A good tutor hides their impatience for about an hour. What changes when a struggling student gets a tutor with no battery to drain? New from BrainStream on infinite patience, swallowed questions, and the kids who stop apologizing: brainstream.ltd/2026/09/28/t…

Last week my manuscript introduced the same engineer twice, 24 scenes apart. Full portrait both times: what he built, why he mattered, down to his sneakers. I would never have caught it rereading. The book has 82 scenes. It's a memoir about building a theme park attraction at Epcot, and the cast appendix lists 44 real names. Real names come with real problems. The source transcripts spell one of those names two different ways; a cast list typed forty years ago settles it six to one. You can't fix that with a typo sweep. It takes someone who has read every scene. So I set rules, in plain English, and my AI assistant enforces them. Every person gets introduced exactly once; the first mention carries the portrait, every mention after is just a name. The women who held serious jobs on that 1982 crew get named, never lumped into "the men." Fold one scene into another and everything downstream renumbers, with the chapter map, cast appendix, and changelog updated behind it. A proofread pass caught 33 typos across 15 files, each fix verified after it landed. My half is the creative half. Which scenes fold, which endings stay, what gets cut. This week I'm making some of those calls from a cruise ship in Alaska. The bookkeeping follows within minutes. The assistant has read all 82 scenes more recently than I have. OpenClaw is a self-hosted AI assistant that lives on a small server, reads my files, remembers every rule I've given it, and treats manuscript consistency as a full-time job. Running your own OpenClaw on an Amazon server costs as little as $20 a month depending on how you use it, and the longer you use it, the better it gets. Setting up OpenClaw wasn't trivial, and lots of people have talked about hitting snags. I wanted to make it as easy as possible, so I created a class that incorporates all the things I learned, best practices, and suggestions about what to do during your first hour and your first day with a full-time AI assistant. aipartnerschool.com #OpenClaw #AIAssistant #AmWriting #Memoir #WritingCommunity #SelfPublishing #Authors #Productivity

Saturday I board a nine-day cruise to Alaska. The office comes with me anyway - it just does not need me. Every morning my AI assistant has already been through the overnight logs: server health, backups, website uptime, the email. The digest is waiting on my phone when I wake up, wherever the ship happens to be. Anything that needs my word arrives as a draft. I approve, edit, or skip - from a deck chair if I like. Everything else it just handles: the scheduled posts, the snapshots, the routine maintenance, the small fires that used to eat the first hour of every day. It has not been flawless. Last week it woke me with an urgent warning that turned out to be its own watchdog watching a retired gauge. It told on itself, and we pointed the sensor at the live one. OpenClaw is a self-hosted AI assistant running on your own server, with your tools and your memory. I teach two classes on setting up and getting the most from OpenClaw: aipartnerschool.com #OpenClaw #AIAssistant #Automation

Publishing a children's picture book on KDP is the hardest self-publishing format there is. I just took a student's book from Word doc to print - and turned what I learned into a service: themeperks.com/from-word-doc…

A children's book author sent me her manuscript: text in a Word file, painted illustrations as scans. Three hours later she had a print-ready 40-page interior, a wraparound cover, and a fixed-layout ebook. She never opened layout software. Neither did I. My AI assistant did the mechanics. It regenerated all 19 of her illustrations at 4K, using her originals as reference so the characters kept their own faces (nothing derails a picture book faster). Scripts typeset the pages to Amazon's print specs and assembled the ebook. The whole image bill came to about $4. Then her proof came back and she caught what the vision checks missed: arm hair mismatched between pages 27 and 29. The first fix overcorrected into dense orange fur. Rejected. The second matched the reference art exactly. The mother read every page; the machine checked every page. Every judgment call stayed human, and the InDesign week never happened. OpenClaw is a self-hosted AI assistant that lives on your own server, connects to your tools, and works around the clock. It built the whole pipeline, which is now a paid service I offer authors. Running your own OpenClaw on an Amazon server costs as little as $20 a month depending on how you use it, and the longer you use it, the better it gets. Setting up OpenClaw wasn't trivial, and lots of people have talked about hitting snags. I wanted to make it as easy as possible, so I created a class that incorporates all the things I learned, best practices, and suggestions about what to do during your first hour and your first day with a full-time AI assistant: aipartnerschool.com #OpenClaw #AIAssistant #SelfPublishing #SmallBusiness #Automation #BookProduction #Publishing #Entrepreneurship

