see it in the right toolbar as a panel: I started off with a simple request: take the quote; I believe it's from Lenin (there's weeks when decades happen or something like that), and organize these seven images in a way that is coherent and legible. Use rounded corners if possible It took 3-4 iterations to get the order correct, but here was the final output: Next, I needed some design help. Here’s a very common slide that I use to explain context scaffolding: [ https://substack.com/redirect/ee06cd7b-f640-47d3-bdc0-db680ccb287a?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] I wanted to pretty it up and gave the following prompt: take this existing slide and make it look better. Use some kind of gradient color. Preserve the histogram nature of it, and make the text better. look at the color scheme from the other slides It fumbled the first iteration but got the second one on the first attempt: I tried it again on a “timeline slide” showing my career evolution into AI, but it couldn’t get it right. My issues mostly related to the placement of images: Logo too high Logo on text Missing logo Should have spelled LaTour Finally, I went for the big kahuna. One-shotting a presentation from a detailed outline. Given the shift away from chat towards agents (for all use cases), I’ve been revamping my curriculum. I dropped in a 1,000 word outline into the prompt and said: Build this Here’s one slide: And another: It was pretty clean, but tbh I didn’t love the design choices that it made. NotebookLM expands its presentation capabilities NotebookLM also made a quiet upgrade announcement: If you’re unfamiliar with this quiet AI workhorse [ https://substack.com/redirect/f1b6fde3-d997-4342-9db1-229d4aa8dbd8?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ], NotebookLM lets you upload up to 500 sources and synthesize them in a variety of different outputs. [ https://substack.com/redirect/b30e38d4-0467-46b9-bc63-3df476ffbd70?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] I plugged in this fantastic post — A Guide to Which AI to Use in the Agentic Era [ https://substack.com/redirect/4c25d3cf-687b-4723-9e32-9a3e4038eb74?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] — and asked NotebookLM to create a slide deck. Here’s one slide on NotebookLM And another featuring a horse [ https://open.substack.com/pub/khemaridh/p/chatgpts-faster-horse-problem?r=bf55&utm_campaign=post&utm_medium=web&showWelcomeOnShare=false ]: Now here’s where it gets interesting. In the past, once the slides were created, you couldn’t edit them — which really made it unusable for anything that required a degree of professionalism and design consistency. Here’s the result: Not bad, it followed the instructions pretty well and the slide is a significant improvement. Now it’s time for a little let down. While the Export to PPTX feature has been announced, it hasn’t been rolled out to my account yet. The final verdict While I enjoyed having the native PowerPoint plug-in, I thought NotebookLM had a significantly better design aesthetic. The revision feature is nice too, particularly since you can modify specific slides and/or the entire deck. The wildcard remains the “export” function which should arrive any day — I can’t wait. Build your “unfair advantage” with Claude Code Our new course teaches non-technical professionals how to build real tools — transcript analyzers, email agents, custom dashboards — without writing a single line of code. Unsubscribe https://substack.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.q1YuD0Prtk7Y1aLi8IZ3Fq5ifxMkn78ENbjRfwaKD8k?

View Email

Have we reached PowerPoint AGI?

