Notes on finding the money lost between teams and systems.
Where money slips between departments, how to check it with records you already have, and what a team should approve before anything changes. Newest first.
Latest posts
Showing 19-36 of 55
Aug 30, 2026·4 min readHow do we measure the ROI of enterprise AI?
Measure enterprise AI ROI from attributable benefit and total cost over the same period. Keep forecasts, observed results, and causal claims separate.
Aug 28, 2026·9 min readAutonomous Security Patching Breaks When the Agent Owns the Merge
Agentic security patching needs a reviewable boundary between candidate fixes and production authority. Inspect the configured permissions, evidence, and recovery path.
Aug 18, 2026·7 min readA three-week model cadence makes hard-coded workflows expensive
Gemini 3.7 Flash followed 3.6 in three weeks. Enterprises need a stable workflow, evaluation, and approval layer around replaceable models.
Aug 16, 2026·8 min readThe agent loop DeepSeek just open sourced was never the hard part
DeepSeek Harness open sources the whole agent runtime as plugins. What ships free is the loop. What still has to be built is the layer standing above it.
Aug 15, 2026·7 min readThe AI gap insurers call a data problem is a context problem
A 2026 insurance AI study names data as the top obstacle to AI progress. What insurers are describing are four systems that never learned to answer one question together.
Aug 7, 2026·5 min readWhat is an approval gate?
An approval gate blocks a designated AI action until the required reviewer approves its current plan. Routine steps can run within standing authority.
Aug 6, 2026·7 min readLeast privilege does not govern AI agents. Approval does.
A new report finds most enterprise AI agents carry more access than their task needs. Scoping that access is not the same as governing the action.
Aug 4, 2026·7 min readThe confidence gap in insurance AI is an evidence gap
A 2026 insurance AI survey found adoption outrunning oversight. The gap the industry calls confidence is a missing evidence trail.
Aug 3, 2026·8 min readThe EU delayed the hard AI Act obligations to 2027. It did not delay the easy one.
The EU AI Act Digital Omnibus delay pushed high-risk hiring and insurance obligations to December 2027. The transparency duty still starts August 2026.
Aug 1, 2026·2 min readCan AI compare the cost of acting versus doing nothing?
Compare acting, waiting, gathering evidence, and making no change over the same period. Show future costs and benefits as estimates until observed.
Jul 29, 2026·4 min readWho owns what AI learns about our company?
Who owns AI agent memory depends on data-use terms, retention, and usable exports. Compare the separable company intelligence with the licensed runtime.
Jul 27, 2026·9 min readUnderwriting agents reached the bind decision
Underwriting agents can carry a policy to bind ready with no human in the workflow. The missing piece is not the model. It is the gate on the decision that commits capital.
Jul 25, 2026·3 min readCan AI learn how our best employees make decisions?
Institutional knowledge includes evidence, human judgments, procedures, and outcomes. A useful company record preserves when a lesson applies and why.
Jul 23, 2026·7 min readFor high-stakes decisions, the boundary comes before the benchmark
For a data-sensitive enterprise, the first model question is where it runs. Keep governed context and measured outcomes inside the customer boundary.
Jul 22, 2026·7 min readWhat is a trust boundary in enterprise AI?
A trust boundary separates permissions and permitted data flows. Review models, tools, logs, backups, and connected services in the actual configuration.
Jul 20, 2026·7 min readThe Hugging Face breach was run by an AI agent. The defense was too.
Hugging Face says an autonomous agent ran a multi-stage intrusion through its dataset pipeline. Agents with credentials are now an attacker class. The answer is architecture.
Jul 19, 2026·2 min readWhat are the hidden costs of using AI at work?
A low model or seat price can hide expensive integration, review, rework, and maintenance. Compare the full cost of an accepted business result.
Jul 13, 2026·8 min readFive questions that test any AI sovereignty claim
Palantir calls AI sovereignty your alpha. The test happens in diligence: five questions that separate sovereign architecture from a slide in the deck.