Anthropic and Google are now watermarking the text their AI models write. Invisible statistical fingerprints baked into the prose, so anyone with the right detector can prove a paragraph came from a machine. Plenty of writers cheered. Finally, a way to tell machine prose from the human kind. I think the question is already aging out. John Gruber has written Daring Fireball, the Apple blog, since 2002. He took the watermark schemes apart last week. The mark works by hijacking the randomness that makes prose good, so watermarked text is slightly worse text. And even a perfect watermark answers a question readers were never going to ask. His advice: "If it's good, read it. If it's not, don't. ... If you read something and enjoy it, and subsequently find out it was generated by an LLM, don't feel bad. You read something good that you enjoyed." Asking who wrote something is about to sound like asking which word processor they used. I run a full-time AI assistant. It pulls my sales data, audits my websites, fixes broken code, publishes my blogs. Everything it hands me gets the same judgment as everything else: did it work, is it true, is it any good. The words I sign are mine; the rest is machinery. OpenClaw is a self-hosted AI assistant that runs on your own server, with your tools and your data. Running your own OpenClaw on an Amazon server costs as little as $20 a month depending on how you use it, and the longer you use it, the better it gets. Setting up OpenClaw wasn't trivial, and lots of people have talked about hitting snags. I wanted to make it as easy as possible, so I created a class that incorporates all the things I learned, best practices, and suggestions about what to do during your first hour and your first day with a full-time AI assistant: aipartnerschool.com #OpenClaw #AIAssistant #AIWriting #ContentCreation #WritingCommunity #AITools #AIPolicy #TechNews

New in Blooloop: my piece on the story under the pyramid. Journey into Imagination looked like a gentle ride through a glass pyramid. Inside, it was one of the boldest engineering projects EPCOT ever attempted. Jim Carstensen and I finally wrote the inside story of the crew who built it. Read the piece: blooloop.com/opinion/steve-a… The book: amazon.com/dp/B0H8B42FW7/?ta… #EPCOT #Imagineering #DisneyHistory

My web server froze for the fifth time in ten days this morning. Each time, my AI assistant's health check caught it within a minute and rebooted the box while I did other things. This morning I finally told it to alert me when it does that. I'd rather know. Five freezes is enough. Rebuild it, I said. Cloning the old machine failed twice; the disk image was rotten at boot. So it did a clean rebuild instead: fresh operating system, 20 GB of files (143,000 of them), the whole database, the security certificates, all eight websites verified before the cutover. Same IP address, no DNS changes. The sites were down for about 40 seconds. My total contribution was a handful of chat messages. It also wrote itself a new rule: one more freeze and it rebuilds on the operating system the old server ran for years without a hiccup. The freezes started the week we switched operating systems. It noticed before I did. OpenClaw is a self-hosted AI assistant that runs on your own server. It does the kind of sysadmin work most small businesses hire out, or put off, or both. Running your own OpenClaw on an Amazon server costs as little as $20 a month depending on how you use it, and the longer you use it, the better it gets. Setting up OpenClaw wasn't trivial, and lots of people have talked about hitting snags. I wanted to make it as easy as possible, so I created a class that incorporates all the things I learned, best practices, and suggestions about what to do during your first hour and your first day with a full-time AI assistant. Details: aipartnerschool.com #OpenClaw #AIAssistant #Sysadmin #ServerMigration #Entrepreneur #SmallBusiness #Automation #DevOps