khemaridh@substack.com2/19/2026

Substack

View this post on the web at https://khemaridh.substack.com/p/this-ai-blog-post-broke-the-internet AI has been on a bender in 2026. It started with Claude Cowork bringing agentic workflows to Excel and PowerPoint. Then Opus 4.6 and ChatGPT 5.3 Codex continued their giant strides in performance and “intelligence.” In late January, folks built their own personal AI agents on dedicated Mac minis. And then last Wednesday, Matt Shumer broke the Internet. In the odd chance you’ve been on a meditation retreat, his post Something Big is Happening [ https://substack.com/redirect/780ccfbd-3d55-4088-a81d-c6cd2dc9136f?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] went absolutely bonkers. The tweet alone has been viewed over 80 million times. Its blast radius was so expansive that old school CEOs “panic-sent” it to their executive teams as required reading. Right AND left-wing influencers used it to rally future anti-AI legislation. The dying media (Forbes and AOL.com) featured the story prominently. Even my meditation teacher sent it to me. (BTW, the post served as a dope marketing campaign for my Claude Code course “Unfair Advantage” [ https://substack.com/redirect/a0139fa4-a412-4b16-9ca5-efc2cf6e192c?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] which also saw its registrations spike!) The tl;dr of the post is that AI is an exponentially powerful technology that is coming for everything you consider sacred: your job, dignity, financial assets and your kids’ futures. It’s the ultimate AI Rorschach Test. It will violently reinforce your existing world view on AI. If you’re bullish on AI, it’ll make you more bullish. And if you think AI will destroy the world, well that destruction has already begun. Personally, I fall into the very excited camp despite modest concerns. Here are my own takeaways. Everything is software One of Shumer’s key arguments is that AI dominates at coding. He’s right and the data overwhelmingly supports this: Spotify declared on an earnings call that its best developers haven’t written a single line of code since December Both the Claude Code and Codex GPT coding agents were predominantly built by themselves A significant percentage of new code on GitHub is AI-generated Coding is an activity that lends itself particularly well to AI. The accuracy of its outputs is directly verifiable. If I want to create an OpenTable clone, there’s a finite set of outcomes (i.e. can I book this reservation) that I can independently check. And coding operates on a very narrow subset of language. (Heck, even I’ve become a coder over the past two years. The most impactful tools I’ve built are a lightweight CRM focused on developing authentic relationships, a sales calls analyzer to improve my close rate and a curriculum generator to enhance my teaching pedagogy.) Shumer’s bull scenario hinges on the belief that all of mankind’s problems (and desires) can be distilled into software problems. Healthcare, finance, education, poverty eradication, entertainment are all fixable with software. And since the cost of building high-quality software is effectively a $100/month Claude subscription — buckle up! I’ve been implementing AI for financial firms for nearly two years and I say hold your horses: Turning non-software problems into software problems is easier said than done. For starters, data access is a huge problem. While software prides itself on transparency, public data sets (like HuggingFace) and open-source code repositories — other industries are more closed off. For these other domains, this means that AI suffers from a sizable data disadvantage. It’s not trained on the most complete data and it can’t use this data as its context. (Not to mention that in industries like finance, the data itself is the competitive advantage.) Furthermore, outside of coding the answers to many questions are not easily verifiable. Let’s say I wanted to answer the question: Should I invest in Google? Let’s try to turn this into a software problem. We could: Load in all of its historical financial statements. Compare the valuation to its peer group. Assess its existing customer base and pricing power. Decompose its factor exposures. Evaluate the composition of the board of directors. To turn this investment question into a software problem, you’d have to identify all the variables that would be required to answer it. And for something like a stock investment there are tens of thousands of factors that influence the outcome. But it’s even more challenging. The question itself is problematic. Are we looking at Google as a 1-day or 100 year investment? Are we looking at the return from an absolute, relative