Writing
I write about software, data, agents, and the work around them.
2026
-
When AI Changes the Work, the Complements Change Too
LLMs can reduce the work required for some tasks while increasing the need for evaluation, exception handling, judgment, and accountability.
-
AI: Token or GPT?
How a general purpose technology became a general purpose token: strategy confusion starts when a media-adapted sign travels faster than the mechanism it names.
-
Agent Work That Changes Behavior
The useful agent story is not smarter chatbots or generic runtimes. It is context-bearing workflows that let people, agents, tests, and domain systems change how work happens.
-
There Is No Agent Workflow Runtime
Most serious agent systems are workflow engines where some steps, some of the time, use LLMs. The runtime should center the workflow, not the agent.
-
Agents Don’t Learn the Domain. The System Does.
Self-improving agent systems are less about models improving themselves and more about domain harnesses where people, code, tests, workflows, and knowledge bases improve each other.
-
When Your Trace Is Lying to You: A Performance Case Study
A real 2–4s → 100–300ms latency case study in what happens when most of the request time is invisible and the trace points you at the wrong work.
-
A Hybrid Search Case Study: Widely Applicable Techniques for Ranking Better Results
A case study in building hybrid search over a large expert corpus, with practical lessons on query analysis, lexical + semantic fusion, reranking, diversification, chunking, retries, and diagnostics that apply far beyond...
-
AI Changes What Counts as Value in Software Creation
AI did not just speed up coding. It changed the software production function, shifting value away from typing and search toward context, architecture, and executable constraints.
-
Velocity Is Limited by the Path to the User
Velocity is limited by the path to the user. Fix the delivery path and you unlock learning.
-
Grounded Execution: Taming the Chaos Dragon
Agents accelerate everything—including mistakes. Without grounding, they accelerate thrash. Open Horizons + Bottle keeps runs aligned by forcing intent + constraints before generation.
-
The Context Stack
How a constellation of tools creates the appearance of infinite memory
-
Ditch the Draft: How to Salvage AI Dev Wins from Context Collapse Hell
Code is now the cheapest artifact. The salvage loop keeps learning durable when AI refactors go off the rails.
2025
-
Splitting Learning from Constraint: Designing AI-Augmented Teams
LLMs make exploration cheap and enforcement automatic. The next step: separate learning from constraint so insights don't freeze into premature rules.
-
Knowledge Work Gets a Reboot
LLMs make exploration cheap, not verification. The reboot is using them to widen options early—without locking into irreversible paths.
-
Beyond the Nearest Peak
LLMs make it cheaper to inspect alternative paths and accelerate correction on the one you choose. They do not replace contact with reality. The leadership job is deciding where learning should compound,...
-
Dissent Mode: Eliminate Mid-Bar Work
Eliminate mid-bar work with barbell strategy, guardrails, and continuous re-alignment.
-
Playing to Win through Managed Disequilibrium
Leadership shakes teams out of don’t‑lose ruts with managed disequilibrium: small, safe disturbances (guardrails + short checkpoints + nimble tests) that move outcomes.
-
Alignment Is the Constraint
Acceleration is a siren call. Alignment is the constraint. A Critical Chain lens for strategy, mechanisms of action, and feedback loops. Lessons from the Arsenal of Democracy episode on EconTalk.
-
Open Horizons: Aim. Do. Reflect.
A flexible, principle-driven framework to help you clarify your direction, nurture your strengths, and maintain momentum in your growth journey.
-
Why Most Operational Analytics Fall Short And What You Can Do About It
“If you can’t measure it, you can’t improve it.” – Lord Kelvin
-
Reducing Transaction Costs with AI-Native Workflows
“Firms exist because the market is expensive.” — Ronald Coase, 1937 “Adding manpower to a late software project makes it later.” — Fred Brooks, 1975 “Institutions are the humanly devised constraints that structure...
-
Growth Is No Longer Optional – It’s a Baseline Expectation
In a matter of weeks, I transformed from a complete novice in digital signal processing (DSP) to a savvy hi-fi audio hobbyist. I didn’t enroll in a university course or hire a...
-
The First 90 Days as a Metis Mission
When you start a new role, it often feels like stepping into a foreign country. You have the skills and experience that got you the job, but suddenly you’re the outsider trying...
-
User Value Comes First
In many organizations, obsession with profit can obscure an aspect of sustainable success: product–market fit. With the rapid rise of generative AI reshaping industries, the window for maintaining a strong connection between...
-
Real-World Application of Strategic Clarity in Platform Leadership
Introduction: Connecting Theory to Practice
2023
-
Outcomes Over Outputs: A book summary
In Outcomes Over Output, Joshua Seiden advocates for a shift from the traditional output-driven approach to a more outcome-centered mindset in product development. He scrutinizes the inflexibility of adhering to predefined outputs...
-
A summary of Strategy & Tactics Trees
Entering the corporate world, we’re often introduced to the concept of strategy, something we are told we should be doing. Yet, for all its ubiquity, a clear definition and actionable guidance often...
-
Planning Not Plans: Rolling Priorities
This is an older version of my planning system. Go over to Open Horizons for the latest version.
-
Implementing SLOs for Data Quality
The most important chapter in Google’s SRE Workbook is on implementing SLOs:
-
Documenting Strategy: Lessons from Leading Data and Engineering Teams
At multiple points in my career, I’ve noticed that teams are more engaged and successful when their strategy is clear, cohesive and inspiring. This realization motivated me to explore various frameworks for...
-
Design in Practice: A Writeup
2025-12 note (LLMs + non‑ergodic risk)
-
LLM Prompt Types
You’ve heard about Prompt Engineering. You know that it’s something buzz wordy that people who’ve bought into the LLM hype go on about. What are prompts? Why do they work? How should...
-
Are you a Data Elbow?
Are you a Data Elbow?
-
A few job search tips
A few job search tips
-
Using the Bridges Model to process Layoff Loss
Using the Bridges Model to process Layoff Loss
-
2023-05-21 Links Post
Some links to things I found interesting this week:
-
Open to Work!
I want to help 10x you through strong data foundations and a focus on the most important.