You can ship real software with AI. This is the method that gets it into production, paid for and safe.
AI coding tools will write you a working-looking app in an afternoon. The hard part is everything after: a spec it can build from, a loop that catches its mistakes, data and permissions that hold up with two users, payments that never release a product before the server confirms the money, secrets that stay out of the browser, a deployment you can roll back, and a bill you understand. This book is that method, written for people who don't code, and it works whichever tool you pick: Claude Code, Cursor, Copilot, Codex, Lovable, Bolt, v0 or Replit.
- The full playbook (about 110 pages, phone or laptop) in 20 parts
- How AI coding tools actually work and the 10 ways AI-built software breaks, with the documented incidents
- Choosing your tools: the three families, what each costs as listed in October 2026, a decision table and a lock-in check
- SPEC.md, PLAN.md, the rules file and DECISIONS.md: the four project files, with templates and a filled example
- The Build Loop: plan, slice, prompt (task cards), build, verify, commit, every day
- Data and permissions, a repeatable debugging method, testing like a second buyer, and the security non-negotiables
- Payments done safely (never release before the server confirms the full price), shipping to production, launch, AI features, costs and maintenance
- The 90-Day Build Plan week by week, plus 18 copy-ready templates, prompts and checklists
- Who it's for
- Founders, operators, freelancers and agency owners who are not software engineers and want software to exist without a development team.
- What you'll learn
- One tool-independent method: a SPEC.md the AI can build from, the Build Loop, architecture on a default stack, data and permissions, debugging, testing like a second buyer, security, safe payments, shipping, launch, AI features, costs and maintenance.
- Why it's different
- It is about everything after the demo: the spec, the verification and the ownership that stop an AI mistake becoming an incident, with a 90-Day Build Plan and 18 copy-ready templates.
What's inside
Read this first
What AI can and can't build for you in 2026, vibe coding vs building, what 'done' means, the method on one page.
How the tools work, and how they fail
Models, context, agent loops, hallucinated packages, context rot, autonomy and guard rails, the failure catalogue.
Choosing your tools
App builders vs AI code editors vs coding agents, prices as listed in October 2026, a decision table, billing models, the lock-in check.
Choosing what to build
Four kinds of software a solo builder can ship, validation before code, scoping v1, kill criteria, an idea scorecard.
Writing the spec
Why a spec beats a long prompt, the SPEC.md template, acceptance criteria as checks, edge and abuse cases, the spec review.
Architecture for non-engineers
What runs where, the default stack and when to deviate, decisions you make once and write down.
Setting up your workshop
Accounts, folders, Git, environment variables and the rules file before the first prompt.
The Build Loop
The daily method: plan, slice, task cards, build, verify, commit; reading a diff with five questions.
Reading code without being a programmer
Enough literacy to find your way around a project, read an error and spot nonsense.
Data, users and permissions
The data model, roles, row-level access rules, and testing them with two users.
Debugging method
A six-step loop for when it breaks or the AI goes in circles, with Git as your safety net.
Testing and quality
Manual test scripts from acceptance criteria, automated tests, the second-buyer pass, release rhythm.
Security: the non-negotiables
Secrets, server-side authorisation, the five common holes, dependencies, prompt injection, the pre-launch review.
Payments and pricing
Processor vs merchant of record, fee maths, the safe payment flow, webhooks, refunds, the loophole checklist.
Shipping to production
Plan limits, domains, live keys, email deliverability, monitoring, the launch-day checklist and rollback plan.
Launch and first paying users
Putting the product in front of people who will pay, honestly and without promises.
Adding AI features to your product
Calling a model from your own app without blowing the budget or the security.
Costs, maintenance and when to hire a human
COSTS.md, free-tier traps, keeping it running, and the moments to pay an expert.
The 90-Day Build Plan
Thirteen weeks from idea to a live product, with checkboxes.
Templates, prompts and checklists
18 copy-ready templates: spec, rules file, task cards, reviewer prompts, security review, webhook logic, rollback plan.
