# gabezen.com — Gabriel J. Pérez Irizarry (Gabe) > You pay for AI and still do the work. I figure out what Claude Team should do for your company, then I make it work. One job live in 14 days, guaranteed. Gabe sets up Claude for teams of up to 20: standard Claude Team, set up to do the busywork overnight. He writes down how the company works (its rules, prices, process and voice: the company manual, delivered as a required Claude plugin from a repo the company owns), turns on Anthropic's own features (enterprise search, projects, scheduled tasks, org skills), connects each person's email, docs, calls, spreadsheets and calendar, and keeps it current as Anthropic ships. Claude then answers questions from those tools with the source attached and runs jobs overnight, like a sales call turned into a priced proposal, lead replies drafted with the company's real prices, or the day's action items turned into tasks. Anything going out waits for a person's OK. Nothing is sent on its own. He runs his own company, Nexrizen (https://www.nexrizen.com), on Claude. 15+ years shipping software, 50+ production apps, Fractional CTO at CITYROW through its acquisition by WaterRower. ## Ways to work with Gabe ### Start here: Hand Off One Job, Free Send me your last proposal, customer reply or weekly update. I build a setup that drafts the next ones on its own, and you set it up yourself. - One job, by email in 48 hours: Free Included: - A setup built from your example, in your voice - A walkthrough video made for you - Your next one drafted in your brand, or a sample - Instructions for your own Claude account No meeting and no card. You'll need a paid Claude plan to run it. Start: https://gabezen.com/hand-off-one-job/ Not sure which job? A prompt at https://gabezen.com/hand-off-one-job/find/, pasted into any AI, finds the work you keep repeating. ### Claude, Set Up in 14 Days I write down how your company works and set up Claude Team to do the work overnight. - One team of up to 20: From $3,997 - Release care, optional: $997/mo Founding price for the first 3 teams, then $5,997. Bigger, or on Claude Enterprise? We'll scope it on the call. Included: - Your company manual, on every seat - Search across your email, docs, calls and chat - A project for each client - Three jobs that run overnight, like call to proposal - Two trainings, an AI policy and an exit doc Guarantee: One job live on your real work by day 14, or all your money back. Book: https://gabezen.com/contact/ ### Build It Yourself I coach your team over screen share. You do the typing, so you know how every piece works. - Four 90-minute sessions, up to 5 people: $1,997 You'll need a paid Claude plan. Which one depends on your team and what you're building, and we pick it at the kickoff. Included: - A kickoff to pick the jobs worth handing off - Your manual and search, set up together - Your first job running on a schedule - Recaps, and up to 5 email questions after Book: https://gabezen.com/contact/?need=learn_to_build What the busywork costs, for example: 5 people × 3 hrs a week × 52 weeks = 780 hours a year. That's a year of proposals and follow-ups written by hand. The setup is a one-time cost. ### A smaller start: One Job, Installed in 7 Days Pick one job: proposals, customer replies, weekly updates, or the one your AI found. I set it up in your own Claude with your rules and prices, and fix it on your real work until you'd send the drafts as they are. "The one your AI found" is a job picked by the prompt at https://gabezen.com/hand-off-one-job/find/. - One job, set up in 7 days: $997 - A 45-minute kickoff, then a first draft within 48 hours of it - Set up in your own Claude account, with your prices, rules and voice - Every note on your real drafts fixed at the rule, so it stays fixed - A handover where it drafts from fresh work while you watch - A week of email support after - Bonuses: your team check reviewed, a map of what to hand off next, and a one-page AI policy Guarantee: If it isn't drafting from your real work by day 7, you get your money back and keep what I built. 