A client's search tool was quietly returning every listing in the database to users in five states. Nobody complained for weeks. Type your zip code, get nearby results. Standard stuff. But for certain zip codes, you got everything. Not an error. Not empty results. The entire database. Which looks correct if you don't know what "correct" is supposed to look like. My AI traced it to two problems stacked inside each other. The zip code database was from 2013. A 33,000-entry government dataset, and some codes created since then simply didn't exist in it. Columbia, SC (29217) was one. When the code couldn't find a zip code, it returned null for coordinates. The distance filter then had nothing to compare against, so it silently skipped. Every listing passed through. No error message, no log entry, nothing. The app looked like it was working. It was handing the entire database to anyone whose zip code wasn't in a 13-year-old dataset. The worst bugs don't crash. They return plausible-looking wrong data and keep going. Your business probably has one hiding somewhere: a form that submits incomplete fields, a filter that skips when input is missing, a price calculation that defaults to something plausible. Fix took minutes once the root cause was visible. Current database with 41,000+ entries, real error message for unknown codes, filter that fails closed instead of open. The AI found it by reading the code line by line, the way a human developer would. No magic. Just patience at a scale humans don't volunteer for. OpenClaw is a self-hosted AI assistant that lives on your server and works through your tools, files, and APIs. Running your own OpenClaw on an Amazon server costs as little as $20 a month depending on how you use it, and the longer you use it, the better it gets. Setting up OpenClaw wasn't trivial, and lots of people have talked about hitting snags. I wanted to make it as easy as possible, so I created a class that incorporates all the things I learned, best practices, and suggestions about what to do during your first hour and your first day with a full-time AI assistant. aipartnerschool.com #OpenClaw #AIAssistant #SoftwareDevelopment #QualityAssurance #SmallBusiness #Debugging #Automation #TechTips

New book! Journey Into Imagination: The Inside Story of the Engineers Who Brought EPCOT's Legendary Pavilion to Life Co-authored with Jim Carstensen, who was there — the engineers, the 3am equipment room sessions, the concrete dust, the phantom E-stops. A firsthand account of building one of EPCOT's most beloved attractions. Now on Amazon: amazon.com/dp/B0H7DGFMNR?tag… #EPCOT #Disney #Imagineering #ThemePark #BookLaunch

I pay $1,200 a year for my AI assistant's brain. No usage meters, no per-token charges, no surprise bills. A year ago that sounded impossible. You rented intelligence by the token. Input tokens, output tokens, cache tokens, each priced separately like a phone bill designed by a casino. Three things changed. Open models caught up. The model running my business comes from a lab most people haven't heard of. Email triage, WordPress publishing, data analysis, course management, ad optimization, social scheduling. I switched from a premium per-token plan in June. Haven't noticed a quality drop. Local inference got practical. A 12-billion-parameter model from Google runs on a standard laptop and matches what required a server rack last year. Free and open-source. Per-token economics flipped. When capable intelligence costs nothing to run, companies charging premium rates for marginal advantages start looking like landlines. I'm writing this from a train somewhere between Banff and Calgary. The server at home doesn't care. OpenClaw is a self-hosted AI assistant that lives on your server and connects to your tools, your APIs, and your files. It runs on whatever model you point it at. Running your own OpenClaw on an Amazon server costs as little as $20 a month depending on how you use it, and the longer you use it, the better it gets. Setting up OpenClaw wasn't trivial, and lots of people have talked about hitting snags. I wanted to make it as easy as possible, so I created a class that incorporates all the things I learned, best practices, and suggestions about what to do during your first hour and your first day with a full-time AI assistant. aipartnerschool.com #OpenClaw #AIAssistant #OpenSource #LLM #LocalAI #AIHosting #TechTrends #Automation

Andrej Karpathy posted something last week I'd been doing for months without a name for it. Long, rambling, voice-dictated prompts work better than carefully constructed ones. You talk. The AI sorts it out. Your mess gives it more signal than a tight prompt. 2.5 million impressions. Because everyone who's tried an AI assistant knows the pressure. Trying to compress a fuzzy idea into the right words before you know what you want to say. I stopped doing that months ago. I dictate. Five, ten minutes straight. The AI gets the tangent about font sizes, the aside about a competitor, the half-formed idea I'm not sure about. It hands back something cleaner than what I started with. Tight prompts make you decide what matters before you know. Voice rambling lets the AI find the shape. OpenClaw is a self-hosted AI assistant on your own server. It handles email, publishing, marketing, research, and dozens of other tasks, the kind of work eating your mornings. Running your own OpenClaw on an Amazon server costs as little as $20 a month depending on how you use it, and the longer you use it, the better it gets. Setting up OpenClaw wasn't trivial, and lots of people have talked about hitting snags. I wanted to make it as easy as possible, so I created a class incorporating all the things I learned, best practices, and suggestions about what to do during your first hour and your first day with a full-time AI assistant. aipartnerschool.com #OpenClaw #AIAssistant #VoiceDictation #PromptEngineering #Productivity #AIWorkflow