or risk-adjusted returns perspective? This is not how much of the finance industry operates. Between old ways of doing things, legacy tools (ahem, Excel) and a general apprehension to anything “coding-related” you can understand the hesitation (and difficulty) of redesigning an entire process to be more software-oriented. The same could be applied to industries like healthcare, law and consulting. How can you turn your work processes into software problems? I believe that Shumer is right that every single worker should be striving to push the limits on transforming their existing workflows into software problems that can be tackled by AI. Yet I also believe tremendous human judgement, intuition and experience are required to integrate all of that information into an actual decision. Can you speed run domain expertise? Here’s the part of Shumer’s post that put the fear of God into millions of knowledge workers: The honest answer is that nothing that can be done on a computer is safe in the medium term. If your job happens on a screen then AI is coming for significant parts of it. The timeline isn’t “someday.” It’s already started. I think this statement requires some clarification. Jamie Dimon’s job “happens on a screen” yet I promise you AI will not be the next CEO of JP Morgan. But, if your job is formatting tables and standardizing charts in PowerPoint — I’d be careful. So what unique advantage does Jamie Dimon have versus our young grad? Domain expertise. And here is where things get interesting. It’s a huge superpower to be fluent in AI. But it pales in comparison to being AI Fluent and having domain expertise. Part of the reason why my nascent business LaTour AI [ https://substack.com/redirect/de2543b5-b66b-4dc6-9f62-a8425a50668c?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ] has been so successful is because of my 15 years of experience in financial services. I have a strong understanding of the outcomes, nuance, constraints, edge cases, workflows and personalities of the industry. (Not to mention tons of relationships to facilitate biz dev.) And when paired with AI, you’ve got yourself a nasty combo. I believe Shumer is correct that AI Fluency will become table stakes. You can already see that happening today. But the real differentiator will be the domain expert who leverages AI to the max. So how does one acquire domain expertise in an AI world? Here’s my mental model for building domain expertise: books versus reps. “Reps” comes from doing the damn thing, whereas “books” comes from learning about the damn thing. Every industry will have its own books/reps ratio. For example, becoming a nuclear physicist probably requires a heavy dose of reading, research and learning. (And then a ton of experience as well.) On the other hand, becoming a great salesperson probably doesn’t require reading a lot of books about sales. Instead you have to go out and experience firsthand the rejection, coalition-building, empathy, social cues and persuasion techniques. AI is an extreme accelerant to the learning component. You have immediate and free access to the world’s knowledge in a customizable and conversational form. It’s patient and adapts to your learning style. However, acquiring the reps — doing the damn thing — is going to be hard to speed run. It’s these reps that will enable you to see where AI falls short, the relationships you’ll need to build and where the big opportunities lie. Should you be scared? Back to the Rorschach test. If you’ve made it this far, here’s what I know about you. I know that you’re wildly capable of taking your existing body of work and converting as much of it as possible into software problems. Is it a new way of thinking? Absolutely. Is it difficult? Not if you truly understand why you do the work you do. I also know that you are a voracious learner who will utilize AI to improve your craft. The more pride you take in your craft, the more motivated you’ll be to get better at it. And then there’s the reps. These reps compound over time. All you gotta do is show up — and the rest takes care of itself. Here’s one easy way to stay ahead A huge part of Shumer’s argument is that everyone can use code to solve problems. It’s cheap, powerful and accessible to everyone via Claude Code. I built this course because I wanted you to add this skill to your AI toolkit. In under a week you’ll learn how to build transcript analyzers that profile prospects after every Zoom call, email agents that surface your highest-leverage replies, a second brain that interacts directly with your knowledge. Unsubscribe https://substack.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.eQOskanisi1JzcEomUYwiRIbN9v1ehvsEPX6gkKWAo0?

