A Headcount Target Is Not an AI Plan
The Dutch plan for about 25,000 fewer full-time positions is a budget target, not a measurement of AI savings. Before linking the two, measure what the redesigned work actually costs.
Read articleArticles.
Perspectives by Adeilson Brito on turning AI capability into dependable enterprise work.
What Can Leaders Conclude From an AI Employee Interview?
A favorable pulse rating can coexist with a real workplace concern. AI interviews add detail, but leaders still need to separate what was reported, how widespread it is, and what action the evidence supports.
Read articleIf Software Becomes Disposable, What Will Customers Still Pay For?
If customers can recreate an interface, what are they still buying? Disposable software puts the spotlight on data, services and expertise, and on whether those capabilities can defend a price.
Read articleHow to Test Agent Recovery From Blocked Tasks
What should an agent do when its approved path stops working? Test retries, uncertain outcomes and missing prerequisites alongside successful tasks, so persistence does not quietly expand its authority.
Read articleWhat Would Change Your Mind About AI Consciousness
An AI can sometimes report on its own internal state. Does that tell us whether it experiences anything? Two studies from one lab show why the answer depends on what we think the evidence establishes.
Read articleCan You Trust the Screen That Shows What Your Agent Did?
The records may be intact while the review screen tells the wrong story. Test how events become status, approvals, summaries and exports, especially when an action is incomplete or its outcome is unknown.
Read articleWhat Must Survive the Meeting?
A few words in an AI meeting summary can turn an assessment into a commitment. Test the record by what readers do next: what they begin, what they wait for, and whether corrections reach the work.
Read articleWhat Makes Someone Ready to Own an AI Deployment?
Who is ready to own an AI deployment? A practical approach for managers and technical leads to assess delivery evidence, choose the right assignment, and provide the support people need to succeed.
Read articleWho Gets the Time AI Saves?
AI can make work faster. Who benefits depends on what happens next. A look at time saved, employee trust, and the choices leaders make about workloads, recognition, and opportunity.
Read articleThe AI Agent ROI Test: What Did the Business Actually Keep?
Measure the value an enterprise actually retains after agent execution, verification, repair and sustained operation—not just the cost of generating an answer.
Read articleWhen Code Gets Cheap, Feedback Gets Expensive
AI makes proposed code abundant. Useful feedback, reliable validation and integration capacity determine how much of it becomes delivered software.
Read articleYour Agent Finished. Can You Reconstruct What Happened?
A completed agent task should leave independently accessible evidence. Working context and retained execution records need separate lifecycles and policies.
Read articleAgent Topology: The Missing Architecture of Multi-Agent Systems
Communication, authority, memory and verification form distinct graphs. Their relationships determine the capability and consequences of multi-agent systems.
Read articleOrganizational Knowledge Is Not Context
Separate organizational knowledge, governed semantics and task-specific context to give enterprise agents information they can responsibly use.
Read articleThe Agent Execution Envelope
Distinguish the agent’s cognitive loop from its technical execution boundary, and make task-specific limits explicit through an execution contract.
Read articleAn AI Model Produces Answers. An AI System Produces Work.
Reliable organizational work emerges from the context, coordination, evaluation and control engineered around a model.
Read articleStop Choosing an AI Model. Build a Model Portfolio.
Why model choice is becoming a governed runtime decision, and how evaluation, routing and enterprise constraints shape a model portfolio.
Read article