4 spots a month. Buy the full setup within 30 days and all $997 counts toward it. Details: https://gabezen.com/hand-off-one-job/installed/ ## Pages - [Home](https://gabezen.com/): You pay for AI and still do the work. I figure out what Claude Team should do for your company, then I make it work. One job live in 14 days, guaranteed. - [A Company Brain for Executives](https://gabezen.com/for/executives/): I teach Claude your company, so work is waiting when you walk in: board prep, follow-ups, status updates. One job live in 14 days, guaranteed. - [A Company Brain for Founders](https://gabezen.com/for/founders/): I teach Claude your business, so the work gets done while you sleep: proposals, investor updates, lead replies. Live in 14 days. - [Claude for Law Firms](https://gabezen.com/for/law-firms/): I teach Claude your small law firm: open cases checked each morning, consult notes for the team, intake replies from your fee schedule. Live in 14 days. - [Pricing](https://gabezen.com/pricing/): Claude Team, set up to do the work overnight, from $3,997, guaranteed. Or set it up yourself with four coached sessions for $1,997. One job free first. - [Company Brain Alternatives Compared](https://gabezen.com/compare/): ChatGPT, Copilot, Gemini, Notion AI, Glean and more, next to Claude Team set up properly. Priced for a team of 10, with where each one wins. - [ChatGPT Business vs Claude, Set Up for You](https://gabezen.com/compare/chatgpt-business/): ChatGPT Business ships the parts: projects, tasks and company knowledge. Someone still has to set it up. How it compares with Claude Team, set up for you. - [Claude Team: Set It Up Yourself or Have It Done](https://gabezen.com/compare/claude-team/): Every Claude Team plan has enterprise search, skills and scheduled tasks. Set them up yourself, with my coaching, or have me do it in 14 days. - [Claude Setup Consultants, Compared](https://gabezen.com/compare/claude-consultants/): Twenty Claude setup consultants, from King Intelligence and Praecipio to Cloudbeast and WotAI: what each costs and promises, next to mine. - [Microsoft Copilot vs Claude, Set Up for You](https://gabezen.com/compare/microsoft-copilot/): Copilot Business is strong inside Microsoft 365 at $21 a seat. A Claude setup is for teams whose work is spread across Zoom, Drive, QuickBooks and more. - [Gemini for Workspace vs Claude, Set Up for You](https://gabezen.com/compare/gemini/): Gemini comes with Google Workspace and its agents live in Google's apps. How it compares with Claude Team, set up for you across all your tools. - [Notion AI vs Claude, Set Up for You](https://gabezen.com/compare/notion-ai/): Notion AI is hard to beat if your company writes everything down in Notion. Claude Team, set up for you, is for knowledge that lives in email and calls. - [Glean vs Claude, Set Up for You](https://gabezen.com/compare/glean/): Glean is enterprise search built for companies with an IT team, and it doesn't publish prices. I set up Claude for teams of up to 20, from $3,997. - [AI Agent Builders vs Claude, Set Up for You](https://gabezen.com/compare/agent-builders/): Dust, Lindy and Zapier Agents can do the work, but you design every agent. My Claude setup comes with the jobs built and your rules written, in 14 days. - [Hiring a Freelancer vs Claude, Set Up for You](https://gabezen.com/compare/freelancer/): I priced 56 Fiverr gigs that promise an AI trained on your business, from $30 to $20,000. What each tier gets you, and how a Claude setup compares. - [A Second Brain vs Claude, Set Up for You](https://gabezen.com/compare/second-brain/): A second brain is one person's notes with an AI on top. Claude Team, set up for you, puts your company on every seat and stays when someone leaves. - [Hiring an Ops Person vs Claude, Set Up for You](https://gabezen.com/compare/ops-hire/): A good ops hire brings judgment and relationships. A Claude setup does the finding and follow-up drafts overnight, for a one-time fee. Often you want