View Email

This AI blog post broke the Internet

khemaridh@substack.com2/15/2026

Substack

View Email

Claude Cowork for Windows is here

khemaridh@substack.com2/12/2026

Substack

View Email

Will AI tear through Corporate America

khemaridh@substack.com2/8/2026

Substack

View this post on the web at https://khemaridh.substack.com/p/how-to-use-clawdbot-with-no-risk “There are decades where nothing happens, and there are weeks where decades happen.” I can’t even keep track anymore. ClawdBot. MoltBook. New Codex app. Nvidia calls off the OpenAI deal. Anthropic Legal. SaaS stocks sell-off. I’m still knee-deep on ClawdBot. While I don’t think the product itself will ever go mainstream — it is a massive inflection point on where things are headed. Last year using o3 and Claude Code for the first time felt like Magic Moments —there was just something about the experience, so unique, so cool, that you could get a little glimpse of the future. Personally, I want to experience that Magic Moment for ClawdBot. I don’t recommend you install it. It’s gnarly. But I still want you to experience it. And here’s how you can do it without putting yourself at risk. It starts with AI Coding Fluency You don’t need to know how to code to make this work. But you need to understand how to navigate a command line, GitHub, and an Agentic IDE (like Claude Code, Cursor or Codex). Thankfully you can literally ask ChatGPT how to do all these things. Or you can sign up to Unfair Advantage — our Claude Code course for smart, non-technical professionals [ https://substack.com/redirect/5a5cc53e-756e-462f-a0b6-0b69fb46935c?j=eyJ1IjoiNzRiZXZxIn0.26TDO4Ygo5l69NnkPzr2Ax0IIjv-K5IWcmNJKcK8PLY ]). Since the project is Open Source you can access it via GitHub. Next, you can copy the code (without running or installing it) to your local machine. It’s just a bunch of text files. The key here is that you’re not running it, it’s that you’re setting it up on your machine. To copy the code, you literally need one command inside a blank Claude Code window. If you’re new here, the easiest way is via the Claude Desktop app. download this repo https://github.com/openclaw/openclaw This will trigger a 3-5 minute download of all the files. Here’s what it looks like in the finder: Deconstruct the project with the help of AI You can now fully interact with this project through simple chat via Claude Code. Here’s my first question: ❯ ELI5, how does this project work? Here’s the response (truncated): What OpenClaw Is Imagine you have a really smart robot friend who can answer questions, help you with tasks, and remember conversations. But here’s the problem: this robot lives in a special building (the AI company’s servers), and you can only talk to it through specific doors (their apps). OpenClaw is like a telephone operator that lets you talk to your robot friend through ANY messaging app you already use - WhatsApp, Telegram, Discord, or even text messages! The Simple Story The Main Characters: 1. You - sending messages from your phone 2. The Messenger (WhatsApp, Telegram, etc.) - where you send your message 3. OpenClaw (The Gateway) - the telephone operator/translator 4. The Smart Robot (Claude, ChatGPT) - the AI that thinks and answers 5. Your Computer - where OpenClaw lives [Truncated] Remember — OpenClaw is not running. The LLM (Claude Code, in this case) is reviewing the code and answering. The questions we can ask are endless: ❯ Can you explain in less than three paragraphs to someone who understands LLMs well but is not a coder what makes this different than just using Claude Code on your machine? Here’s the response: OpenClaw vs Claude Code: Different Tools, Different Jobs Claude Code is an interactive coding assistant that runs in your terminal - you invoke it when you need coding help, it completes a task, then exits. OpenClaw is an always-on gateway server that connects messaging platforms (WhatsApp Telegram, Discord, etc.) to LLMs, acting as a persistent personal assistant. Think of Claude Code as a specialized IDE tool, while OpenClaw is more like running your own hosted chatbot service that routes messages from any channel to any LLM provider. The core difference is architectural: Claude Code manages terminal sessions and file operations for development workflows, while OpenClaw manages persistent conversational sessions across multiple messaging channels with features like pairing/authentication, group chat policies, media handling, and cross-platform message routing. OpenClaw lets you text your AI assistant from your phone via WhatsApp using the same conversational thread that remembers context across days, whereas Claude Code gives you a powerful coding agent that lives in your terminal for the duration of your development session. In short: Claude Code = coding assistant in your terminal, OpenClaw = personal AI infrastructure that makes LLMs accessible through the messaging apps you already use, running 24/7 on your hardware with full control over routing, permissions, and data storage. Next, I wanted to “leg into” the project with the appropriate security guardrails: ❯ On a 1-10 regarding security, I’m a 2 (risk averse). In short bullets, explain to me how I can “leg in” to this and make it more secure Phase 1: Isolated Testing Run OpenClaw in a Docker container or VM first (not your main machine) Use a dedicated, limited-scope API key (not your main Anthropic/OpenAI account) Start with just ONE channel (Telegram recommended - easiest to isolate) Set dmPolicy: “pairing” so only