AI Automation Agency LaunchBag

Build small, reliable automations for small businesses, and know which ones to refuse.

Find the repetitive work inside a real business, judge honestly whether it should be automated at all, build the smallest thing that survives its own exceptions, hand it over so the client owns it, and let each project sharpen the next quote.

Real pages from the bag. Open one to read it.
Introductory price $29. $39 starting Oct 25. Introductory price through October 24, 2026. From October 25, 2026, the price is $39. One purchase, yours to keep.

Checkout on Payhip. Your download link arrives by email right after payment. Refundable until you download, final after that (refund policy).

What’s in the bag ↓

What’s in the bag

Numbered in the order you’ll use them. One guide that teaches the decisions in the order they happen, one workbook the projects run on, the documents a client reads and signs, and the prompts that remove real work. Five files, and the tools carry most of the weight.

00Start Here & The Two Gates

PDF · 8 pages

What this business really is, why most of the work is not AI, the two gates, whose accounts everything lives in, and the honest page about what it will not help you build.

  • 00-Start-Here-and-The-Two-Gates.pdf

01The Automation Operator’s Playbook

PDF · 19 pages

The durable part, in the order the decisions happen, from finding a workflow in a real business to reading your own delivery record. No tool tutorials, on purpose.

  • 01-The-Automation-Operators-Playbook.pdf

02The Project Workbook

Excel / Google Sheets · 6 connected tabs

The file the projects run on. The screens gate the estimate, the test set gates the launch, one row follows each project to a year after handover, and everything volatile sits on one dated sheet.

  • 02-The-Project-Workbook.xlsx

03Client & Project Kit

Word / Google Docs · 7 documents

What a client reads or signs, from the audit to the handover, including the data and access agreement and the register that proves the accounts are theirs.

  • 03-Client-and-Project-Kit.docx

04Prompt Patterns

Word / Google Docs · 7 patterns

Prompts attached to specific steps rather than kept as a library. Each says what to check before trusting it, and names what the model must never be allowed to decide.

  • 04-Prompt-Patterns.docx
  • 5files
  • 27PDF pages
  • 1spreadsheet
  • 2Word files
  • Researched September 2026
All 5 file names
  • 00-Start-Here-and-The-Two-Gates.pdf
  • 01-The-Automation-Operators-Playbook.pdf
  • 02-The-Project-Workbook.xlsx
  • 03-Client-and-Project-Kit.docx
  • 04-Prompt-Patterns.docx

All 5 files are in the ZIP; nothing is held back for an upsell.

A closer look inside

Pages and spreadsheet views from the files you download, not mockups. Open any one to read it.

What you’ll be able to do

Each one names the file that does it.

  • Find the work by watching, not by asking

    People answer “what would you like to automate?” with what annoys them, which is not what costs them. An hour of paying attention gives you named workflows with a frequency and a time cost, which is also a deliverable you can charge for.

    Playbook ch. 2 · Client Kit: audit findings · Prompt Patterns 1

  • Say the sentence nobody else in this field says

    Fix the process first. If a workflow only runs the same way 60% of the time, the exceptions are the process, and automating the tidy majority hands the business the mess plus a bill. The screen gives you that verdict, and it is the most valuable thing you can tell a client in month one.

    Playbook ch. 3 · Workbook: Screen & Estimate

  • Decide whether any step actually needs a model

    Kept as a separate question on purpose, because the whole category collapses the two. A model earns its place on three jobs (classify, extract, draft for a person), and on none of them does it take the final action. Most good first projects contain no model at all.

    Playbook ch. 4 · Workbook: Screen & Estimate

  • Know what you should not take on yet

    Anything that moves money. Anything regulated. Anything the client could not do by hand tomorrow if it broke. Refusing work is a skill, and the screen makes it a decision rather than a feeling.

    Playbook ch. 5 · Workbook: Screen & Estimate

  • Price a project from the job instead of a price list

    Build hours, debugging, the running cost the client will pay on their own account, and the support you will still be doing in a year. The workbook will not finish an estimate while support says zero, because that is the line that quietly turns a business into unpaid work.

    Playbook ch. 10 · Workbook: Screen & Estimate

  • Make an AI step reliable enough to sell

    Asking a model for JSON gets you the wrong shape often enough to break things. Enforcing a schema is two orders of magnitude better. Every model step in this bag has the same three parts: a schema, something that checks the output fits, and a defined behavior when it does not.

