A fifteen-year study at the intersection of frontier AI research and enterprise value. The author, a researcher, founder, operator, and advisor to Fortune 500 C-suites, finds that durable advantage emerges only when deep technical authorship meets commercial command: 11 patents, 20+ peer-reviewed publications, contributions to 9 books, and $300M+ in strategic deals closed across Fortune 500 enterprises spanning nine industries including aerospace, media, technology, financial services, healthcare and beyond. Meetings were harmed in the making of these results.
Note. Deal value reported in aggregate. Results significant at the board level (p < .01). Replication attempts welcome, see §18.
Research programs
Four questions we work on.
Each program publishes named frameworks a team can put to work the same week, backed by cited evidence and open tools.
Program 01
Owned intelligence
When a capable model runs on hardware you already own, renting every answer stops being the default. This program maps what to own, what to rent, and what the switch is worth.
Software that acts on its own needs an identity, a boundary and rules it cannot forget. This program defines how agents are attacked, contained and governed.
An agent is only as good as what it remembers and how it is tested. This program measures memory strategies under the budgets real deployments run with.
A confident answer is not a defensible one. This program builds the evidence layer that lets an AI-assisted finding survive an auditor, a board or a peer.
capabilitiesagentic AI · frontier reasoning · compound AI systems · RAG · cybersecurity · decision intelligence
alignmentP&L outcomes · empathy-first leadership · trust by design
deploymentproduction since pre-ChatGPT era
deployed_forFortune 500 · Federal · Defense · Aerospace · Healthcare · Media · Energy
Known limitations: will not present a pilot as a strategy. Refuses to ship hype without an architecture. Allergic to recurring meetings with no agenda. Hallucination rate on market foresight: anomalously low. Claims are grounded in peer-reviewed research.
§ 03 · How I think
Bring me a hard problem. Watch the read.
Pick the question keeping you up. The diagnosis assembles step by step, the same first-principles read you would get in the room.
reasoning trace · idle
The read
AI solutions engineering
A hard problem, engineered end to end.
The read is one thing. This is the build. A market problem taken apart and rebuilt as a working solution, each one a full-screen presentation.
The architecture The architectureThe architecture The architecture The architecture
§ 04 · Architecture
Fluent at every layer.
Most careers cover two layers of this stack. Strategy that has never touched the stack is a guess. Pick a layer to see the work done there.
value flows uptrust flows down
The work at this layer
§ 05 · Reference architecture
Own your intelligence, drawn live.
One harness runs any open model. Memory, evaluation, training and an evidence record sit on top of it, inside your walls. Choose a job and watch it route.
What the stack gives you
Reference implementation. Zorp, an open-source stack under the MIT license. zorp.dev ↗
§ 06 · Method
The training run, 2011 to now.
Fifteen years of compounding reps. Pretraining as a founder, scaling inside global enterprises, an alignment phase in security, and deployment at the frontier. Drag along the curve.
§ 07 · Incident log
Known failure modes of the enterprise.
Fifteen years of field observation across nine industries. None of these are technology problems. All of them are why AI programs stall. Open an incident and apply the patch.
Fig. 6 · Illustrative
Meetings held against decisions made
0of 6 patched
Meeting volume climbs every quarter. Decisions stay flat until the failure modes are patched. Each patch you apply above lifts the line.
§ 08 · First principles · The inner stack
The part of the architecture nobody audits.
"The intelligence economy is not a gold rush. It is a consciousness shift. Companies chasing money will crash. Companies chasing research, with empathy in the architecture, will lead the market."field notes · repeated to boards until it compounds
Fig. 7 · The crash curve
Money is a lagging indicator of research
Research firstMoney first
Part III
The instruments The instrumentsThe instruments The instruments The instruments
§ 09 · Live model
Rent or own, on your numbers.
Renting meters you forever. Owning costs the hardware once, then power. Move the sliders to your own workload and watch where the lines cross.
Rentinga month
Owninga month after hardware
Pays back inmonths
Over 24 monthssaved
Arithmetic on the numbers you enter. Not a quote and not a measurement.
§ 10 · Diagnostic
How much of your intelligence do you own?
Six questions, one per layer of the stack. Thirty seconds, no email gate. You leave with a posture and the one move that matters most.
Your stack so far
0%owned
§ 11 · Interoperability
One program, eight seats.
An AI mandate only works when every seat at the table gets something it can use. Pick your seat.
of 08 seats
You bring
You leave with
§ 12 · Generated on demand
Your one-page brief, built live.
Choose who is reading and what is on their mind. The brief assembles itself from the research, with the three moves and what to read next.
Who is reading
What is on their mind
§ 13 · Interactive session
Query the author.
Documentation is fine. A conversation is better. Ask the console anything a diligence team would, or tap a prompt.
aj-1 · live sessiononline
§ 14 · Evaluations
The body of work, counted.
No self-awarded scores. Only what is published and free to read. Every bar opens its index.
Counts from the site indexes.
§ 15 · Value mapping
How research becomes outcome.
A framework is only worth the system it becomes and the result a board can see. Hover any node to follow its path.