explicitly approved contacts can message Don’t enable group chats yet Phase 2: Lock Down Access Store bot tokens in separate files with chmod 600 (not in config or environment variables) Use allowFrom with explicit user IDs only (no wildcards, no usernames) Set groupPolicy: “disabled” to block all group messages Enable only the tools/actions you need: channels.telegram.actions.reactions: false, etc. Review logs daily initially: openclaw logs --follow Phase 3: Network Hardening [Truncated] Since ClawdBot has an entire ecosystem of plug-ins, skills, and customizations — I wanted to understand where I could extend it: ❯ I know that there's a rich ecosystem of add-ons and customizations. Can you also look at the documentation in their GitHub repo and make three brief recommendations about safe extensions? They could really help me better understand what's so special about this Each example merits its own section, starting with Memory Core Plugin: Memory-Core Plugin (Safest - Built-in, No External Dependencies) What it demonstrates: OpenClaw’s contextual memory across conversations - the AI remembers what you told it weeks ago. Security profile: ✅ Ships with OpenClaw (already vetted) ✅ Local-only (no external API calls) ✅ Read-only search through your workspace files ✅ No network exposure, no credentials needed ⚠️ Stores indexed data in ~/.openclaw/memory/ (keep permissions tight) Why it’s special: Unlike Claude.ai where each conversation is isolated, this lets your assistant remember “I prefer Python 3.11” or “my dentist appointment is Feb 15th” across all channels and conversations. It’s like giving the AI a notebook. Next up was something on diagnostics so I skipped it 😜. The next tip was effectively Longer Term memory. After all, LLMs are the worst co-workers since they don’t learn from their mistakes. This integration uses RAG techniques to access your old conversations: Memory-LanceDB Plugin (Moderate Risk - Requires OpenAI API) What it demonstrates: Long-term semantic memory - the AI can recall “that conversation about security in October” without you specifying it. Security profile: ⚠️ Requires OpenAI API key (sends embeddings to OpenAI) ⚠️ Stores vector database locally in ~/.openclaw/memory/lancedb/ ✅ Auto-capture/auto-recall can be disabled (start with both off) ℹ️ Creates embeddings of your conversations (consider privacy implications) 🔒 Mitigation: Use a dedicated OpenAI key with spending limits ($5/month cap) Why it’s special: This is what separates OpenClaw from simple chatbots. It builds a searchable knowledge base of your conversations. Ask “What did I tell you about my travel plans?” and it semantically searches months of history instantly. For my last query, I tried to peek into the future: ❯ If you had to predict the future and the big foundation model companies implement this into their offering, what would it look like and where would they struggle? Here’s what they’d build: Anthropic/OpenAI/Google Could Launch in 6-12 Months: “Claude Connect” or “ChatGPT Channels” - A hosted service where: You authenticate once with OAuth Connect your messaging accounts (WhatsApp, Telegram, Discord) Their servers handle routing between your messages → their AI Unified conversation history across platforms Persistent memory (they already have Projects/Memory features) Beautiful web dashboard for configuration [Truncated] And here’s where the foundation labs would struggle in building an OpenClaw alternative: 1. The Trust Paradox OpenClaw runs on your machine keeping data local, but foundation companies would route all your messages through their servers where they can see everything. Risk-averse users and enterprises won’t trust sending personal WhatsApp/Signal conversations through Anthropic’s infrastructure, especially when compliance requires data sovereignty. 2. The Business Model Conflict OpenClaw is provider-agnostic (switch between Claude, GPT, local models freely), but foundation companies need vendor lock-in to maintain revenue. If they support competing models they become a commodity router; if they don’t, users choose more flexible alternatives. 3. The Platform Access Problem WhatsApp Business API costs money per message and Meta would block mass personal routing, while iMessage has no official API (Apple would sue for reverse-engineering). OpenClaw individuals can use personal-use APIs and reverse-engineered protocols that don’t scale commercially but work fine at small scale. 4. The Compliance Nightmare Routing millions of users’ messages makes them legally responsible for data residency, content moderation, age verification, wiretapping laws, and breach liability. They’d need to scan and log everything for compliance, contradicting privacy promises and creating massive insurance costs. [Truncated: Customization Gap, Multi-Tenancy Technical Challenge] Now it’s your turn Fire up your GitHub and kick off your own ClawdBot “simulation.” What burning questions will it answer? Drop your responses in the chat. If you’re looking to level-up your career with Claude Code, look no further than our new course Unfair Advantage: Claude Code Mastery (for Non-technical smart people). Not only will you whip up your own “ClawdBot Simulator” you’ll build custom KPI dashboards to outperform your goals, transcript analyzers to generate more sales and intelligent email agents to identify the highest-leverage opportunities. Unsubscribe https://substack.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.SQMuiA8LQl-7uYTy3Ww1AlppBpBNtzjn5g2OIPopwKo?

