Tech blog
Notes on the tech, product design, and the thinking behind key features.
The judgment line moves: evidence, ceilings and the delegation ladder
The judgment line is drawn on the evidence you have on day one. As the record builds, some steps earn the right to finish on their own, some never will, and some should move back. How to move the line, and where it stops.
judgment linedelegation laddergoverned delegationai adoptionbusiness processThe delegation ladder: what to hand over to AI first
Delegating work to AI is built in steps. The delegation ladder sets out seven kinds of work, from finding answers to working with partners, and what it takes to move up each rung.
governed delegationdelegation ladderai governanceai adoptionautomationGoverned delegation: the future of business software
AI can now do the work between systems that rules never could. Letting software exercise judgment on our behalf is not an intelligence problem but a delegation problem, and delegation needs authority, boundaries and accountability.
governed delegationai governanceai agentsautomationauditBlocking is half the answer: AI usage control, and what comes after it
AI usage control is the fastest-growing part of the market for securing AI. What these tools do, why stopping unapproved AI is only half of the problem, and where a company-approved AI workspace fits.
ai usage controlshadow aiai governanceapproved aiconfidential documentsA memory is not a record
Portable memory layers for agents are shipping, and they clarified a line I drew too firmly last year. Let the agent remember what it likes. The part someone has to answer for is a different artifact.
agent memorygovernanceauditai agentsHermes and Forge: where the designs differ, and why
Hermes is an agent harness. Forge is a governed workspace for confidential work. Comparing profiles, skills, memory and autonomy, and the reasoning behind each decision we took.
hermesagent harnessskillsjudgment linebusiness aiThe judgment line, applied to a whole process
A method for deciding where AI belongs in a business process. How to shortlist the processes worth the effort, why cycle time matters more than volume, and how to choose between rules, programs, AI and people for each step.
judgment linebusiness processai adoptionworkflow automationmethodologyA plain guide to governing AI in your business
What governing AI actually means, in four simple questions, why it gets harder as AI grows more capable, and how it connects to emerging standards like ISO 42001 and the NIST AI RMF.
ai governanceiso 42001nist ai rmfeu ai actshadow ai“We don’t store your documents, we store vectors” is usually false
A comforting sentence AI vendors say about your data, why it is almost never true, and a two minute check you can run on any vendor including this one.
privacyembeddingsvendor diligenceai governanceThe memory a personal agent needs is the memory a business agent can't have
Longitudinal memory is the point of a personal AI agent and a liability for a business one - on harness engineering, governed memory, and auditability.
harness engineeringagent memorygovernanceai agentsContext debt, and how Forge handles it
An AI assistant is only as trustworthy as the documents underneath it. How Forge surfaces context debt, keeps unconfirmed knowledge out of answers, and leaves the judgment to your people.
contextknowledge basegroundinggovernanceBefore you build AI agents, build your organizational memory
Everyone wants AI agents that act, but acting needs organizational memory that is current and confirmed. Why you build the assistant that answers first, then act.
memoryai agentsorganizational memoryknowledge base
