# Abel Dawit > Helping you automate your work and grow your career. Newsletter and free guides. Abel Dawit teaches practical AI workflows for people who want AI to do more of the work they already do. Every Instagram post teaches one workflow with the exact prompt, and each comes with a free guide. The Scale on Autopilot newsletter walks through one full build every week. Abel has about a decade of experience in data and AI, implements AI-native systems for large enterprises in healthcare, pharma, life sciences, and financial services, holds a B.A.Sc. in Bioengineering from UPenn and a MicroMasters in Statistics and Data Science from MIT. Website: https://abeldawit.com Email: hello@abeldawit.com Instagram: https://instagram.com/abeldawit.co TikTok: https://tiktok.com/@abeldawit.co X: https://x.com/abeldawitco Threads: https://www.threads.com/@abeldawit.co ## About I'm Abel. I've spent about a decade putting data and AI systems to work inside large, complex organizations. Now I build AI automation for real businesses and share the exact workflows here, so you can use AI to get more done in the work you already do. - AI Implementation: I implement AI-native systems and agentic workflows for large enterprises in healthcare, pharma, life sciences, and financial services. - Research: I worked as a clinical data scientist in academic radiology, analyzing large imaging datasets in PHI-governed environments. - Data Foundation: About a decade in data strategy and analytics. B.A.Sc. in Bioengineering from UPenn. MicroMasters in Statistics and Data Science from MIT. ## Resources ### The Claude Skills Guide URL: https://abeldawit.com/resources/claude-skills-guide Type: Guide Price: Free with a newsletter sign-up Length: 10 pages Turn the workflows you repeat every week into Claude Skills and run them on command. What's inside: - What Claude Skills are and how Claude picks the right one - How to build your own Skill, two ways - How to make your Skills trigger reliably - The /morning-brief Skill, ready to copy Who it's for: Anyone who repeats the same instructions to Claude and wants to set them up once. ### The Resume Guide URL: https://abeldawit.com/resources/resume-guide Type: Guide Price: Free with a newsletter sign-up Use Claude to check your resume against a real job posting and see exactly what to fix before you apply. What's inside: - How to install the tailored-resume-generator Claude Skill - How to run it against a real job posting to optimize your resume Who it's for: Job seekers who want a resume that matches each posting they apply to. ### The Muse Starter Guide URL: https://abeldawit.com/resources/muse-starter-guide Type: Guide Price: Free with a newsletter sign-up Length: 16 pages What Meta's personal AI agent does, what it can handle for you, and how to start using it. What's inside: - What Muse actually does - How to connect your apps - The workflows worth trying first - The honest limitations Who it's for: Anyone curious whether Meta's personal AI agent is worth setting up. ### The Website Redesign Guide URL: https://abeldawit.com/resources/website-redesign-guide Type: Guide Price: Free with a newsletter sign-up Length: 23 pages Use Claude Design to set up your brand once, then redesign your current website from it. What's inside: - How to set up your brand assets in Claude Design - How to publish your design system so every design uses it - How to start the redesign from your current website Who it's for: Business owners and creators who want a better website without starting from a blank page. ## Newsletter Scale on Autopilot: Practical automation tips and workflows for business owners who want to scale on autopilot. Free, every week. https://abeldawit.com/newsletter ## Services Paid 1:1 sessions on one focused topic or service. Pick the session that fits and book it on Topmate. https://topmate.io/abeldawit A free 30-minute call for general questions, introductions, and partnership conversations. https://calendly.com/abeldawit/30min ## Articles ### The Checklist Big Companies Use Before They Trust AI, Shrunk Down for a Small Business URL: https://abeldawit.com/articles/ai-automation-checklist Published: 2026-10-02 In healthcare and finance, AI doesn't go live because the demo looked good. Here's the checklist those teams use, cut down to ten minutes per automation. In healthcare and finance, an AI system doesn't go live because the demo looked good. It goes live after someone answers a short list of unglamorous questions. What goes in? What does good look like? Who checks it? How will we know when it breaks? Most small business automations skip all of them. That's usually why they quietly stop working three weeks in, and nobody notices until a customer does. You don't need a compliance team to borrow the list. Here's the version I'd use for a one-person business. It takes about ten minutes per automation. ## 1. Write Down What Goes In and What Comes Out One sentence each. "In: a customer email. Out: a label and a draft reply." If you can't write those two sentences, it isn't ready to build. Big teams call this a spec. Yours can be a note on your phone. ## 2. Save Ten Real Examples, Including Ugly Ones Demos always use the clean example. Real inputs have typos, forwarded threads, three questions in one email, and the customer who writes in all caps. Pull ten from your actual inbox or files, and make at least three of them messy. Larger teams