ResearchSystemOutcome
§ 16 · Capability surface
The capability surface, with receipts.
Eight capabilities. No scores, only the published work behind each one. Pick a point on the orbit.
AJ-1capability map
The receipts
Part IV
The research The researchThe research The research The research
§ 17 · Research
Research from the frontier.
Updated this month
Annual reports, long reads, and field notes grounded in primary sources. No hype. Written for executives who need to act.
White Papers
Practitioner-ready research published in print-quality format. Download and share.
Pilot Purgatory Is a Governance Problem, Not a Model Problem
Industry projections show 40% of enterprise apps embedding agents by year-end, up from under 5%. Most of those projects will still die in pilot. The reason isn't the model. It's that nobody put guardrails, owners and a P&L line on the output path.
agenticgovernanceproduction
read note →field note 01
Enterprise ValueJun 2026 · 3 min
The 75% That's Trapped in Your Silos
The headline number this month: up to three-quarters of AI value stays stuck because organizations optimize functions, not flows. Orchestration, not more automation, is the unlock.
orchestrationvalue flows
read note →field note 02
World ModelsJun 2026 · 3 min
Why World Models Are the Real Next Frontier
Pattern recognition plateaus on more compute and more data. The next leap is systems that model how the world actually works, causal, physical, social, and the enterprises that prepare for it now will not be caught flat.
world modelsreasoning
read note →field note 03
ComputeJun 2026 · 3 min
A Trillion-Fold Ramp, and What It Means for Your Roadmap
Training compute has grown a trillion-fold, with another thousand-fold expected in three years. "Superintelligence labs" are now a category. Strategy built for today's model sizes is already obsolete.
scalinginfrastructure
read note →field note 04
StrategyJun 2026 · 4 min
The Frontier Just Fractured Into Four
There is no single frontier anymore: regulatory, efficiency, cost and capability are now separate races. Picking the right one for your use case is the new core competency. Hype picks all four and wins none.
frontier modelscostregulation
read note →field note 05
Context EngineeringJun 2026 · 4 min
Prompt Engineering Is Dead. Context Engineering Is the Job Now.
The model is no longer the bottleneck. What you put around it is. Context engineering - designing what the model knows, when it knows it, and how it forgets - is the discipline separating teams shipping value from teams iterating on demos.
context engineeringRAGsystem design
read note →field note 06
MCPJun 2026 · 3 min
MCP Is the USB-C Moment for Enterprise AI. Most IT Teams Are Not Ready.
Anthropic's Model Context Protocol is becoming the de facto standard for connecting AI agents to tools, data, and APIs. The land grab is already underway. Vendors who own the MCP server own the integration layer - and the margin that comes with it.
MCPagenticintegration
read note →field note 07
Vibe CodingJun 2026 · 3 min
Vibe Coding Is Not the Threat to Engineering Teams. The Backlog It Creates Is.
AI-generated code ships faster than it can be reviewed, tested, or secured. The velocity is real. So is the debt. Enterprises adopting vibe coding without a governance wrapper are building a liability that will arrive in the form of a breach, not a sprint retrospective.
vibe codingAI devsecurity
read note →field note 08
Reasoning ModelsJun 2026 · 4 min
Your Organization Does Not Need a Reasoning Model. It Needs to Know When Not to Use One.
o3, Gemini 2.5 Pro, Claude Opus 4 - the reasoning tier is powerful and expensive. Most enterprise use cases do not need it. The real skill in 2026 is model routing: matching task complexity to model capability without paying frontier prices for commodity work.
reasoning modelsmodel routingcost
read note →field note 09
AI ROIJun 2026 · 4 min
The CFO Will Ask for the AI ROI Number. You Should Have It Before They Do.
Boards are done funding AI on faith. The new pressure is proof: cost per outcome, not cost per token. Enterprises without an AI measurement layer are one budget cycle away from a hard conversation. The ones with it are compounding advantage while everyone else recalibrates.
AI ROIfinancemeasurement
read note →field note 10
Note. Perspective pieces, the author's own views on live industry themes, written deliberately ahead of the consensus. That is the job. Follow along via LinkedIn.
▮ field note · jaggi-lab
§ 18 · Research in Practice
What to Build Next — Your Enterprise AI Playbook.
Solutions I would put on the board agenda. Each is grounded in peer-reviewed research, scoped for Fortune 500 deployment, and designed to move a metric your CFO tracks. Click any solution to see the full brief, architecture, and deployment plan.
Calculators, templates and frameworks grounded in real research. Built for leaders who need to move fast and get the numbers right.
§ 20 · Selected output
Research and recognition.
Strategy without research is opinion. The counsel given to boards is grounded in original, published, peer-reviewed work, and in hundreds of essays and white papers.
0Patents
AI systems, security and enterprise intelligence. Granted, USPTO and international.
0Peer-reviewed publications
IEEE and Scopus venues on agentic AI, machine learning and applied intelligence.
0Books
Contributing author on applied AI and enterprise innovation. In print.
0International awards
For AI innovation and research. Global recognition, 2018 to present.
Invited keynotes and global speaking at international innovation conferences across multiple continents.