AI

Models, agents and the decisions that turn AI research into useful products.

Hands compare matching highlighted passages in two paper source documents.

AI · Oct 10

Data Provenance in RAG Answers: What a Citation Proves

A citation points to a passage. Provenance also records which document version was used and who published it.

A hand selects a tool-access card from a small organizer.

MCP · Oct 6

MCP authorization: who can access an agent's tools

A closer look at access tokens, operation permissions and user consent in MCP connections.

Researchers checking an AI paper against source documents

Research workflows · Sep 25

The AI Research Stack: What Small Teams Actually Need

A source library, a repeatable question and a way to check the answer are a useful place to start.

Two workshop workers use a tablet and a physical task board while sorting materials.

AI agents · Sep 13

Agentic Workflows: Where They Earn Their Keep

An agent earns its place when the task needs decisions that a fixed sequence cannot sensibly anticipate.

An editor reviews printed pages on a light table with a red pencil.

Human oversight · Sep 1

When an AI Product Needs a Human in the Loop

Human review works best when it happens at a decision someone can still understand and change.

A technician checks cables and a clipboard in a server room.

AI economics · Aug 20

The Cost of Intelligence Is Moving to Operations

The useful unit is the cost of an accepted result, including the work needed to correct and maintain it.

A researcher draws connections on a whiteboard in a shared workshop.

Open models · Aug 8

Open Models, Closed Advantages: What the New Divide Looks Like

A downloadable model changes what a team can control. It does not settle who owns the rest of the system.

A researcher compares three compact computers on a workshop bench.

AI Models · Jul 10

GPT-5.6 Makes Model Choice a Workload Decision

OpenAI's three-model launch puts task difficulty and the cost of checking results into the same conversation.

An editor arranges film stills on a light table beside a tablet.

Creative Tools · May 27

Gemini Omni Makes the Video Edit a Conversation

Google's new video model accepts mixed references and follow-up edits, changing how a creative brief can develop.

A researcher retrieves a folder from a tall library shelf.

AI Models · Feb 20

Claude Opus 4.6 Takes on Longer Pieces of Work

A larger context window and new agent features expand the work Claude can attempt, with review still essential.

A research team organizes photographs and notes on a wall.

AI Models · Nov 19

Gemini 3 Arrives With a Wider Research Brief

Google's new model preview spans text, images and video. The useful test is whether its answers stay traceable.

The AI desk explains how models and agents behave inside real products and workflows, what they cost to run and where human review still matters, with sources linked beside each claim. Dated events are collected under AI news.