call this a test set. If I could keep only one step on this list, it would be this one. ## 3. Decide What "Good" Means Before You Look Write it down first, or you'll talk yourself into whatever the AI produced. For a reply draft, "good" might mean: uses only facts from the price list, under 80 words, never promises a date. Run your ten examples and score them. If more than one fails, fix the prompt and run all ten again. Not just the one that failed. ## 4. Put a Person Where a Mistake Costs You AI drafts, you approve. That's the default for anything that reaches a customer or moves money. Enterprise teams call it "human in the loop." For you, it's just reading before you hit send. Loosen it later, after weeks of outputs you didn't need to change. ## 5. Make It Tell You When It Fails The worst automations fail silently. A login expires or an app updates, and the whole thing just stops. Zapier and Make can both notify you when a step fails. Turn that on before anything else. ## 6. Read Five Outputs Every Friday "Monitoring" sounds like a big-company word. Here it means five minutes on Friday, reading five things the automation made that week. AI models get updated, and the same prompt can start behaving differently. This is how you catch it early. ## 7. Know How to Turn It Off Write down how to pause it, and who does the work if you do. "Me, by hand, the old way" is a fine answer. Just know it before you need it. ## Copy This for Every Automation ``` Automation: In: Out: Good means: Ten test examples saved in: Passed: / 10 Who approves before it goes out: How I'll hear if it fails: Friday check: yes / no How to pause it: ``` Questions: Q: How do I test an AI automation before using it? A: Collect ten real examples, including at least three messy ones. Write down what a good result looks like before you run them. Then score every output, fix the prompt, and run all ten again until it passes. Q: Why do AI automations stop working? A: Usually because they were tested on clean examples, nobody reads the output, or something changed, like an expired login or an updated AI model. Without an error alert and a weekly check, nobody notices until a customer does. Q: Should AI automations run without any human review? A: Not at first. Have a person approve anything that reaches a customer or moves money. Loosen that only after weeks of outputs you didn't need to change. ### How Healthcare Keeps Patient Data Safe With AI, and the Version Your Business Needs URL: https://abeldawit.com/articles/keep-customer-data-safe-with-ai Published: 2026-10-02 I spent years working with patient data under strict privacy rules. Four of those habits work just as well when you paste customer details into ChatGPT. Before I built AI systems, I was a clinical data scientist in academic radiology, working with patient imaging data under strict privacy rules. You learn quickly that the useful question isn't "is this tool safe?" It's "what's the least I can give it and still get the job done?" Most small businesses never ask that. Customer names, emails, order histories, sometimes worse, all get pasted into a chat window because it's faster than cleaning them up. Usually nothing bad happens. But "usually" isn't a policy. Four habits from healthcare carry over almost directly. ## 1. Give It the Minimum US health privacy rules have a principle called "minimum necessary." You use only the data a task actually requires. It translates perfectly. If you want help replying to an upset customer, the AI needs what they said and what you can offer. It doesn't need their name, email, phone number, or order history. Cut those before you paste. ## 2. Swap Names for Roles In research, data is de-identified before anyone analyzes it. Your version takes ten seconds. Before: ``` Sarah Jenkins (sarah.j@example.com, 555-0142) ordered the large planner on March 3 and says it arrived damaged. She wants a refund. ``` After: ``` A customer ordered the large planner last week and says it arrived damaged. She wants a refund. ``` The reply comes out just as good, and nothing personal left your computer. Add the name back when you send it. ## 3. Know Where It Goes In healthcare, nobody touches patient data without knowing where it's stored, who can see it, and how long it's kept. For your AI tools, that's one settings page. Check whether your chats are used to train the company's models, and turn that off if you can. See how long chats are kept and how to delete them. And look at what your plan covers, because business and team plans usually come with stronger privacy terms than personal ones. ## 4. Decide Who Can Use What Hospitals give people access to the data their job needs and nothing more. A three-person team needs the small version: which tools are approved, which apps can be connected, and what never goes in. Write it down once. Here's a one-page policy you can copy: ``` Our AI Rules Approved tools: [for example, ChatGPT Team or Claude] Never paste: card numbers, passwords, government ID numbers, health information, or anything a customer told us in confidence. Before pasting customer messages: remove names, emails, phone numbers, and addresses. Connected apps: only [list]. Ask before connecting anything new. Anything AI writes to a customer gets read by a person before it's sent. Questions go to: [name] ``` ## What This Isn't This isn't legal advice. If you handle health information, financial records, or