Philosophy
How I think about AI
AI is easy to demo and hard to live with. A good answer in minutes is not the same as a workflow your team can run every week. This page states how I think about AI, so you can judge whether my approach fits your company.
1. The curse of the learning curve
Simple AI output is easy. Durable operational leverage is not. It takes sustained learning and effort, and your staff cannot give that effort while doing their actual jobs. That gap is the curse of the learning curve: the tool is easy to start and hard to live with.
Angora method: many AI stalls in a small office are not technology problems. They are ownership problems. No owner, no list, no due date, no weekly report, and a new tool added without fixing the gap between systems. The workflow never becomes a habit, so the team drifts back to the old way.
McKinsey's March 2025 survey of nearly 1,500 organizations found workflow redesign had the strongest association out of 25 attributes tested. That is consistent with my method: the work is redesigning the workflow, not buying a tool.
How I deliver it: Map measures the workflow. Build builds it inside tools you already own. Care checks it. Your staff use it. They do not have to learn AI, and they do not have to become AI consultants.
2. Accuracy over everything
Efficiency can only be built on accuracy. That is the rule I build by.
Angora method: a fast wrong answer costs more than a slow right one, because every downstream step inherits the error. So I treat accuracy as the gate, not a feature. Nothing gets built before it passes a human gate, and nothing counts as accepted until it works in live use against criteria agreed in writing before building. I name what not to automate. No work that only a licensed professional can judge. No whole company rebuild. No outcome promise.
3. Start small and build momentum
Users have very little patience for AI. A massive overhaul asks for more patience than users have.
Angora method: momentum comes from small wins, not from scope. One workflow, finished and used, earns the next one. A long program with no early win teaches your team that AI is another project that stalls.
How I deliver it: one workflow. One list, one owner, one due date, one weekly report. About 4 weeks per build, sequential, remote only. I build it.
In November 2025, McKinsey found 39% of respondents reporting some EBIT impact. That is survey context, not a promise about your company, and not a result I claim for Angora.