Models, pricing and the enterprise AI tools IT teams are buying, building and governing.
One release describes the same 6,000 three different ways, and none of them is a catalogue you can check. There is no price because this is sold as an engagement, and the only named client is exploring it. Five questions for the first meeting.
A pilot signed off on November's rates costs twice as much in December. Two levers change the answer, and one of them is that Google prices its cheaper-per-job model identically.
Approve a command once with a password in it and Claude Code can save the password to a project file. One managed setting stops it saving any new rules. Where that setting goes on each OS, and the order to search, rotate and clean up.
Revolut's AI operations lead says the bank serves GLM 5.3 to about 2,000 engineers on 56 Nvidia B200s at a blended 30 cents per million tokens. Here is how that compares with the list prices most IT teams are paying.
Bloomberg reported on November 2 that Goldman now lets only engineers use Claude Code and GitHub Copilot. For an IT team that wants the same split, the two products work very differently: Copilot Business can be licensed by team, while some Claude plans put Claude Code in every seat.
A fake update to the Claude Code extension sat in Cursor's marketplace for about 11 hours on October 29. Most reports name one version. Cursor's own notice names two, and removing the extension leaves the backdoor in place.
Dropping the $19 or $39 seat also drops the credits that came with it. Teams that used their whole pool pay the same; teams with idle seats save the difference, as long as someone sets a budget.
Most of the advice so far is about cleaning up the bad package. If you run a LiteLLM proxy, two more checks matter: which image your containers pulled, and what your proxy writes to its logs.
Dynamic permissions, risk scores and Guardian Agents arrive with no release date. The two licenses Microsoft ties to agent security cost $15 and $99 a user a month.
The notice gives three months, half the six OpenAI's own policy sets for generally available models. Fine-tuned versions shut down on the same day.
The introductory rate matches mid-tier models from Anthropic and OpenAI. From December 1 it moves to $4 and $20, the same as their top models.
Paid plans get GPT-6 Sol and free users get GPT-6 Luna. For Enterprise workspaces, admin settings decide when staff see it.
GitHub Copilot, OpenAI Codex and LM Studio already run inside them. Central management through Intune is not ready yet.
Most developers now use AI tools, and half of professionals use them every day. Fewer trust the output than a year ago. That gap is what IT policy has to manage.
Staff are already using AI tools, approved or not. A strategy is how IT decides which uses to back, which to block and who answers for the results.
The EU's AI literacy rule has applied since February 2025, and enforcement powers start in August 2026. It asks for less than most vendors' training pitches suggest.
Nearly nine in ten companies now use AI. Only 37% see any effect on profit. The gap between using AI and getting a return from it is the main story of 2026.
An AI audit asks what data went in, how the system was tested, and who is accountable when it's wrong. Three frameworks now give companies a common way to answer.
Employment for young workers in AI-exposed jobs has fallen behind older peers. For junior developers, the response is to get good at the parts of the job AI handles badly.
Each major cloud now sells one place to pick models, connect company data, build agents and monitor them. The differences are in model choice, integration and where your data already lives.
NLP is the field that taught computers to work with human language. Chat assistants are its most visible product, but most enterprise uses are quieter.
Machine learning finds patterns in past data and uses them to predict new cases. Most business AI that isn't a chatbot is still this kind.