Spec-Driven Development for Tech Teams: Ship Real Apps with BMAD
Dheeraj Sharma
AI systems builder & educator
Make AI agents build what your team decided, in new code and existing code
AI coding agents are fast, and they guess. Every decision nobody wrote down, the agent makes for you, and the team finds out in review or in production.
Spec-driven development fixes the order: decide first, write it down, then let the agent build from it. This course teaches it in two steps. First by hand with Claude Code, so you see the loop. Then with BMAD, a free, open source method that interviews you like a product team, splits the work into stories, builds each one in a fresh chat, reviews it, and looks back at the end.
You watch one real app go from a rough paragraph to 7 stories, 524 passing tests and a retrospective. Then the same loop goes into an existing codebase: a small change in about 3 minutes, and a new feature built from a spec that read the code first.
I have spent 20 years building software platforms for enterprises. Every number here comes from one recorded run, so you can check it on the kit yourself.
What you’ll learn
Go from prompting and fixing to a loop your team can repeat: idea, spec, stories, tested code and review, in new and existing code.
Spot the five problems vibe coding creates, so you can trace each bug back to a decision the AI made for you.
Organise a project around foundation files and one spec per feature, approved before any code exists.
Turn a prompt you type twice into a Claude Code skill, and run the same loop in code that already exists.
Pressure-test a one-paragraph idea with BMAD, one question at a time, so the rules nobody wrote down get decided.
Write a five-field spec (why, capabilities, constraints, non-goals, success signals) with every open question closed.
Decide when a PRD is worth writing, using the test from BMAD's own docs.
Order stories so each needs only the ones before it, so a team can build them one at a time without collisions.
Pin shared decisions (stack, port, database, data shape) in an architecture spine before code exists.
Run every build with a six-line checklist and read what the agent flags before you approve it.
Sort review findings into fix now, defer or decide, and catch the repeated finding that points to a gap upstream.
Run a thorough review and a retrospective that checks the whole epic against the spec.
Prove the app with test days you chose before running them, and leave with a report your team can read.
Size the work first: skip BMAD, go straight to a build, or plan an epic.
Ship a small change in one sentence, then read the diff for changes nobody asked for, and build a feature from a spec.
Compare BMAD with Spec Kit, OpenSpec and plan mode on one feature, so you can choose for your team (coming soon).
Learn directly from Dheeraj
Dheeraj Sharma
AI systems builder & educator. 20 yrs enterprise software, 18K+ audience
Who this course is for
Developers and tech leads who get working code from AI agents that still misses what was decided, and lose the time to review and rework.
Product managers, product owners and analysts whose job is now the spec, and who want agents to build from it without guessing.
Technical founders building their own product with AI, who need it to hold up past the demo, story after story, in one codebase.
Prerequisites
A paid Claude plan (Pro or Max)
Every lesson runs in Claude Code. My full run used a Max plan with no limit hit; on Pro, expect to pause between stories.
Claude Code, in the desktop app or a terminal
No code to write. The lessons show a terminal. In the Claude Code desktop app, Claude runs the same commands for you.
A laptop where you can install software
You install Node.js, Git and uv. Some work laptops block installs, so a personal laptop can be easier.
What's included
25 video lessons, about 4 hours 30 minutes
Five modules: spec-driven development by hand with Claude Code, then BMAD from a rough idea to a reviewed epic, then BMAD in an existing codebase. A bonus lesson comparing BMAD with Spec Kit, OpenSpec and plan mode is coming soon.
The course kit, with 22 saved states
The starting files, the finished app, and a saved state at every step of my run (15 for the new app, 7 for the existing one). If your run drifts, load the state and rejoin where the lesson starts.
5 hands-on exercises
One per module, on your own copy of the kit or your own project. Each has a "done when" list, so you know when the module is finished.
Monthly live group call
One call a month with me for everyone enrolled. Bring a spec, a stuck story or your own codebase. The first call is Tuesday 17 November, 12:00 ET, and recordings stay in the course.
Lifetime access
Keep the lessons, the kit, every call recording and the comparison lesson when it lands. When a BMAD release changes a step, the lesson notes say what changed.
Certificate of completion
Finish the five exercises and get a Maven certificate to share on LinkedIn or with your employer.
Maven Guarantee
Your purchase is backed by the Maven Guarantee.
Free resource
Build an SEO + AEO Research Agent in 45 Min (No Code)
01 - Get Cited by ChatGPT, Claude AI
Assemble a working agent with no code, then read its citation_opportunities so ChatGPT, Perplexity, and Claude cite you
02 — Turn raw keyword data into a content strategy
Write two short context files (your business + your content strategy) so every keyword is filtered through your audience
03 — Wire one MCP that replaces your SEO tool stack
Connect Tavily as a single MCP server and get live web search, SERP analysis, and deep research in one tool
The course outline: 5 modules, 25 lessons, 5 exercises
Module 0: Spec-Driven Development by Hand (about 1 hour)
Vibe coding, and the five problems it creates
What a spec is, and how a spec-first project is organised
A new project, spec first: the foundation and the first feature
The second feature, and turning a repeated prompt into a skill
Spec-driven development in existing code, and where frameworks fit
Exercise: one feature, spec first, no framework
Module 1: Decisions Before Code (about 57 min)
The finished tracker, and what it cost me
Install BMAD and read what landed on disk
From spec driven development to BMAD: one loop, five fields
The interview: from a rough paragraph to a spec
Recap: module 1 in four points
Exercise: your spec
Module 2: From the Spec to the First Story Built (about 55 min)
Too much for one request: stories and the fresh chat
Break it into stories, then write what they must share
What Build does before it writes, and who checks it after
Tracker v2: story 1, built and checked
Recap: module 2 in five points
Exercise: the spine and story 1
Modules 3 and 4: finish the epic, then existing code
Module 3: Finish the Epic, and Know When to Stop (about 48 min)
The rest of the stories, on a clock
Thirty findings a story: which ones get fixed
Prove it: a thorough review, the retrospective, and three test days
Recap: module 3 in four points
Exercise: the whole epic, proven
Module 4: Small Changes and Code That Already Exists (about 66 min)
The sizing rule: enter the loop where the work needs it
A small change in existing code, straight to Build
What the code cannot say: project context
A feature in existing code: the streak freeze
Recap: module 4 in four points
What you have, what breaks next, and the next step
Bonus: BMAD vs Spec Kit vs OpenSpec vs plan mode (coming soon)
Exercise: existing code
Frequently asked questions
Maven for Teams
Reimbursement
Get your company to pay
Everything L&D needs: email template, receipts, and certificate of completion.
Get reimbursedPrivate cohort
Run a cohort for your org
A dedicated cohort with a custom schedule and curriculum, tailored to your team.
Book a private cohort