both. - [Interactive demo](https://gabezen.com/try/): A made-up agency's Claude, set up properly. Have it write a proposal from this morning's call, or ask it anything, with the source behind every line. - [Hand Off One Job, Free](https://gabezen.com/hand-off-one-job/): No meeting. Pick the proposal, reply or update you're tired of writing and send me your last one. In 48 hours, a Claude setup drafts the next ones. Free. - [I install one job in your Claude and tune it on your real work.](https://gabezen.com/hand-off-one-job/installed/): I set one job up in your Claude (proposals, replies or updates), load your rules and prices, and fix it on real work. Live in 7 days for $997, guaranteed. - [Jobs Claude Can Do for Your Business](https://gabezen.com/jobs/): Over 30 jobs Claude already does at my company and my clients': proposals, client updates, books, hiring. What you get from each, and what you stop doing. - [What Is a Company Brain?](https://gabezen.com/company-brain/): A company brain is an AI with your company's context, tools and skills, so it does the work, not just answers. What's in mine, and what still breaks. - [Why I Only Set Up Claude](https://gabezen.com/why-claude/): I set up one AI platform for companies, and it's Claude. What Anthropic ships for teams, how it treats your data, and where Claude still falls short. - [Book a free 30-minute call](https://gabezen.com/contact/): Pick my brain for 30 minutes, free. Bring the work your team still does by hand and leave knowing which job Claude should take over first. - [Free guide: Build It Yourself with Claude Code](https://gabezen.com/guide/): Build and publish a dashboard from a spreadsheet with Claude Code, no engineering background needed. About two hours, with every prompt ready to copy. - [Free setup plan](https://gabezen.com/setup/): Six quick questions about your team's tools. Get a free plan for setting up Claude Team, with the hours it takes and where people get stuck. - [Free Claude Team check](https://gabezen.com/check/): Paying for Claude Team and still doing the work? Ten quick questions, then your fix plan: the exact clicks, templates to copy, and a PDF to keep. - [What 33 Claude Connectors Can Change](https://gabezen.com/connectors/): Free table, with CSV and JSON. Read 33 Claude connectors' tools and sorted them by what each can edit, send, share, delete or pay. Tick yours and get what to set first. - [Can My Employer See My AI Chats?](https://gabezen.com/can-my-employer-see-my-ai-chats/): Free table, with CSV and JSON. On all eight work plans compared, someone at your company can get your chats. Who, in Claude, ChatGPT, Copilot and Gemini, each with a source. - [Claude Code Cheat Sheet](https://gabezen.com/claude-code-cheatsheet/): The Claude Code commands you'll use every day, plus skills, subagents, hooks, MCP and plugins, with working examples. Every claim links to the docs. - [What I've Learned](https://gabezen.com/blog/): Every post. Notes from running my company on a company brain built with Claude Code, with real numbers and what broke. Older posts cover AI coding and Django. - [What I've Built](https://gabezen.com/projects/): 50+ apps shipped. Led tech at CITYROW through acquisition. A look at what I've built and how. - [About Gabe](https://gabezen.com/bio/): 15+ years building software and 50+ production apps. Now I set up Claude for small teams, starting with one job of yours, free in 48 hours. ## Topics - [Is It Safe?](https://gabezen.com/blog/topic/is-it-safe/): What Claude can read, change and delete in each tool you connect, who at work can see your AI chats, and what to set before you turn it on. - [AI at Work](https://gabezen.com/blog/topic/ai-at-work/): Bookkeeping, onboarding, SOPs, custom GPTs and more: notes on where AI helps a small business, where it falls short and when to skip it. - [Claude Code](https://gabezen.com/blog/topic/claude-code/): How our team uses Claude Code day to day: the setup we share, the prompts that work, what non-developers do with it and what we