Every technique inside, by name
The 46 named techniques, frameworks and tools this book teaches, taken from the book itself. £500.00 one-time, instant PDF download.
- The Build Loop: plan, slice, prompt, build, verify, commit
- SPEC.md: the spec an AI can build from and you can test against
- PLAN.md: ordered slices, each with a "done when" line
- The rules file (CLAUDE.md, AGENTS.md or your tool's rules)
- DECISIONS.md: the dated log of choices and why
- The four project files
- Task cards: one slice per session
- Vertical slices, not horizontal layers
- Acceptance criteria as checks you can perform without reading code
- The spec interview prompt
- Edge cases and abuse cases at spec time
- Context rot and the fresh-session rule
- Models, tokens and context windows
- Chat versus agent
- The failure catalogue: ten ways AI-built software breaks
- The three tool families: app builders, AI code editors, coding agents
- The lock-in check
- The tool choice table (prices as listed in October 2026)
- Four kinds of software a solo builder can ship
- Start one row higher than your ambition
- The idea scorecard, the "not now" list and kill criteria
- What runs where: browser, server, database, third-party services
- The default stack: Next.js on Vercel, Supabase, Stripe or a merchant of record, Resend, GitHub
- The architecture decision table
- Reviewing a diff as a non-coder: the five questions
- Row-Level Security by owner or organisation ID
- The two-user test
- The six-step debugging method: reproduce, isolate, read, hypothesise, fix, confirm
- The debug report prompt
- Testing ranked by the cost of being wrong: money paths first
- Manual test scripts from acceptance criteria
- The second-buyer pass: double submit, parallel attempts, retries
- The fresh-session code review and security review prompts
- Secrets on the server, never in the browser or the repo (GitHub push protection)
- Processor versus merchant of record
- The safe payment flow: server-set price, access only after the webhook confirms the full amount
- Webhooks done right: verify the signature, store processed event IDs, check amount and currency
- The payments loophole checklist
- Free-tier traps and upgrade triggers (Vercel Hobby is non-commercial; Supabase Free pauses)
- The release checklist and the one-action rollback plan
- Launch in three rings and the ring-one message
- Call the model from the server, never the browser
- Cost per request arithmetic and prompt caching
- COSTS.md: the monthly cost sheet
- The maintenance calendar and the hiring brief
- The 90-Day Build Plan
Last updated:
Is it for you?
Who it's for
- Founders and operators without a technical co-founder who want to build their own product instead of waiting for one.
- Freelancers and agency owners who want to deliver small software and automations for clients with AI tools.
- People who have already built something with Lovable, Bolt, Cursor or Claude Code and are stuck between 'it works on my screen' and 'people pay for it'.
- Anyone who wants one method that survives switching tools.
Not for you if
- You want a book about one tool's buttons. The book teaches a method and uses the tools as examples; the tools' own docs cover their menus.
- You are a professional engineer looking for advanced engineering. The book is written for non-engineers; you will know most of it.
- You want a promise of income or a guaranteed launch. The book gives you the method and the checks; results depend on the problem you pick and the work you do.
Who's behind this
Build Software With AI is written by Manthan Bhendkar, founder of Asendify, whose playbooks take one decision, strip out the hype, and give you the numbers and a method.
He is not a trained software engineer. He built and runs his own products mostly with AI: an ebook shop with three payment gateways, a cold-email software product, a Notion-like workspace app, an AI assistant on the Claude API and a set of paid web tools, on a standard stack of Git, Vercel and Supabase. The book is the method he uses, including the mistakes that only a second buyer or a loophole check would have caught.
Checked against the sources
The plan limits, security rules and payment rules in the book come from the providers' own documentation and pricing pages, as listed in October 2026. Four of them:
Vercel's Hobby plan is non-commercial
Vercel describes its free Hobby plan as for personal, non-commercial use, capped, with no way to buy more usage; production for a business needs Pro ($20 a month as listed). The book's upgrade trigger: before the first paid customer.
