AI Generalist Playbook

Let’s lock onto one workflow: a repeatable process that starts with a clear input and ends with a useful output. For Nadia, 34, who owns a busy restaurant, the workflow can’t depend on perfect information or extra staffing. It must run on what she already has: menus, reservation notes, supplier emails, and customer messages.

The reason this works is simple: you reduce risk by narrowing the scope to one job your business already does. AI becomes a “second set of hands” inside that job, not a new business department. When you scope the workflow correctly, you can test it in a week, learn from failure without burning cash, and improve it without changing your whole operation.

Here’s The One-Workflow Sprint Map, built for speed and clarity. Use it to pick, scope, and ship your first workflow:

1. Pick the workflow that touches money or time every week - Choose a process that happens repeatedly and affects either revenue (sales, bookings, upsells) or time (quotes, scheduling, answering questions). If it only happens once a month, you won’t learn fast enough.

2. Define the “input bundle” and “output promise” - List what you feed the system (for example: customer message text, reservation details, menu item names) and what you want back (for example: a draft reply, a suggested special, a corrected reservation summary). Write the output in plain business language your team can use immediately.

3. Set a “first version” rule - Build the smallest useful version that still creates value. For a restaurant, that might mean: “Draft replies for common questions” instead of “Fully automate customer service.”

4. Choose one measurement you can check weekly - Pick one success signal tied to your output promise. Measure it consistently. If you can’t measure it weekly, you probably can’t manage the workflow.

5. Create a safety boundary - Decide what the AI must not do. For example: “Never quote pricing from memory; always ask the manager if the customer asks for a discount.” Safety boundaries stop you from shipping confident nonsense.

6. Plan the sprint: 5 working days to a usable version - You don’t need a “perfect system.” You need a version you can run with real inputs for a week and improve based on results.

The differentiator here is that you scope the workflow around a specific weekly reality. Nadia doesn’t have time for a sprawling automation project. She needs a workflow that improves a weekly bottleneck, like handling customer questions or turning slow reservation inquiries into booked tables.

---

Putting the One-Workflow Sprint Map to Work (Nadia’s Restaurant Example)

Nadia runs a restaurant where customer questions show up everywhere: phone messages, reservation emails, and short texts from people asking about allergies, parking, and whether she can accommodate a birthday. Her biggest frustration isn’t that customers are “hard.” It’s that she answers the same questions repeatedly and still misses details during peak hours.

She chooses one workflow: Draft replies to customer questions about reservations and menu needs. The key is scoping it so she can ship a first version quickly and keep control.

Follow these steps to build your own scoped workflow.

Step-by-step build (use this exact order)

1. Write the workflow in one sentence - “When a customer sends a message asking about reservations, allergies, or special requests, I want a draft reply I can send or quickly edit.”

2. List the input bundle you already have - Start with what you can collect today without extra software: - Customer message text (copy/paste) - Any known reservation details (date, time, party size) - Your menu item names (as a reference list) - A short “house rules” list (for example: “We can confirm allergens but we cannot guarantee zero cross-contact.”)

3. Lock the output promise (what you want back) - Make the output specific: - A draft reply in your restaurant’s voice - A short clarification question when details are missing - A safety line when customers request something you can’t guarantee

4. Create a “golden response template” for 5 common situations - Pick the questions you answer most often. For Nadia, examples might include: - “Do you have vegetarian options?” - “I have a nut allergy. Is it safe?” - “Can you seat us near the window?” - “We’re celebrating a birthday - do you do anything special?” - “What time do you close on weekends?” - You’ll use these templates as ground truth so the AI stays consistent.

5. Add a safety boundary - Nadia sets rules like: - Never invent pricing, availability, or guaranteed allergen safety. - Always ask for a missing detail instead of guessing. - Include a standard disclaimer line for allergy questions.

6. Run a 1-week “shadow mode” - You generate drafts, but you still review and send manually. - Track how often the draft gets edited, rejected, or needs a manager decision.

7. Improve only one thing after you review the week - If the drafts sound too generic, tighten the restaurant voice. - If the drafts miss reservation details, adjust the input bundle or add a checklist.

Expected outcomes (what “value fast” looks like)

By day 5, Nadia can produce usable drafts for most messages. By the end of week 1, she should know: - Which question types the drafts handle well - Which ones need extra rules or missing info - How much time she saved during peak hours (even if it’s rough)

Quick checklist (actions you can take today)

• Choose one workflow that repeats weekly (not monthly). - Define your input bundle and output promise in plain language. - Write a safety boundary for what the AI must not do. - Create 5 golden response templates for your most common customer questions. - Run shadow mode for 7 days and review every draft you generate. - Pick one weekly measurement and track it consistently.

---

What to Watch For: Common Mistakes That Break Your First Workflow

Even when you start with a great workflow, a few failure patterns show up fast. Fixing them early prevents you from burning time and losing trust in the process.

Scope creep kills speed Do this: Keep your first version narrow. Generate drafts for a limited set of question types (for Nadia: allergy questions and reservation clarifications), then expand later. Not this: Automate “everything customer-related” including complaints, refunds, and special catering requests on day 1. You’ll drown in edge cases and lose control.

Weak input bundle leads to confident errors Do this: Provide the AI with a clean reference list (menu item names, house rules, and a short “what we can promise” list). If you don’t know something, you should ask. Not this: Paste random notes, partial messages, or outdated menu text and then expect the AI to “figure it out.” You’ll get replies that sound right but contain wrong details.

No safety boundary makes you uneasy to use it Do this: Write explicit “must not” rules (no invented pricing, no guaranteed allergen safety, no availability claims). If a customer asks for something outside your rules, you trigger a manager follow-up. Not this: Let the AI respond freely because “it sounds helpful.” You’ll spend time second-guessing every message, which defeats the point.

---

Your Takeaway and the Next Move

Your first AI workflow should feel like a controlled experiment, not a leap of faith. When you use The One-Workflow Sprint Map, you stop guessing and start building something measurable: one input bundle, one output promise, one safety boundary, and one weekly check you can repeat.

If you want a simple starting point, Nadia’s example shows the pattern: pick a workflow that repeats, draft outputs you can review quickly, and ship a first version that saves time without creating risk.

Next, you’ll turn your choice into a concrete sprint plan: we’ll map your week, define your exact “golden templates,” and set up your shadow-mode review so you can improve the workflow using real results instead of opinions.

End of chapter one. 4 more chapters in the full book.