removed. - [Company Brain](https://gabezen.com/blog/topic/company-brain/): Notes from running my company on a company brain built with Claude Code: what it does all day, what we threw out and what still goes wrong. - [Building Software](https://gabezen.com/blog/topic/building-software/): Notes on building software with AI: unit tests, agents that get stuck, Django, microservices and when craft still pays. ## Elsewhere - [Nexrizen](https://www.nexrizen.com): Gabe's company, run on Claude - [LinkedIn](https://www.linkedin.com/in/gabrieljperezirizarry/) - [X](https://x.com/gabezen) - [Instagram](https://www.instagram.com/elgaboulder/) ## More - [Full version](https://gabezen.com/llms-full.txt): this file plus the FAQ, case studies and every post's text - Any one post as Markdown: add index.md to its URL, or request it with Accept: text/markdown ## Optional Every post, newest first. Skip this section if context is short. - [What a Claude Setup Costs: Seats, Help, Our Prices](https://gabezen.com/blog/what-a-claude-setup-costs/): Claude Team is $20 to $25 a seat a month. Help runs from free to five figures. The extras on the bill, every published setup price, and mine. - [Month-End Close with Claude, Xero and QuickBooks](https://gabezen.com/blog/month-end-close-claude-xero-quickbooks/): How our QuickBooks close uses Claude, and what it caught. What Xero's and QuickBooks' connectors can change, and how to check the numbers. - [Is Claude Team Worth It for a Small Team?](https://gabezen.com/blog/is-claude-team-worth-it/): Is the Claude Team plan worth it? For two or more people doing client work, usually yes, but not for the usage. What a seat adds, and when Premium pays. - [Does Claude Team Train on Your Data? Plan by Plan](https://gabezen.com/blog/does-claude-team-train-on-your-data/): Claude Team doesn't train on your chats unless someone clicks thumbs up or down. Training, retention, ZDR, the DPA and HIPAA on every Claude plan. - [Claude Team Plan Setup: The Admin Settings to Change](https://gabezen.com/blog/claude-team-plan-setup/): Claude Team plan setup, owner side: who's Primary Owner, how people join, SSO, connector approvals, rules on every seat, and the defaults worth changing. - [Claude Code in the Desktop App, for Non-Developers](https://gabezen.com/blog/claude-code-desktop-app/): Claude Code in the desktop app is the Code tab: your files, no terminal, every paid plan. How to turn it on and how it differs from chat and Cowork. - [Can Teammates See My Claude Chats on a Team Plan?](https://gabezen.com/blog/can-teammates-see-my-claude-chats/): No, not unless you share one. What coworkers on Claude Team can see anyway: project files, shared chats and their files, artifacts and Slack. - [Can My Employer See My Gemini Chats?](https://gabezen.com/blog/can-my-employer-see-my-gemini-chats/): On a work Google account, yes. Any super admin can export your Gemini chats, and Vault can keep ones you delete. Who sees what, and what to tell staff. - [Can My Employer See My Copilot Chats?](https://gabezen.com/blog/can-my-employer-see-my-copilot-chats/): On a work account, yes. Copilot files every chat in your work mailbox, temporary ones too, and deleting doesn't always clear it. Who can look. - [Can My Employer See My ChatGPT Chats?](https://gabezen.com/blog/can-my-employer-see-my-chatgpt-chats/): On a company ChatGPT seat, an owner can export your chats, and Enterprise logs catch temporary and locked chats too. Who sees what. - [Is Claude Safe for Business? What I'd Check First](https://gabezen.com/blog/is-claude-safe-for-business/): Yes for most small-company work on Team or Enterprise, once you set what each connected tool can change. When you need Enterprise, and what to keep out. - [Can Claude Open a Google Drive Link?](https://gabezen.com/blog/can-claude-open-google-drive-links/): Only through the Google Drive connector, signed in as an account that can open the file. Why pasted links fail, four ways that work, and what Claude gets. - [Claude Desktop App Update: What's Worth