192 em-dashes to convert. 13 chapters opening with "I was." 400 words to trim from one chapter. Two characters who needed more texture. One scene that needed dialogue after the disaster. Twelve revisions across 32 chapter files. Each one simple. Together, a week of keyboard work. I'd gotten developmental notes on a manuscript. The kind every writer recognizes: clear, specific, and exhausting to execute. So I recorded a voice memo. Twelve instructions, the way I'd describe them to an editor across a desk. Expand Catharine's mother: coupons, hex codes, the receptionist job, the memory-care angle. Benjamin Lein should be a biker who sends emails about aliens. The balloon scene needs the post-disaster dialogue, not just the disaster. My AI assistant transcribed the memo, broke it into 12 line items, and produced a new draft with every change implemented. Em-dashes converted. "I was" openings reworked. Characters expanded with the details I'd dictated. Word counts trimmed where I flagged them. Then it ran an editorial evaluation. 8.5 out of 10, up from 8. Ninety-four to ninety-five percent publication-ready. The em-dash reduction alone moved the needle. I made the creative calls. The AI handled the mechanical work. Twenty minutes of dictation replaced a week of tedium I would have hated every minute of. OpenClaw is a self-hosted AI assistant that lives on your server and remembers your work. Running your own OpenClaw on an Amazon server costs as little as $20 a month depending on how you use it, and the longer you use it, the better it gets. Setting up OpenClaw wasn't trivial, and lots of people have talked about hitting snags. I wanted to make it as easy as possible, so I created a class that incorporates all the things I learned, best practices, and suggestions about what to do during your first hour and your first day with a full-time AI assistant. aipartnerschool.com #OpenClaw #AIAssistant #BookEditing #WritingProcess #Productivity #ManuscriptRevision #EditingWorkflow

I have 152 students in one of my courses. Until last week I had no idea who was actually watching the lessons. My AI assistant found a complete progress-tracking database on my course server. Lesson-level data, watch position, completion percentages, timestamps. It had been quietly collecting for months. Nobody was reading it. So I said: build me a report. One conversation later I had enrollment counts by school, active students over 7 and 30 days, a leaderboard, per-course engagement, revenue. The AI pulled the data via SSH, formatted the whole thing, and scheduled a weekly cron job. Every Sunday at 7 AM, the numbers land on my phone. Along the way it found a bug. The school detection code was defaulting every student to "writing" regardless of which school they'd enrolled through. One-line fix. Without it, the entire report was scrambled. Then: certificates. A completion certificate template existed but was broken, redirecting students to a 404. My AI fixed the template, redesigned it with custom border art, made it school-aware (different name and URL per domain), optimized for print, and linked it to the account page. Students at 90%+ completion now see a View Certificate button. I spent maybe 20 minutes on review decisions. The rest was the AI working through the stack: discover the data, build the report, fix the bugs, ship the feature. OpenClaw is a self-hosted AI assistant running on your own server. It connects to your tools, reads your files, and handles real work while you do other things. Running your own OpenClaw on an Amazon server costs as little as $20 a month depending on how you use it, and the longer you use it, the better it gets. Setting up OpenClaw wasn't trivial, and lots of people have talked about hitting snags. I wanted to make it as easy as possible, so I created a class incorporating all the things I learned, best practices, and suggestions about what to do during your first hour and your first day with a full-time AI assistant. Easy OpenClaw: udemy.com/course/easy-opencl… Get Real Work Done: udemy.com/course/get-real-wo… #OpenClaw #AIAssistant #OnlineCourses #StudentEngagement #EdTech #CourseCreation #DataAnalytics #Automation