data about children, the rules where you operate may ask for more. That's worth a conversation with someone who knows them. Questions: Q: Is it safe to paste customer information into ChatGPT or Claude? A: It depends on your plan and settings. Remove names and contact details first, check whether your chats are used to train the company's models, and never paste card numbers, passwords, or government ID numbers. Q: What does "minimum necessary" mean? A: It's a principle from US health privacy rules: use only the data a task actually needs. For AI, it means cutting names, contact details, and history the AI doesn't need before you paste anything in. Q: Does a small team need an AI policy? A: A short one helps. One page listing the approved tools, what never gets pasted, which apps can be connected, and that a person reads anything AI writes to a customer covers most small teams. ### What to Automate First in a Small Business, From Someone Who Builds AI Systems for a Living URL: https://abeldawit.com/articles/what-to-automate-first Published: 2026-10-02 Skip the flashy stuff. The new-hire test finds the work AI can take off your plate this week, plus the task log and starter prompt I'd use. Most small business owners start automating in the wrong place. They pick the task they hate most, usually sales emails or social captions, and hand it to AI. It comes back sounding like everyone else's, and they decide AI isn't for them. Big companies almost never start there. In my work building AI systems for healthcare and finance companies, the first projects are boring on purpose. High volume, done the same way every time, and easy to check. That's not a lack of imagination. It's where AI pays off first and fails safely. You can borrow the same approach with a spreadsheet and an afternoon. ## The New-Hire Test Ask this about every task in your week: could you hand it to a new hire on their first day, with one page of instructions? If yes, AI can probably do it now. If you'd need to sit next to them for a month before you trusted them with it, AI isn't ready for it either. That second pile is your judgment, your relationships, and your taste. Keep it. This one question filters out most of the flashy ideas. Writing in your brand voice? That's a month of training. Sorting the inbox into pricing questions, booking requests, and complaints? Day one. ## Keep a Task Log for One Week Before you open any AI tool, write down every task you repeat. Five columns are enough. Here's what it might look like for a small service business: | Task | Times a week | Minutes each | One page of instructions? | If AI gets it wrong | |---|---|---|---|---| | Reply to "are you available?" emails | 15 | 4 | Yes | Wrong date, caught before sending | | Turn call notes into a follow-up email | 6 | 10 | Yes | A missed action item | | Write the weekly Instagram caption | 3 | 20 | No | An off-brand post | Multiply the first two number columns to get minutes per week. Cross out anything that fails the new-hire test, or where a mistake would reach a customer before you saw it. The biggest number left is your first automation. In this example it's the availability emails, at an hour a week. ## Do It by Hand First Don't start with Zapier. Start with a chat window and a written prompt, and run it on real examples until you'd send the result with only light edits. For the inbox example, a starting prompt looks like this: ``` You sort customer emails and draft replies for [business name], [one line on what you do]. For each email I paste: 1. Label it: booking, pricing, complaint, or other. 2. For booking and pricing emails, draft a reply using only the facts below. If the answer isn't in the facts, write "Let me check and get back to you today." Never guess. 3. For complaints and "other", don't draft anything. Write one line on what the customer wants. Facts: - Hours: [...] - Prices: [...] - How to book: [...] Tone: warm and short. No exclamation points. Sign off as [your name]. ``` Two lines in that prompt do most of the work. The facts list stops it from inventing a price. And telling it *not* to answer complaints keeps the risky emails with you. ## Connect Your Apps Last Once the prompt holds up on twenty or so real emails, it's worth connecting it to your inbox with a tool like Zapier or Make. Not before. Most automations that break were connected first and tested later. ## What I'd Leave Alone for Now Anything that sends money. Anything that talks to a customer without you seeing it first. Anything that failed the new-hire test. Those can come later, once the boring stuff has earned your trust. Questions: Q: What should a small business automate first with AI? A: Work you could hand to a new hire on their first day with one page of instructions, that you do many times a week, and where a mistake gets caught before it reaches a customer. Sorting and drafting replies to common emails is a typical first pick. Q: Do I need Zapier or coding to start automating with AI? A: No. Start in a chat window with a written prompt and real examples. Connect your apps with a tool like Zapier or Make only after the prompt holds up on about twenty real cases. Q: What shouldn't a small business automate with AI? A: Anything that moves money, anything that reaches a customer without a person reading it first, and anything you'd need a month to train a new hire on, like your brand voice or pricing decisions.