Using](https://gabezen.com/blog/claude-desktop-app-update/): How to update Claude Desktop, then two months of changes checked against Anthropic's docs: what to use, what to skip, what Team plans don't get yet. - [Is It Safe to Connect Claude to Stripe?](https://gabezen.com/blog/is-it-safe-to-connect-claude-to-stripe/): Safe in a sandbox or with tight permissions. Stripe's human check covers refunds, not voided invoices, canceled subscriptions or new webhooks. - [Is It Safe to Connect Claude to Ramp?](https://gabezen.com/blog/is-it-safe-to-connect-claude-to-ramp/): Safe on Read-only. The listing says read-only, but Full access is Ramp's default, and a bill approved there can start the payment. - [Is It Safe to Connect Claude to Notion?](https://gabezen.com/blog/is-it-safe-to-connect-claude-to-notion/): Safe if you check first. On Business or Enterprise with Notion AI, it can reach linked Slack, Gmail and Drive. Notion's docs list trashing databases. - [Is It Safe to Connect Claude to Jira?](https://gabezen.com/blog/is-it-safe-to-connect-claude-to-jira/): Safe once an admin sets it. Connecting for Jira tickets can also open Confluence public links and code merges unless you turn those groups off. - [Is It Safe to Connect Claude to Gusto?](https://gabezen.com/blog/is-it-safe-to-connect-claude-to-gusto/): Safe if one admin connects it and approves every call. The listing says query and analyze. The tools can run payroll and link a bank account. - [Is It Safe to Connect Claude to Google Calendar?](https://gabezen.com/blog/is-it-safe-to-connect-claude-to-google-calendar/): Safe if writes need your approval. An event, change or RSVP from Claude emails every attendee by default, and one Anthropic page calls it read-only. - [Is It Safe to Connect Claude to Fireflies?](https://gabezen.com/blog/is-it-safe-to-connect-claude-to-fireflies/): Safe once you change one default. Out of the box any teammate can open your meetings, and Claude likely inherits that. Fix it before anyone connects. - [Is It Safe to Connect Claude to Dropbox?](https://gabezen.com/blog/is-it-safe-to-connect-claude-to-dropbox/): Safe if you block delete first. Claude's Dropbox connector can remove files but can't restore them, and on a team the big undo belongs to an admin. - [Is Claude Code Safe to Use?](https://gabezen.com/blog/is-claude-code-safe/): Yes, once you set it up. It runs as you, beyond any one folder, and new sessions let a second model approve actions in your place. - [Can I Put Client Data Into Claude?](https://gabezen.com/blog/can-i-put-client-data-into-claude/): Only if the contract allows it and you're on a company plan. Training off doesn't release an NDA, and deleting a chat can leave what Claude remembered. - [Is It Safe to Upload Bank Statements to Claude?](https://gabezen.com/blog/is-it-safe-to-upload-bank-statements-to-claude/): Safe enough for your own statements if you strip account numbers, skip training and memory, and check totals. Incognito still keeps a copy for 30 days. - [Is It Safe to Connect Claude to Xero?](https://gabezen.com/blog/is-it-safe-to-connect-claude-to-xero/): Reasonably safe on Xero's read-only connector, going by its docs. I haven't tested it. Its two MCP routes can write. Read the consent screen first. - [Is It Safe to Connect Claude to Slack?](https://gabezen.com/blog/is-it-safe-to-connect-claude-to-slack/): Safe once you set its tools. Blocking private search isn't enough: Claude's channel and thread tools read private messages too, and it can send. - [Is It Safe to Connect Claude to Shopify?](https://gabezen.com/blog/is-it-safe-to-connect-claude-to-shopify/): Safe enough through a limited staff login. Shopify says orders and customers are read-only, yet its GraphQL tool declares write access to both. - [Is It Safe to Connect Claude to Microsoft 365?](https://gabezen.com/blog/is-it-safe-to-connect-claude-to-microsoft-365/): I'd connect it for reading, once you've checked write tools are off and searched SharePoint as your most junior employee. - [Is It Safe to Connect