Installing an AI assistant takes 20 minutes. Making it useful takes longer. Phase 1 is mechanical: install the software, pick a model, connect it to a chat app. Phase 2 is tools: API keys for WordPress, Dropbox, email, image generation. An hour, maybe. Phase 3 is the part I had to figure out myself. I think of it as the brain. The brain is six text files. A personality file covering how the assistant talks and what it values. A user file with who I am and how I work. A memory file holding every person, project, and lesson from months of collaboration. A playbook of behavioral rules. A daily journal for handoff between sessions. And a guide tying it together. Phase 3 took longer than the first two combined. Not because the files are hard to write. Because they force you to decide what kind of assistant you actually want. (The AI will do exactly what you tell it; the problem is figuring out what to tell it.) Plenty of people stop at Phase 2. They install the software, plug in some tools, start chatting. Things go well for a few days. Then the assistant starts forgetting preferences, asking the same questions, giving generic answers. Same software, same model. Six text files are the difference between a chatbot and something that resembles a colleague. I've done this five times now. The personality, memory, and playbook from assistant #1 became the template for #2 through #5. Each got customized for a different job. By the fifth setup, I was cloning a trained employee and assigning them a new role. OpenClaw is a self-hosted AI assistant on your own server. You control the model, the memory, the tools, the schedule. Over months, it accumulates enough context to function like a colleague who's been there since the beginning. Running your own OpenClaw on an Amazon server costs as little as $20 a month depending on how you use it, and the longer you use it, the better it gets. Setting up OpenClaw wasn't trivial, and lots of people have talked about hitting snags. I wanted to make it as easy as possible, so I created a class that incorporates all the things I learned, best practices, and suggestions about what to do during your first hour and your first day with a full-time AI assistant. Easy OpenClaw: udemy.com/course/easy-opencl… Get Real Work Done: udemy.com/course/get-real-wo… #OpenClaw #AIAssistant #AIAutomation #Productivity #SmallBusiness #SelfHosted #AIAgents #Entrepreneur

The cost of running a full-time AI assistant snuck up on me. Not the server. The tokens. I run two AI assistants on my server. They handle email, social media, course management, infrastructure monitoring. Around the clock. The conversation context for a persistent agent grows to hundreds of thousands of tokens. Prompt caching is what makes that affordable. Cache reads cost roughly 12x less than cache writes with Anthropic. As long as your stable context hits cache, each turn is pennies. Mine stopped hitting. Calls that should have been cache reads were billing as full writes. A session costing pennies was costing dollars. I tried Google's Gemini 3.5 Flash. Its OpenAI-compatible endpoint wasn't caching at all. Every call, full price. OpenAI's GPT models worked but had their own pricing curves. After weeks of auditing token usage and testing alternatives, I moved everything to GLM-5.2 from Z.AI. Flat-rate coding plan. Zero per-token cost. All 11 cron jobs, both servers, every sub-agent. The part I didn't expect: GLM is genuinely good at agentic work. Complex tool chains, long context, error recovery. I kept Anthropic and Gemini as fallbacks. Haven't needed them. OpenClaw is a self-hosted AI assistant that runs on your server, reads your code, and handles tasks while you sleep. Running your own OpenClaw on an Amazon server costs as little as $20 a month depending on how you use it, and the longer you use it, the better it gets. Setting up OpenClaw wasn't trivial, and lots of people have talked about hitting snags. I wanted to make it as easy as possible, so I created a class that incorporates all the things I learned, best practices, and suggestions about what to do during your first hour and your first day with a full-time AI assistant. Easy OpenClaw: udemy.com/course/easy-opencl… Get Real Work Done: udemy.com/course/get-real-wo… #OpenClaw #AIAssistant #GLM #LLM #PromptCaching #AgenticAI

