Code is cheap now. Static requirements processes are too slow for radical challenges.
Engineers need to be closer to the source problem space. Stakeholders need to feel the product as it is being built. Feedback needs to move immediately.
Every serious product pushing boundaries needs a living, breathing PRD: a continuously evaluated system of assumptions, feedback, measurements, failures, and decisions.
The question is no longer:
Did we build what the document said?
It is:
Are we learning the truth fast enough to build something that actually matters?
What I do
I turn open questions into proven answers.
That means taking ideas like autonomous investigation, analyst augmentation, agentic workflows, and operational AI, then reducing them to the questions that actually matter:
- Will users trust it?
- Does it improve the workflow?
- Can it handle real-world complexity?
- Does it produce better decisions?
- Can we verify what it did?
- Does it move the north star?
My job is to find fit.
To do that, I start by building systems that are just real enough to test the truth, and no realer.
Principles
Be wrong cheaply. Be right quickly.
Innovation is not about protecting ideas.
It is about exposing them to reality as fast as possible.
A cheap failure is progress. A slow maybe is waste.
Just real enough, and no realer.
Every piece should be built only as far as it takes to generate honest signal.
If a mock proves the point, mock it. If a script proves the point, script it. If production is required to learn the truth, go to production carefully.
Build for evidence first. Scale second.
You cannot build autonomous operations in a lab.
When a system's value depends on real-world complexity, fit cannot be proven in a sandbox.
Real autonomous operations depend on messy data, ambiguous workflows, imperfect users, business context, escalation paths, trust boundaries, and institutional habits.
You are not the expert.
The customer is. The analyst is. The operator is. The workflow is.
Test with real users, real data, and real operational pressure, or you are measuring nothing.
Low tech, high value.
Use the least technology required to answer the open question.
Do not build architecture to support assumptions. Do not scale technology that has not proven itself worth scaling.
The best prototype is the smallest system that can produce an honest answer.
Pick a measurable north star and relentlessly hill climb.
Identify ground truth and run toward it.
Judge the system against reality: better investigations, faster decisions, fewer misses, cleaner handoffs, reduced toil, higher trust, stronger outcomes.
If the metric cannot survive contact with ground truth, it is not worth measuring.
Earn trust through verifiability.
What a user cannot verify, they will not trust.
What they do not trust, they will not adopt.
Autonomous systems need to show their work: what they saw, what they inferred, what they did, what they skipped, and where the human should care.
An honest failure beats a fake metric.
A failed experiment that tells the truth is valuable.
A green dashboard that hides uncertainty is dangerous.
Innovation work should produce clarity: kill it, change it, scale it, or keep learning.
False confidence is the enemy.
Operating model
- Start with the open question.
- Build the smallest useful thing.
- Get as many eyes on it as early as possible.
- Put it in the real workflow quickly.
- Measure against ground truth.
- Keep what works.
- Cut what does not.
- Hill climb.
- Repeat until success, then scale.