Claude to HubSpot?](https://gabezen.com/blog/is-it-safe-to-connect-claude-to-hubspot/): Safe enough for CRM work with writes on approval. It can also publish to your website, and HubSpot Starter's audit log skips CRM changes. - [Is It Safe to Connect Claude to Google Drive?](https://gabezen.com/blog/is-it-safe-to-connect-claude-to-google-drive/): Claude's directory lists tools to share and trash Drive files. Assume it can read what you can. Fix old sharing and block those two tools first. - [Is Claude Team SOC 2 Compliant?](https://gabezen.com/blog/is-claude-team-soc-2-compliant/): Yes, Claude Team is in scope, and it can't do HIPAA. The audit window closed before Cowork existed. What to write on the security questionnaire. - [Is Claude Cowork Safe for Business Use?](https://gabezen.com/blog/is-claude-cowork-safe-for-business/): My read of the docs: fine on Team for low-stakes office work with the defaults changed. Not for regulated work, because audit logs are Enterprise only. - [Can My Employer See My Claude Chats?](https://gabezen.com/blog/can-my-employer-see-my-claude-chats/): On a company seat, at least one person can export your Claude chats, incognito too. Who sees what, and what to tell your team. - [AI Project Failure Rates: Where the 95% Comes From](https://gabezen.com/blog/ai-project-failure-rate/): I traced the famous AI failure rates to their sources. RAND's 80% is a footnote to a CEO profile, and MIT's own chart points nearer 75% than 95%. - [How to Set Up Claude's Free Small Business Plugin](https://gabezen.com/blog/claude-small-business-plugin/): Anthropic's free Small Business plugin interviews you, then drafts proposals and follow-ups. How to set it up well, and what it won't do for a team. - [Is It Safe to Connect ChatGPT to Gmail? What to Check](https://gabezen.com/blog/is-it-safe-to-connect-chatgpt-to-gmail/): ChatGPT can read and draft in your Gmail, and since June 2026 send. Who approves depends on a setting. What it can touch, and what to ask before Allow. - [Is It Safe to Connect Claude to QuickBooks? What It Can Touch](https://gabezen.com/blog/is-it-safe-to-connect-claude-to-quickbooks/): Claude's QuickBooks connector can create, send and delete invoices. Ours only reads reports. What each route can touch, plus a one-page connector policy. - [Claude Code as a Chief of Staff: 39 Runs, and Where Claude Duet Wins](https://gabezen.com/blog/claude-code-chief-of-staff/): Our Claude Code chief of staff runs at day's end: 39 runs, 626 owed replies found. How to build yours, and where Claude Duet's morning brief wins. - [How to Train ChatGPT on Your Own Data, and When Not To](https://gabezen.com/blog/train-chatgpt-on-company-data/): Asking how to train ChatGPT on company data usually means you want it to look things up. Here are the four routes, priced, and where each broke for us. - [How Much Does an AI Chatbot for Your Business Cost? 56 Gigs](https://gabezen.com/blog/hire-someone-ai-knowledge-base/): I priced 56 Fiverr gigs that promise an AI trained on your business, from $30 to $20,000. What each tier buys, and six questions to ask before you pay. - [Custom GPTs Are Retiring: What to Use for Your Business](https://gabezen.com/blog/custom-gpt-for-business/): OpenAI retires custom GPTs on December 11, 2026. One worked fine for one person and a few files. For a team it goes stale. Here's what to move to instead. - [Claude Code vs Cowork: Why Our Team Still Uses the Terminal](https://gabezen.com/blog/claude-code-vs-cowork/): Claude Code vs Cowork: same agent, different home. Why our team picked the terminal, and how scheduled tasks and Claude Duet run while you're away. - [Claude Code for Non-Developers: Four Installs, One Team](https://gabezen.com/blog/claude-code-for-non-developers/): Claude Code works for non-technical people when the work is documents. Our team installs four things. What we took off their plate, and what's still hard. - [Is It Safe to Connect Claude to Gmail? What It Can Touch](https://gabezen.com/blog/can-claude-read-my-email/): Claude's