    Playbook ch. 6 · Prompt Patterns 6–7

  • Prove it before it goes live

    The characteristic failure here is not a crash, it is a confident wrong answer nobody notices for a month. So you write the expected answers before you build, run the exceptions, and a launch gate refuses to clear while a case is failing or an irreversible step has no one approving it.

    Playbook ch. 7 · Workbook: Test & Launch

  • Handle the one attack that matters

    The classic first project reads incoming email and does something about it, which is the exact shape of the top-ranked security risk in this field. You learn to recognize it, and the three defenses that contain it, before you build one.

    Playbook ch. 8 · Workbook: Test & Launch

  • Get the accounts right so you are a supplier, not a hostage

    Production in the client’s accounts and name, development in your own sandbox with made-up data, your access a named seat someone can revoke. Handover means one thing: they could remove you tomorrow and it keeps running.

    Playbook ch. 9 · Client Kit: access register, handover

  • Find out what the work actually paid

    Your build factor, your debugging ratio, your support factor: each one a number you entered divided by another. And the question that decides whether this is a business: do the recurring fees cover the support you are still doing?

    Playbook ch. 13 · Workbook: Project Log, Calibration

Who it’s for

  • You are comfortable learning software but you are not a programmer, and you do not want to become one.
  • You can see that businesses waste time on repetitive work, and you want a way to find it and price it.
  • You have watched a lot of AI-agency content and come away with a stack but no method.
  • You would rather tell a client the truth about a bad idea than sell them a build.

Who it’s not for

  • You want a tool tutorial. There is none here, and the guide explains why.
  • You want to build autonomous agents. This teaches you to recognize that shape and stay out of it for now.
  • You want a price list. There are no prices anywhere in this bag.
  • You want an income figure. There is not one, and the honest page says why.

Six phases, one gate

Start Here gives you the roadmap and every file shows where you are on it. Phase 4 is the gate, clear to build: the workflow passed the screen, the AI line is settled, and the scope is written.

Read the Playbook chapter for the phase, fill the Workbook tab it points to, and move on. The Client Kit is filled in once and used per project; the Prompt Patterns are attached to specific steps rather than kept as a library.

  1. Understand what this business actually is

    You can explain why most of what businesses pay to automate is not AI, and why that is good news for a beginner.

    Start Here (00) · Playbook ch. 1
  2. Set your rules and costs, once

    Your rate, your platform, your model and every number that moves, each with somewhere to check it and the date you last did.

    Workbook (02) Tools & Rules
  3. Audit one real business

    Named workflows with a frequency and a time cost, written up as something you could charge for. Most will fail the screen, and that is the phase working.

    Playbook ch. 2–3 · Client Kit (03) · Prompt Patterns (04)
  4. Clear to build

    One workflow that passed the work screen, whose AI line is settled, whose blast radius you can live with, scoped in writing with a price built from the job.

    Workbook (02) Screen & Estimate · Playbook ch. 3–5, 10
    Clear to build
  5. Build it in a sandbox, prove it, hand it over

    It survives its own exceptions, a person confirms anything irreversible, it runs in the client’s account, and they could remove you tomorrow without it breaking.

    Playbook ch. 6–9 · Workbook (02) Test & Launch · Client Kit (03)
  6. Keep it running and read your record

    Your own build, debugging and support factors replacing the shipped ones, and an honest answer to whether the recurring fee covers the work you are still doing.

    Playbook ch. 12–13 · Workbook (02) Project Log, Calibration

What it doesn’t do

  • No income figures and no prices. Every published figure in this field comes from an agency advertising itself or somebody selling a course.
  • No tool tutorials. They would be the fastest-rotting thing in the bag; you get a dated sheet and a way of choosing instead.
  • No autonomous agents with tool access. That shape is the top-ranked security risk in this field, and it is taught as a boundary rather than a technique.
  • No work that moves money on its own, and nothing clinical, legal, financial-advice or hiring.
  • No legal, tax or security advice. Where obligations attach, the bag says what the rules require and what you must verify for yourself.
  • No fabricated clients, projects or results. There are no invented case studies anywhere in it.
  • And the one that matters most: most workflows you screen should fail. A kit that made every opportunity look good would be selling you the same thing the videos are.