Every course platform I'd ever used was built for the platform's benefit, not the student's. Teachable. Thinkific. Podia. They all look the same: a list of courses on the left, a cramped video player in the middle, progress buried in a database you can't see. Customize anything? Support tickets. Want to understand what students are doing? Good luck. I had a different picture in my head. Something that worked more like Netflix. Students pick a course, it plays, they come back. No clutter. No friction. Clean on mobile. So I asked my AI assistant to build it from scratch. Fourteen hours later, I had a live site. Not a mockup. Not a plugin with 47 settings. A full course platform: 22 courses, 640+ lessons, passwordless one-tap login, a custom lesson player, progress tracking, mobile-responsive design. My AI handled the architecture, wrote the code, debugged the authentication, built the admin tools, and launched the whole thing at writingacademy.com. I handled the decisions: which features mattered, what it should feel like, when it was done. The AI costs paid for themselves in the first 24 hours. Ten new members enrolled at launch. This is what a full-time AI assistant actually does. It doesn't just answer questions. It builds things. Running your own OpenClaw on an Amazon server costs as little as $20 a month depending on how you use it, and the longer you use it, the better it gets. Setting up OpenClaw wasn't trivial, and lots of people have talked about hitting snags. I wanted to make it as easy as possible, so I created a class that incorporates all the things I learned, best practices, and suggestions about what to do during your first hour and your first day with a full-time AI assistant. Easy OpenClaw: udemy.com/course/easy-opencl… Get Real Work Done: udemy.com/course/get-real-wo… #OpenClaw #AIAssistant #CourseCreator #OnlineLearning #AIAgents #Automation #Solopreneur #IndieHackers

OpenAI recently shipped computer-use for Windows. Every tech influencer is posting clips of an AI as it moves a mouse, clicks dropdowns, and fills out spreadsheets. It looks like magic. But mouse-moving is a terrible way to run a business. When you automate a task by pantomiming a human, you inherit every bit of friction built into the desktop. The AI moves slow; it gets confused when a button shifts by three pixels; it gets stuck clicking 'Yes' on permission prompts. Every action requires a full screenshot round-trip to the cloud. I run my business with a self-hosted AI assistant. It doesn't drive my mouse, and I don't want it to. Instead, it works behind the scenes. It talks directly to my databases, calls WordPress REST APIs, and runs command-line scripts. It rebuilt a friend's DJ directory in a week; it audited my books and optimized my ad campaigns; it updates my courses without opening a browser. This is the difference between pantomime and capability. One looks great in a twenty-second demo on X. The other actually gets work done when you aren't looking. Yes, setting up API connections and background workers takes more effort than watching a mouse move. But once they run, they don't break. You want business outcomes, not an AI pretending to be an administrative assistant. Running your own OpenClaw on an Amazon server costs as little as $20 a month depending on how you use it, and the longer you use it, the better it gets. Setting up OpenClaw wasn't trivial, and lots of people have talked about hitting snags. I wanted to make it as easy as possible, so I created a class that incorporates all the things I learned, best practices, and suggestions about what to do during your first hour and your first day with a full-time AI assistant. Easy OpenClaw: udemy.com/course/easy-opencl… Get Real Work Done: udemy.com/course/get-real-wo… #OpenClaw #AIAssistant #AIAgents #Automation #Solopreneur #IndieHackers #Tech #SoftwareEngineering

Twenty-two scanned memos from 1983, sitting in a folder for years. Primary sources for a book I've been meaning to write. Typewritten, faded, a few with handwritten notes in the margins. I'd tried processing them before. Got through three, then something more interesting came along. Last Saturday I pointed my AI at all 22. It OCR'd them, read them alongside seven other reference documents, and in one session produced a book plan that went through seven revisions as it found problems. Final output: 18 chapters, 44 interview questions organized by priority, and a corrections log flagging where sources contradict each other. It caught a key person who vanishes from the memos in April and a timeline two published accounts tell differently. My Saturday: correcting OCR errors and saying yes or no to chapter decisions. The interviews start Tuesday. OpenClaw is a self-hosted AI assistant that lives on your own server and learns your work over time. Running your own OpenClaw on an Amazon server costs as little as $20 a month depending on how you use it, and the longer you use it, the better it gets. Setting up OpenClaw wasn't trivial, and lots of people have talked about hitting snags. I wanted to make it as easy as possible, so I created a class that incorporates all the things I learned, best practices, and suggestions about what to do during your first hour and your first day with a full-time AI assistant. udemy.com/course/easy-opencl… udemy.com/course/get-real-wo… #OpenClaw #AIAssistant #Research #AI #Productivity #OCR #DocumentProcessing #Writing