What I do, 2025
How I described what I do in October 2025.
This page is deprecated.
AI-Native Engineering
Engineering practices rebuilt for the AI age: past the 'almost done' trap, to a disciplined AI-native culture.
The problem
The age of AI has broken the old rules of software development. The bottleneck is no longer writing code; it is defining the spec and validating the output. That opens two traps:
- The “almost done” loop. LLMs generate code that is 80% correct and often carries subtle bugs or inconsistencies. Teams get stuck chasing AI-generated errors through fixes that create new problems.
- The productivity illusion. A team can write more code than ever, but if it is the wrong code, or not truly done, real productivity falls. Old metrics fail, and timelines and quality become impossible to predict.
The way out is a move from a code-first culture to a disciplined, AI-native one.
How I approach it
I install the AI-Native Operating Model: the frameworks and hands-on practice that turn AI into an advantage rather than a source of rework.
- The Six-Gate Loop: a spec-driven, validation-first delivery process that makes quality the path of least resistance.
- Architectural guardrails: “Frozen Zones” and other guardrails that protect critical assets and keep the system stable.
- Automated validation: a practice of deep, automated validation that catches errors and regressions before production.
- Metrics that matter: regression rates and time-to-green, instead of vanity metrics like lines of code.
What it produces
- A clear roadmap from a traditional to an AI-native engineering culture.
- Higher “done-done” velocity, out of the “almost done” loop.
- A team that predictably delivers high-quality, AI-augmented work, and can estimate it.
Engineering Leadership
Strategic technology leadership for organizations navigating the complexity of the AI age.
The problem
Technology leadership has never been harder. Leaders are caught between the pressure to innovate with AI and the responsibility of running existing systems. The result is methodological inertia: teams apply old development lifecycles to a new class of AI-driven problems. That produces a productivity illusion: more code shipped, and with it more bugs, more rework and more technical debt.
How I approach it
I lead at the level of the organization, installing the AI-Native Operating Model across it:
- The Six-Gate Loop: a spec-driven, validation-first delivery process that makes quality the path of least resistance.
- System Directors: coaching senior engineers to design and validate systems, not only write code.
- Guardrails: “Frozen Zones” and other architectural guardrails that protect critical assets.
- Measuring what matters: regression rates and time-to-green, not lines of code.
What it produces
- A clear transition plan from a traditional to an AI-native engineering culture.
- Higher “done-done” velocity and lower regression rates.
- An engineering team that operates with discipline and confidence.
- Technology investment aligned with real business value.
IP & Prototyping
Validating new ideas and building valuable IP without the risk of over-investment.
The problem
In the rush to innovate with AI, many companies see a demo that looks 95% complete and over-invest in a full-scale product, only to find themselves in the “95% Trap”, where the last 5% is a resource-draining slog. The better path is to prove the concept first, with a rigorous process that builds intellectual property along the way.
How I approach it
I run an idea through the first stages of the Six-Gate Delivery Loop, which validates the concept and leaves behind a well-documented prototype:
- Spec gate: the vision becomes a testable spec that defines the prototype’s behavior and constraints.
- Scaffold and test gate: the architectural skeleton and the tests come before the implementation.
- Gen/Edit gate: the prototype is built inside those guardrails: a functional asset, not a brittle demo.
- IP strategy: the documentation, specs and tests from this process form the core of the intellectual property, worth far more than a standalone piece of code.
What it produces
- A working prototype that rigorously tests the feasibility and value of the concept.
- A clear view of the technical risks and a reliable path to production.
- A foundation for the IP: specs, tests and architecture.
- Far less time and cost to validate a new idea.
Applied AI Systems
Production-ready AI that cuts through the hype and delivers real-world value.
The problem
Most AI initiatives never move from a promising demo to a robust production system. Frontier models generate code that is 95% correct, and that creates the “95% Trap”: teams stuck in a perpetual “almost done” loop, applying old methods to a new class of problems, producing brittle systems that never truly ship.
How I approach it
I build AI systems with a lightweight version of the AI-Native Operating Model, the antidote to the hype cycle:
- Testable specs: goals turned into a clear, testable contract the output is validated against.
- Architecture for validation: “Frozen Zones” and clear interfaces that isolate AI-generated code from critical components.
- Guardrails in the build: a spec-driven, test-first workflow that catches errors and hallucinations early.
- Teams that keep it going: the people who maintain the system learn the same AI-native practice.
What it produces
- Production AI systems that escape the “95% Trap” and deliver measurable value.
- A spec-driven development process that is predictable and keeps quality high.
- An architecture whose “Frozen Zones” protect the critical assets.
- A repeatable framework for the next system.