I spent five years making new technology legible to markets. Now I make it absorbable by companies.
Most recently that technology was the infrastructure for autonomous agents. The bottleneck was never the technology. It was the organization on the other end. I embed with teams to rebuild the value chain with an AI-native operating model and infrastructure underneath.
- Currently
- Engagement lead at a pre-IPO consumer brand, roughly $200M revenue and 1,200 employees across 98 locations. Co-founder, Compression.
- Before
- Agent execution primitives at Ritual, shared security at EigenLayer, agentic governance at Inversed. Four years of research on the intersection of emerging technology and organization design at Fielding.
- Writing
- On the bleed, adaptation under load, and why added intelligence makes some organizations worse. Read the essays →
Companies do not have an AI problem. They have a metabolism problem.
AI floods an organization with cheap cognition. The approval chains are the resistant tissue: they cannot absorb the signal, so output pools as queue depth instead of becoming work. The company reads the stall as a tooling problem and buys more, which is more insulin into a body that has stopped responding.
Meanwhile BCG, across about twelve thousand workers in fourteen markets, finds 42 percent of regular frontline AI users handing back a full workday every week, and two thirds of them given little or no guidance on where it should go. That unclaimed capacity is the bleed, and it is the largest unowned asset on most balance sheets.
Organizational metabolic syndrome
The root cause under the returns numbers. Same body, new mind: agents raise cognitive supply while the organization's demand structure stays fixed, so the surplus pools instead of moving. Three fixes: shorten the uptake pathways, move decision rights to the work, and turn humans from muscle into nervous system.
Read →The Bleed
BCG priced it across about twelve thousand workers in fourteen markets. At a thousand-person company that is roughly sixty full-time employees' worth of capacity, freed by AI and never put on the books. No CFO would leave a recovered budget line unallocated. Most companies are doing exactly that with time.
Read →Time Under Load
Environmental change is exponential and organizational adaptation is linear, so the gap widens every period and no amount of additional effort closes it. Capacity comes the way it comes for any body: specific load, held in place until the structure adapts. Incumbents are not weak. They are deconditioned.
Read →I build the machine I work inside.
A governed repository holds the entire practice: engagements, methodology, contracts, decisions, and a structured deposit from every meeting that happens. Roughly 3,300 documents and 430 meeting records, each carrying frontmatter so machines can read it as well as people.
On top of it: retrieval across three layers with glossary expansion and citation-graph reranking, twenty-one portable skills that run recurring work end to end, fifty scripts, an evaluation harness for retrieval quality, linters that enforce the house style at commit time, and validation that runs in CI on every change.
What it buys is time. Discovery synthesis that takes a team a week takes me a day, because the previous engagements are queryable rather than remembered. Findings carry their provenance, so a challenge resolves in minutes. Every project deposits back, which is why the fifth engagement costs less to run than the first.
This is the argument applied to my own practice. Short seams, decisions next to the work, and nothing re-derived that was already learned.
Infrastructure first, then the other end of it.
Five years inside the infrastructure for autonomous systems, then a move to the side of the problem that was actually stuck. Nine years before that building brands and communities, which is where the facilitation came from.
Compression
Inversed
Ritual
EigenLayer
Spring Health
Signify Health
If you have bought the technology and nothing has changed.
That is the conversation I am here for. The failure is almost never the model. It is that the body underneath was built for a different kind of work, and nobody has taken it apart. Diagnostics start at two weeks.