01 / 04
MOST AI STRATEGIES OPTIMIZE FOR USE CASES. THE WINNERS OPTIMIZE FOR WORKFLOWS. THE FIRST CREATES DEMOS. THE SECOND CREATES LEVERAGE. WE START BY MAPPING WHERE THE WORK ACTUALLY HAPPENS.
// 01 — ABOUT
About
Most companies are using AI to do their existing work a little faster. A faster draft, a quicker summary, a chatbot bolted onto the website. That is the small version of this, and it is the version most of the market has settled for.
The larger shift is that AI changes what work is worth doing at all, and almost no one is organized to act on that. The technology is no longer the constraint. The models are good and getting cheaper every quarter. The constraint is organizational: the habits, the incentives, and the workflows nobody wants to rebuild. That is the harder problem, and it is the one I work on.
I spend my time at the point where the building meets the adoption: designing the systems, then doing the slower work of getting an organization to actually change how it operates. Most people in this field can do one side of that. The work that matters happens when someone does both.
This site is where I write about it.
// 02 — DOMAINS
Where the work lives.
AI Deployment
From prototype to production: integration, volume handling, organizational adoption.
The model is the commodity and deployment is the product. Organizations that treat the last mile as cleanup work are spending on the 80 percent that was already solved.
Evaluation Design
Frameworks for what "good enough" means across high-stakes professional workflows.
Good enough for what, compared to what, according to whom, at what cost. An unmeasured deployment is unfalsifiable, and unfalsifiable investments are how budgets die.
Agent Architecture
Systems that automate and augment expert work without removing human judgment.
The question is never what can be automated. It is what the human should still own, and the best agents compress everything around that judgment instead of replacing it.
Enterprise Strategy
Technology decisions inside complex organizations with competing stakeholders.
Most technology decisions are stakeholder decisions wearing a technical costume. The architecture diagram is rarely the hard part of the architecture.
Change Management
Why organizational adoption fails and the structural reasons it gets ignored.
Adoption is a behavior problem, not a feature problem. Trust travels through colleagues, never through mandates, which is why champions outperform announcements every time.
Knowledge Work
Making professional expertise scalable without diluting the expertise itself.
Scaling expertise is easy if you are willing to flatten it. The discipline is capturing the judgment behind the output, not just the output.
Governance & Risk
Moving fast without breaking trust: compliance, accountability, audit trails.
Governance done right is a speed enabler. Clear rules are what let cautious people move fast without fearing they will be next month's cautionary tale.
Product Leadership
Building what users need, not what stakeholders ask for, inside enterprise constraints.
Users tell you their requests and their workflows tell you their needs. Those are different documents, and the second one is where the value lives.
AI Literacy
Closing the gap between capability and understanding before it becomes a liability.
Fluency fails as a training program and works as an operating model. The gap between capability and understanding is where fear, workarounds, and confident bad decisions live.
// 03 — MANIFESTO
What I believe about AI in enterprise.
The unit of analysis is the deployment, not the model.
What separates AI investments that compound from those that stall is never the model choice. It's the deployment decision: which workflow it integrates with, what judgment it augments, what feedback loop it creates.
Most enterprise AI fails before it reaches the user.
Strategy documents, proof-of-concepts, and pilots are not AI deployment. They're AI theater. Real deployment means production systems that handle volume, edge cases, and organizational resistance, and keep working when the novelty wears off.
The evaluation threshold problem is the most underreported question in enterprise AI.
Every enterprise AI project eventually hits the same wall: how good is good enough? The absence of rigorous evaluation frameworks isn't a research problem. It's a leadership problem — and it's causing billions in misallocated investment.
AI literacy fails as a training program. It works as an operating model.
When fluency is a course, you get completion certificates and zero behavior change. When AI capability outpaces organizational understanding, you get fear, workarounds, and poor decisions made confidently. Closing that gap is not an event on a learning calendar. It is role-anchored benchmarks, rebuilt workflows, and leaders who model the behavior, owned deliberately and measured continuously.
Internal AI leadership requires a different playbook than AI-native leadership.
The strategies that work at AI-first companies create technical and organizational debt when applied inside traditional enterprise. Different constraints, different stakeholders, different definition of done. The playbook transfer problem is real and almost never discussed.
Strategy without execution infrastructure is a liability.
Strong AI POV without deployment infrastructure doesn't create competitive advantage. It creates technical debt, failed pilots, and organizational cynicism. The execution layer isn't implementation detail. It is the strategy.
// 04 — DEPLOYMENT
How AI actually gets deployed.
Four phases. Each one harder than it looks. The work that separates AI investments that compound from those that stall.
// 05 — PUBLIC PROJECTS
Building in the open.
Documenting what I build outside of work — production systems, not prototypes. Shipped publicly with the code, the reasoning, and real performance data.
01/02 Live
Autonomous Market Monitoring System
A production AI agent that observes market conditions, evaluates signals, manages risk, and executes paper trades without human intervention — every hour the market is open.
PythonAlpaca APIpandasFlaskPlotly
View details → 02/02 Live
US Data Center Power Map
Interactive map of every US data center overlaid with power plant fuel mix, NERC grid zones, and water stress data. Built to visualize the infrastructure dependencies of AI compute at national scale.
AstroReactMapbox GL JSEnergy Infrastructure
View project → // 06 — WRITING
Long-form essays on AI deployment.
Arguments about what actually works. Written from inside the problem, not above it. Follow on LinkedIn for shorter takes between essays.
Essay 01 Published
The Last Mile Problem in Enterprise AI
Why most enterprise AI investments fail not in model selection or integration, but in the final 20% between working prototype and organizational adoption.
Deployment Enterprise
Essay 02 Published
The Evaluation Gap
The absence of rigorous evaluation frameworks is the most underreported problem in enterprise AI deployment, and it is causing billions in misallocated investment.
Evaluation Strategy
Essay 03 Published
Internal vs. AI-Native: Why the Playbooks Don't Transfer
Why strategies that work at AI-first companies create debt when applied inside traditional enterprise, and what actually works instead.
Leadership Enterprise
// 07 — CONNECT