View Email

How to use ClawdBot (with NO risk)

khemaridh@substack.com2/5/2026

Substack

View Email

Should we be scared now?

khemaridh@substack.com2/1/2026

Substack

View Email

Will Claude Eat Excel?

khemaridh@substack.com1/29/2026

Substack

View Email

The next hot thing in AI (is a Mac Mini)

khemaridh@substack.com1/25/2026

Substack

View Email

6 Tips to Unlock NotebookLM's superpowers

khemaridh@substack.com1/22/2026

Substack

View Email

Move over ChatGPT

khemaridh@substack.com1/18/2026

Substack

View Email

Will Claude Cowork change work forever?

khemaridh@substack.com1/15/2026

Substack

View Email

AI: The double-edged sword

khemaridh@substack.com1/11/2026

Substack

View Email

2025 AI Retrospective: The Tools, the Workflows, the Reality

khemaridh@substack.com1/8/2026

Substack

View Email

Just don't call me a "vibe coder"

khemaridh@substack.com1/4/2026

Substack

View Email

Welcome to Future-Proof Your Career with AI

khemaridh@substack.com1/3/2026

Substack

Explore all emails from Future-Proof Your Career with AI

Jensen makes the case for open source AI

AI is coming for your Slack workspace

Using ChatGPT 5.6 for knowledge work

Will Claude Cowork get replaced (by ChatGPT)?

Chat is dead. Agents are the new default.

Is Fable worth the hype?

How to get the most out of Fable 5

Meet my new AI CFO

The rebel alliance

I built a self-improving "company brain"

Codex takes control of your computer

The power user's guide to Claude skills

The AI Shift in Investing

I built a "headless CRM" in Codex

Your AI budget is growing. Your returns aren’t.

11 ready-to-use AI skills to help you work faster

This agent handles iMessage, Gmail and WhatsApp

Here's how AI helps me read more

Lessons in "Founder Mode" and AI

Is Notion AI ready for prime time?

How to design agent skills (like a pro)

AI now integrates seamlessly with Excel

How to use the new Codex Super-App

AI is coming for finance

AI is coming for your mind

I tried ChatGPT's new image model

Why your firm is struggling with AI adoption

I tried OpenAI's new "Super App"

Claude Cowork gets some competition

What milkshakes can teach us about AI

How to get your company AI-pilled

AI just got way less fun

The Ultimate AI Catch-Up Guide

Claude now multi-tasks with Excel and PowerPoint

A complete list of Claude's new features

Claude Cowork: How to use the new Projects feature

Claude is coming for your messaging apps

The Complete Guide to Claude Skills

The Rise of the AI Chief of Staff

Claude can now run scheduled tasks

The "Cursor Moment" for Finance

I paid my daughter $7 to learn Claude

Another viral doomsday post?

Stop using AI like it's 2025

The $1 Trillion SaaS Reckoning

Have we reached PowerPoint AGI?

This AI blog post broke the Internet

Claude Cowork for Windows is here

Will AI tear through Corporate America

How to use ClawdBot (with NO risk)

Should we be scared now?

Will Claude Eat Excel?

The next hot thing in AI (is a Mac Mini)

6 Tips to Unlock NotebookLM's superpowers

Move over ChatGPT

Will Claude Cowork change work forever?

AI: The double-edged sword

2025 AI Retrospective: The Tools, the Workflows, the Reality

Just don't call me a "vibe coder"

Welcome to Future-Proof Your Career with AI