Row-Level Security is yours to get right
CVE-2025-48757 (May 2025) recorded insufficient Row-Level Security policies in Lovable-generated apps letting unauthenticated users read or write tables. Whatever the tool, the book makes you enable and test RLS on every exposed table with two users.
Push protection is off for private repos
GitHub's push protection blocks known secret formats going to public repositories by default; for private repositories it is off unless you have GitHub Secret Protection. The book's rule: rely on neither, and keep secrets out of the repo entirely.
Verify the webhook signature first
Stripe's documentation: without verification, an attacker could send fake webhook events to trigger actions like fulfilling orders or granting access. Every provider's webhook gets the same treatment in the book, before anything else happens.
Not just SaaS
The worked example is a client-approval portal, but the method (spec, loop, verify, own) is the same for the other three kinds of software a solo builder ships:
An internal tool
A lead tracker that replaces the spreadsheet for a two-person business. One user you can ask, no payments, your own data: the place the book tells first-time builders to start and learn the Build Loop.
Client work
Portals and tools for agency clients: scope by slices with their own price, put in writing who owns the code and the keys, and price in a qualified human review for anything touching payments or sensitive data.
An automation
A weekly report pulled from three sources and emailed on Monday. It runs unattended, so the method is about noticing when it fails: the "done when" line, logging and the maintenance calendar.
AI inside an existing business
Part 17's patterns for adding a model to a product you already run: call it from the server, start with fixed-input features, do the cost-per-request arithmetic before you build, and keep a human on anything that acts.
Questions
Which AI tool does the book use?
None in particular. It covers the three families (app builders like Lovable, Bolt, v0 and Replit; AI code editors like Cursor, Copilot and Windsurf; coding agents like Claude Code and Codex) with October 2026 prices as listed, then teaches one method that works in any of them. Part 3 helps you pick one in an afternoon.
Do I need to know how to code?
No. Part 9 teaches just enough code literacy to find your way around a project, read an error and spot nonsense in a change. You will not write code; you will specify, direct, verify and decide.
Why is it £500?
It is a complete method, not an introduction: 20 parts, about 110 pages, the 90-Day Build Plan and 18 copy-ready templates covering specs, the Build Loop, security, payments, deployment, costs and maintenance. One avoided payment loophole or leaked key costs more than the book.
Is there a guarantee?
No. Software outcomes depend on the problem you pick and the work you do, so the book makes no promise of results or income. Each copy is licensed to one person and watermarked.
Will it go out of date?
Tool names and prices will; the method will not. Prices and limits are labelled 'as listed in October 2026' and the book tells you to check the provider's page. Specs, the Build Loop, verification, permissions, safe payments and rollback do not change with the tool.
How do I get it?
Right after you pay, a download button appears on this page, and the link is emailed to you. The link works for 30 days and up to 10 downloads.
Why is my name on the PDF?
Each copy is licensed to one person. Your name and a unique licence ID are printed in it, which stops copies being passed around and resold.
How do I know a payment request is really from you?
We only take payment through this checkout or an invoice emailed from an @asendify.co address. Our only official WhatsApp is +91 90960 82894. We never ask you to pay a personal account, UPI ID, crypto wallet or gift card. If anyone does, don't pay, and email montybhendkar@gmail.com.
Get the book
The method, the four project files, the Build Loop, security, payments, deployment and costs, with a 90-Day Build Plan and 18 templates.
Build Software With AI is general information about building software with AI coding tools, accurate to the author's knowledge as of October 2026. AI tools, their features, terms and prices change often; prices and limits given are as listed by providers at that date and examples are hypothetical. AI-generated code can contain errors and security flaws: test and review all code, and have qualified people review anything that handles money, personal data or regulated activity, before you rely on it. It is not legal, security, tax or financial advice, and no result, income or outcome is guaranteed.