Gmail connector can read, draft and send, with your OK on each by default. What it can touch, what it keeps, and the rules we run. - [A Company Brain Example: The One That Runs My Company](https://gabezen.com/blog/company-brain-that-runs-a-company/): A real company brain example: the folders, the jobs that feed it and the commands that read it, after 3,474 commits. And what still breaks. - [Self-Improving AI: Our 3 Feedback Loops Collected Nothing](https://gabezen.com/blog/feedback-loops-collected-nothing/): We built three ways for our AI to learn from its own work. Two months later two logs were empty and a hook never fired. Loops that fill ride on daily work. - [AI Note Taker Privacy: 46 Transcripts Got Past Our Gate](https://gabezen.com/blog/confidential-transcripts-reached-main/): AI meeting transcripts hold salaries and passwords. We built a four-layer gate to hold them, and two days later a docs PR merged 46 right past it. - [How to Build a Company Brain in Claude Code (and What Broke)](https://gabezen.com/blog/company-brain-claude-code-setup/): Start with a folder of markdown and a CLAUDE.md. Once a team wrote to ours, we added folder rules, a hook and a nightly janitor. It still has 245 findings. - [CLAUDE.md Best Practices: What Our 223-Line File Taught Us](https://gabezen.com/blog/claude-md-file-best-practices/): CLAUDE.md best practices from a 223-line file: what goes in, what belongs in a skill, where nested files go, and a starter template under 25 lines. - [Decision Log: Write Down the Rulings Your AI Follows](https://gabezen.com/blog/write-down-the-rulings/): Our decision log is one append-only file of rulings our AI loads before drafting proposals. Six of the first came from one person's private AI memory. - [What Is Tribal Knowledge? Five Places Ours Hid From Our AI](https://gabezen.com/blog/tribal-knowledge-ai/): Tribal knowledge is what your team knows and never wrote down. Ours hid in five places, including a README that said 73 rules when the file had 151. - [We Deleted 52 Claude Code Skills in Three Months](https://gabezen.com/blog/deleted-52-skills/): Our Claude Code skills and commands went from 41 to 103 in a month. We deleted 52, the one we missed came back better, and the pile grew back anyway. - [AI for Onboarding New Employees: What Worked on Three Hires](https://gabezen.com/blog/ai-employee-onboarding/): We used AI to onboard three hires with kits written from docs we already had. It caught stale docs before they did and got day one wrong. - [Why AI Agents Fail in Production: 8 Times Our Tools Lied](https://gabezen.com/blog/small-ways-tools-lied/): Our company brain broke more often from bad inputs than from the model. Eight times a tool returned a wrong answer with no error, and the check we added. - [Can Claude Send Emails for You? Ours Drafts and Never Sends](https://gabezen.com/blog/ai-drafts-every-email-sends-none/): Claude can send Gmail for you as of August 2026, with approval on by default. Ours found 626 replies we owed, drafts them, and sends none. - [AI Bookkeeping With Claude: 15 Months, Then QuickBooks](https://gabezen.com/blog/ai-bookkeeping-small-business/): Can AI do your bookkeeping? Ours, built on Claude, sorted 559 transactions with none left uncategorized. We still moved to QuickBooks. Here's why. - [AI for Writing SOPs: Draft From What Actually Happened](https://gabezen.com/blog/ai-for-sops/): AI writes useful SOPs when it drafts from real projects. How we wrote a 435-line kickoff SOP from three retros, and what we don't let the AI write. - [Claude Code Use Cases for Business: What Our 66 Commands Do](https://gabezen.com/blog/claude-code-use-cases-for-business/): Claude Code use cases for business that pay off put documents in and get documents out. What our 66 commands do by department, and where each falls short. - [AI Meeting Action Items: Ours Made 1,042 Tasks in 4 Months](https://gabezen.com/blog/ai-created-1042-tasks/): Our AI turned meetings and email into 1,042 tasks in four months, mostly noise. Telling it