I kill bad ideas fast. That is a feature, not a bug. Here is every major project — the ones that flew and the ones that died.
Memory support for daily tasks — built for cognitive challenges. 40% improvement in task completion. Took 9 weeks instead of the 6-month estimate from traditional development.
The challenge was not building the interface. It was understanding what "support" actually means when someone is struggling with executive function. I spent more time on user research than on prompting AI.
50+ stakeholders on a banking merger with no clear communication strategy. Plotted on Mendelow's Grid with auto-generated comms plans. Used on a $50M+ project with 73 stakeholders.
The breakthrough was realising that most stakeholder conflicts are not about disagreement — they are about mismatched expectations of who decides what. The tool makes that visible.
Feature prioritisation driven by the loudest voice in the room, not data. Now teams compare up to 10 features with Reach, Impact, Confidence, and Effort scores.
The hardest part was getting teams to agree on what "Impact" means. The tool does not solve that — it forces the conversation by making the definition visible before the scoring starts.
Drafting diplomatic emails took 2 hours because turning frustration into professionalism without burning bridges is an art. Multiple tone options now cut that time significantly.
I built this after sending an email I thought was professional but was read as passive-aggressive. The tool offers 5 tones, but the real value is showing you how the same message lands differently.
Portfolio needed to prove AI-augmented execution capability. 16 pages, 54 tools, 4 Mission Control mockups, 8,000+ words. Built and iterated in days, not quarters.
This site is the case study. Every decision — from the aurora background to the "killed" section — was deliberate. It proves that AI + human strategy = world-class output at fraction of traditional cost.
Teams guessing velocity based on "how they feel." Now visual trends, rolling averages, and completion forecasting based on actual data.
The insight: most velocity problems are not capacity problems. They are scope problems disguised as speed problems. The calculator makes that visible.
I kill ideas in 1-2 days. That is cheaper than shipping something no one needs.
Too subjective. What counts as "healthy" varies by team and context. No universal rubric meant the score was always argued, never trusted.
Lesson: Metrics that require interpretation are not metrics. They are opinions with numbers attached. I killed it before anyone got attached to the dashboard.
Edge cases exploded: timezones, recurring meetings, overrides, room bookings, dietary restrictions for lunch meetings. Complexity exceeded value by an order of magnitude.
Lesson: Calendar scheduling is a solved problem by Google and Microsoft. Building a worse version because "AI can do it" is not strategy. It is vanity.
Teams gamed the metrics. Optimised for the score instead of actual quality. A classic Goodhart's Law failure.
Lesson: When a measure becomes a target, it ceases to be a good measure. I should have known this on day one. I did. I built it anyway to prove the point to myself.
Fully automated requirements generation from a one-sentence prompt. Sounded amazing. Produced grammatically perfect, strategically useless documents.
Lesson: Requirements are not documents. They are shared understanding. AI can format understanding. It cannot create it from a prompt. The human still has to think.