How it was made

  • Researched to a saturation rule rather than a source quota, and stopped when the two findings the product turns on were both primary, quantified and consistent with each other.
  • The load-bearing claims were taken to primary sources: the security guidance, the privacy obligations, the email law and the process-selection research all read at source.
  • The screening system comes from twenty years of process-selection research that predates the current field entirely, including the verdict it never gives: fix the process first.
  • Every volatile fact (platform, model, prices, terms) is isolated on one dated sheet, so a change is a row edit rather than a rewrite.
  • The workbook is generated by a script and tested by 28 independent scenarios computed by hand, plus 85 formulas evaluated in a separate engine.

Made and supported by Frog Labs LLC, the company behind LaunchBag.

More on the process: how a LaunchBag is made.

Before you buy

The questions people ask about this bag. More on the support page.

Do I need to be able to code?

No. You need to understand one shape: something happens, you fetch data, you decide, you act, a person confirms anything irreversible, you write down what happened, and you handle the failure. The bag teaches that shape and only the technical detail that affects whether a client’s automation keeps working.

Which platform does it teach?

None, in the guide. That is deliberate. Tool specifics in this field date in months. The workbook names a sensible starting platform on a dated sheet with the reasoning and the criteria for choosing differently, so when something changes you edit a row instead of discovering the book is wrong.

Is this really about AI?

Less than you would expect, and that is the most useful thing in it. Most of what businesses pay to automate is the plumbing between systems they already own. The bag keeps two questions separate (should this be automated at all, and does any step actually need a model?) because collapsing them is where the money goes.

What will I actually be able to sell?

A workflow audit, a defined first automation, and keeping it running. The bag does not tell you what to charge, because every published figure in this field comes from someone selling something. It does show you how to build a number from the job and correct it from what the work actually cost you.

How much does it cost to run an automation?

It depends on the platform and how much a model is used, and both change. The workbook models the shape (charged per million tokens, output costing more than input) and keeps the actual prices as dated rows you check. That cost is paid by the client on their own account, and the bag is explicit that it is not your margin.

Whose accounts does the automation live in?

The client’s, and this is settled before your first project rather than during it. Production in their accounts on their billing; development in your own sandbox with made-up data and never their live keys. Handover means one specific thing: they could revoke your access tomorrow and the automation keeps running.

What if I break something in a client’s business?

That is what the blast-radius screen and the launch gate exist to prevent. Nothing irreversible happens without a person confirming it, nothing goes live until it has survived the exceptions including the one the client says never happens, and anything they could not redo by hand is not a first project.

Is it just a folder of prompts?

There are seven prompts, and each one is attached to a specific step: turning audit notes into a process map, generating the edge cases you would not think of, drafting a scope. Every one says what to check before you trust it and names what the model must not be allowed to decide. The decisions live in the workbook, not in the prompts.

Does it work in Google Sheets?

Yes. The workbook uses only functions that exist in both Excel and Google Sheets, and that is checked by script. Every formula is evaluated in an independent engine, and every dropdown, validation message and print setting is verified before release.

It is only 27 pages. Is that enough?

The page count is the wrong measure here. The decisions in this business are made in the screens, the estimator and the launch gate, not in prose, and the guide is exactly as long as it needs to be to make those tools usable. Padding it would have made the bag look bigger and work worse.

Get the AI Automation Agency LaunchBag

5 files: 2 PDF guides (27 pages), 1 spreadsheet for Excel or Google Sheets, and 2 editable Word files. One download, yours to keep.

Introductory price $29. $39 starting Oct 25. Introductory price through October 24, 2026. From October 25, 2026, the price is $39.

By clicking “Get the bag” and completing checkout on Payhip, you agree to LaunchBag’s Terms and refund policy. This is a digital download: refundable until you download a file, and all sales are final after that, with the exceptions in the refund policy. Not legal, tax, insurance, or safety advice, and the templates are not a substitute for the advice of an attorney; no income promises.

Sold by Frog Labs LLC. To receive our legal name and business address by email, write to contact@froglabsllc.com.

  1. Pay on Payhip’s checkout page.
  2. Payhip emails your download link to the address you used, right away.
  3. Unzip and open 00 Start Here & The Two Gates. The files are numbered in reading order.
Price $29 Intro price · $39 starting Oct 25
Get the bag