How It Works
The Process Behind Every Tool on This Site
Not magic. Not automation. A repeatable process for turning ideas into tested solutions in days.
Identify
Define
Prototype
Test
Ship or Kill
Identify the Problem
Not "we need a feature."
A real pain point you face as a delivery practitioner.
Example from my work:
"Sprint planning takes 4 hours because we can't split oversized user stories consistently."
The test: Can you explain the problem to a colleague in one sentence?
If you can't, you don't understand it yet.
Define Success
What does "solved" look like?
Not vague aspirations. Specific outcomes.
Example:
"A tool that takes a large user story and suggests 3-5 smaller stories using proven splitting patterns. Output should be copy-pastable to Jira."
The test: Would you know success if you saw it?
If not, refine your definition.
Prototype with AI (1-2 days)
Tools I Use:
- Perplexity Pro — Academic research, framework discovery ($20/month)
- Claude Code — Primary building tool ($20/month)
- Claude CLI — Command-line interface for rapid iteration (free)
- Gemini AI Pro — Comparative reasoning + NotebookLM for ingesting many documents at once ($20/month)
Total cost: ~$60/month. Not expensive. Just strategic.
What I Ask AI to Do (example prompt):
"Build an HTML form where users can input a user story. Show 6 splitting pattern options (workflow steps, business rules, data variations, operations, performance, platforms). When user selects a pattern, generate 3-5 smaller stories based on that pattern. Make output exportable to CSV."
What AI Assists With: Drafting HTML, CSS, JavaScript, and implementation options.
What I Direct: Workflow logic, acceptance criteria, testing, prioritization, and delivery judgment.
This is the same process I used as a Product Owner for years — just faster. I define the workflow, decisions, and success criteria. I test what comes back. I iterate until it is useful. The only difference is that instead of waiting weeks for a development cycle, I can validate a working version in minutes.
Average iterations per tool: 10-20
Time per iteration: 5-20 minutes
Total time: 1-2 days of focused work
Test with Real Users (1 day)
Show it to 3-5 colleagues.
Don't explain it. Just watch.
What to observe:
- Do they understand it without explanation?
- Do they get stuck anywhere?
- Does the output match what they need?
- Do they actually use it or just say "interesting"?
Take notes. Don't defend your choices.
User confusion = design problem, not user problem.
Iterate based on real feedback, not assumptions.
Ship or Kill (immediate)
Ship if:
- Solves the problem
- People actually use it
- Feedback is positive
- Maintenance is manageable
Kill if:
- Doesn't solve the core pain point
- Too complex for the value
- Better solutions already exist
- Requires ongoing maintenance you can't provide
No attachment to ideas. Attachment to outcomes.
Some of my killed projects:
- Backlog health analyzer (too subjective)
- Automated meeting scheduler (too many edge cases)
- Requirements quality scorer (people gamed the metrics)
Killing bad ideas fast is a feature, not a bug.
Real Examples
Ignite Intelligence v15
Problem: No off-the-shelf platform existed for an operator-grade command centre that fulfils my personal requirements and use cases. So I built one — with autonomous agents and structured morning briefings.
- Months 1-2: Architecture from first principles using Claude Code — agent registry, autonomy tiers, model selection
- Months 3-4: Build the orchestration layer (Olympus Chairman, Jarvis dispatch) and core agents (Finance, Opportunity Scout, Brain Coach, Coder, Researcher, Outreach, System Guardian)
- Months 5-6: Real-time telemetry, FastAPI + MongoDB + Qdrant integration, WebSocket realtime channel, scheduled routines
- v15: Production deployment. Multi specialist agents in daily use. Chairman briefing every morning.
Shipped: Multi-agent autonomous platform in production daily use. The skill was systems thinking, not syntax.
User Story Slicer
Problem: Sprint planning wasted 4 hours on story splitting debates
- Day 1: Researched story splitting patterns (Richard Lawrence's work)
- Day 2: Built HTML tool with 6 patterns using Claude Code
- Day 3: Tested with my team
- Day 4: Refined based on feedback
- Day 5: Shared with 5 colleagues outside my team
Shipped: Free tool, available to community.
Stakeholder Map Generator
Problem: 50+ stakeholders on banking merger, unclear communication strategy
- Day 1: Designed Power/Interest matrix framework
- Day 2: Built interactive tool
- Day 3: Tested on real project stakeholders
- Day 4: Added communication plan automation
Used on: large-scale project with 73 stakeholders.
This Website
Problem: Portfolio needs to prove product thinking + AI execution capability
- Week 1: Positioning strategy
- Week 2-3: Content creation across all pages
- Week 4-5: Build using Claude Code
- Week 6: Testing and refinement (300+ iterations)
Shipped: This site you're reading.
The Division of Labor
What AI Assists With:
- Drafts HTML, CSS, JavaScript, and implementation options quickly
- Helps assemble forms, dashboards, workflows, and calculators from defined requirements
- Applies changes when I describe the exact behavior or acceptance gap
- Reduces the technical delay between idea and prototype
What I Direct:
- Decide what features to build
- Prioritize competing stakeholder needs
- Translate business strategy into usable workflows
- Say no to weak feature requests
- Make product decisions and validate the output
AI is a tool. Like Excel is a tool for financial analysts. Like Figma is a tool for designers.
AI is the prototyping tool for Product Owners.
How the work actually gets done
What I use:
Research & reasoning: Perplexity Pro · Gemini AI Pro (NotebookLM)
Build & execution: Claude Code · Claude CLI
What these tools don't do: they don't decide what to build. They don't write the acceptance criteria. They don't decompose ambiguous business need into a structured spec, or hold the architecture in working memory across months of build, or know when to stop.
That's the work. The tools execute against a brief. The brief is mine. The architecture is mine. The judgment is mine. Eight years of $50M+ enterprise delivery exposure across banking, insurance, government, and core platform programs is what shapes the brief these tools receive.
Ignite Intelligence v15 — a multi-agent autonomous operations platform — was architected from first principles by a Business Analyst and Product Owner. 1 hour a day for 6 months. The discipline is Business Analyst / Product Owner. The output is AI infrastructure. Same skill, new surface.
Your Turn
Try this experiment this week:
- Identify one repetitive problem you face as a delivery practitioner
- Spend 1 hour with Claude Code, Claude CLI, or Gemini AI Pro
- Test it with one colleague
- Iterate once — fix the biggest confusion point
- Decide: Ship or Kill. If it works, share it. If it doesn't, try a different problem.
Useful AI-assisted prototypes still need product thinking: know the problem, describe it clearly, define acceptance criteria, test, iterate, and make strategic decisions. That's Product Ownership.
Download the Complete Guide
The AI Execution Guide (PDF)
Complete step-by-step process including:
- • Problem identification framework
- • How to describe what you want to AI (with examples)
- • Testing checklist
- • Ship/kill decision criteria
- • 15 real examples from my work
- • Common pitfalls and how to avoid them