to bundle failed. A script that does the counting worked. - [Two Founders, One Company Brain, and the Duplicate Problem](https://gabezen.com/blog/two-founders-one-brain/): My co-founder and I each run an AI day-end process over the same clients. Keeping it from filing everything twice took a shared table and cross-inbox keys. - [Claude Code for Project Managers: The Weekly Update We Kept](https://gabezen.com/blog/claude-code-for-project-managers/): Claude Code for project management: ours drafts weekly client updates and retros, refuses to draft while a video link is missing, and never presses send. - [Should I Automate This? First Ask Whether It Should Exist](https://gabezen.com/blog/should-it-exist-before-automating/): Eliminate before you automate. Our AI build gate said DON'T to a vacation reminder for the wrong reason, so I added a question zero. Here it is. - [Two Bots Argued on a Pull Request for 21 Hours](https://gabezen.com/blog/two-bots-argued-21-hours/): Our Claude Code GitHub Action reviewer and a nightly routine looped on one PR for 21 hours: 147 review runs, about $500. The fixes we made on both sides. - [Claude Code Team Setup: Shared Settings, /setup, What Broke](https://gabezen.com/blog/claude-code-team-setup/): Claude Code team setup for people who don't code: one repo, shared settings in git, a /setup command that reads our roster, and what broke on day one. - [Knowledge Graph vs RAG: We Piloted a Graph and Killed It](https://gabezen.com/blog/knowledge-graph-tested-and-killed/): We piloted an LLM knowledge graph on 46 documents with kill criteria set first. 89% of edges were true, but a rerun matched 16% of labels. Search won. - [How to Prompt Claude Code: 5 Lessons From 356 Sessions](https://gabezen.com/blog/making-of-guide/): Five Claude Code prompting lessons from 356 sessions building an interactive guide with SkiFree and a music engine. The biggest: name the reference. - [Software Craftsmanship in the Age of AI: When Craft Pays](https://gabezen.com/blog/ai-craft-vs-outcomes/): AI made the choice between clean code and shipping visible. Where craft still pays for itself, where it's a trap, and what the METR study measured. - [Microservices vs Monolith: What Segment's 140 Services Cost](https://gabezen.com/blog/microservices-coordination-tax/): Segment folded 140+ microservices into one monolith and shipped shared-library fixes faster. When microservices pay off and when a small team pays the tax. - [Why AI Agents Get Stuck in Loops (and Quit Too Early)](https://gabezen.com/blog/llms-control-flow-problem/): AI agents get stuck in loops when they have to decide how many times to do something. A Collatz study shows why, and how to make an agent stop on time. - [Is AI Making Developers Worse? Yes, and They're Worth More](https://gabezen.com/blog/ai-agents-force-multipliers/): AI coding agents write code that's almost right. Catching what's wrong takes fundamentals, so developers who have them are worth more now. - [How Law Firms Are Using AI: Notes From a Legal Conference](https://gabezen.com/blog/ai-adoption-legal-tech/): Law firms are using AI for research and first drafts, with a person checking every citation. What I saw at a legal conference, and what the surveys say. - [Panel of Experts Prompting: Templates and What Research Found](https://gabezen.com/blog/panel-of-experts-prompting/): Panel of experts prompting asks AI to simulate specific experts who argue. Copyable templates, what the debate research found, and when it's overkill. - [Best Backend Framework for Vibe Coding: Why I Pick Django](https://gabezen.com/blog/django-llm-native-era/): LLMs have read 20 years of Django code that puts every file in the same place. That's why Claude writes better Django than React, and why I add HTMX. - [Can AI Write Unit Tests? 200 Tests, 15% Caught Real Bugs](https://gabezen.com/blog/ai-vibe-coding-django/): Yes, with a check. Claude wrote about 200 Django tests for me in a month, and roughly 15% failed on the first run because of real bugs. My workflow.