The thinking
Frameworks we work by
These are the models we use inside engagements, published in full. They are our point of view made concrete, and they are yours to use whether or not we ever work together.
Framework
The Production Gap
The distance between a working AI demo and a system in production, described as five specific gaps.
For: Teams whose pilots keep working in the room and failing in the wild.
A demo proves a model can do something once. Production means it does that thing reliably, safely, and affordably, thousands of times, inside a real business. The reason most generative AI pilots never ship is that they close none of the five gaps that separate the two.
1. The evaluation gap
A demo is judged by whether it looked good. Production needs a way to measure whether the system is right, consistently, before and after every change. Without evaluation you cannot improve a system or trust it.
2. The integration gap
A demo runs in a sandbox. Production has to live inside real systems, data, and permissions. Most of the work, and most of the risk, is here, not in the model.
3. The cost gap
A demo ignores unit cost. At scale, inference and data costs decide whether the use case has a business at all. This has to be modeled before you build, not discovered after.
4. The trust gap
A demo can be wrong and no one cares. Production needs guardrails, human review where it matters, and a clear answer to what happens when the model fails. Trust is designed, not hoped for.
5. The ownership gap
A demo is owned by whoever was excited. Production needs a named owner, a runbook, and a team accountable for it on a bad day. Systems without owners quietly rot.
How to use it
Score any AI initiative against these five gaps before you invest. The ones that will ship are the ones where you can name how each gap gets closed.
Framework
The AI-Native Ladder
Four stages of AI maturity, from ad-hoc experiments to an operating model rebuilt around AI.
For: Leaders who need to know honestly where their organization stands.
AI-native is not a tool you buy or a team you hire. It is a stage of maturity, and most organizations are lower on the ladder than their slides suggest. Naming the rung you are actually on is the first honest strategic act.
Rung 1: Ad-hoc
Individuals use AI tools privately. There is energy but no direction, no shared standards, and nothing that compounds. Value is anecdotal.
Rung 2: Assisted
AI is added to existing workflows to make them faster. The process is unchanged, so the gains are real but capped. This is where most companies plateau.
Rung 3: Integrated
Individual workflows are redesigned around AI, with evaluation and guardrails in production. The division of labor between people and models is deliberate. Value is measurable.
Rung 4: AI-native
AI is part of how the organization works, decides, and builds, by default. New products and processes assume it. The advantage compounds because the operating model itself has changed.
How to use it
You climb this ladder one workflow at a time, not with a company-wide program. Pick the rung you want to reach, then earn it on a single workflow before you scale the pattern.
Framework
The Use-Case Value Map
A way to rank AI opportunities on value and feasibility so priority is a decision, not a preference.
For: Teams with more AI ideas than focus.
The problem is rarely a shortage of AI ideas. It is the absence of an honest way to choose between them. The Value Map scores every candidate on two axes and turns a long list into a sequence.
Value
How much this moves cost, revenue, or cycle time in your specific business. Estimated in real numbers, not adjectives, even if the range is wide.
Feasibility
How ready the data, the technology, and the organization are to ship this, and how close current model capability is to what the use case needs.
Prove first (high value, high feasibility)
Your opening moves. These fund credibility and momentum for everything after them. Start here.
Invest and sequence (high value, low feasibility)
The bets that matter but are not ready yet. Do not start them. Sequence the data and capability work that makes them feasible later.
How to use it
Kill the low-value quadrants without guilt, prove the easy wins first, and treat the hard high-value bets as a roadmap, not a to-do list.
Framework
The Wrapper Test
Five questions that separate a durable AI advantage from a thin layer over someone else's model.
For: Investors and acquirers evaluating AI companies.
The hardest question in AI diligence is simple to state: is this a real company, or a wrapper that the next foundation-model release will erase? These five questions get you to an honest answer.
1. What survives the next model?
If the leading foundation model gets twice as good next year, does this product get stronger or redundant? Durable companies ride the wave. Wrappers get swallowed by it.
2. Where is the proprietary advantage?
Is there data, distribution, workflow depth, or a system that a competitor cannot copy in a weekend? If the only asset is a prompt, there is no moat.
3. What is the dependency risk?
How exposed is the business to one model provider's pricing, terms, and roadmap? A single point of dependence is a single point of failure.
4. Do the unit economics hold at scale?
When inference volume multiplies, does the margin survive? Many AI products are demos that cannot afford their own success.
5. Can this team build the next version?
The advantage is rarely the current product. It is whether the team can keep building faster than the platform commoditizes them.
How to use it
A company that answers all five well is riding the model wave. One that answers none is standing in front of it. Most sit in between, and the score tells you where the real risk lives.
Want these applied to your situation?
Frameworks are a starting point. The value is in using them against your real business. That is what an engagement is for.