In partnership with
The AI bill lands after you sign.
Three things from yesterday sit on the same desk. One company switched AI on for more than a third of a million people. Half of the CIOs in a survey said out loud that they expect it to cost more rather than less. And a vendor started shipping a tap you can turn down. None of that is a story about models.
Advertisement
Find what’s slowing your GTM.
Most GTM problems aren’t one big failure. They’re one or two weak spots quietly dragging down the rest.
The free GTM Audit scores your go-to-market across four dimensions, then sequences a 90-day roadmap for your stage.
Drop your URL. Get your score in about 10 minutes.
1. Volkswagen switched Copilot on for 330,000 people
Volkswagen Group has extended Microsoft Copilot to more than 330,000 employees. The same write-up has Microsoft and Volkswagen planning an Intelligent Data Foundation on Microsoft Fabric.
The seats are live. The shared data layer is a plan.
The figure itself was already out on 21 October, when Microsoft said more than 330,000 Volkswagen employees were actively using Copilot. On 27 October the same number came back as a customer story rather than a vendor one.
I read the sequence as backwards, and it is the part I would argue about in the room. A third of a million people are now putting questions to a data estate nobody has finished describing, and when the answers come back wrong the repair is not in the licence. It sits in the Fabric work that has not started yet. I would put that data foundation on the same board as the seat count, reported in the same weekly number, because only one of those two things has a vendor pushing it along.
2. Half of CIOs expect AI to raise what technology costs
Big Technology's enterprise AI guide carries the number that ends the savings argument: 51% of CIOs expect AI to increase the total cost of ownership of their technology solutions.
Beside it, Futurum Group's read of Oracle's agentic applications found that 21.1% of buyers who pay for generative AI as an add-on prefer outcome-based pricing, while 11.2% say their vendor offers it.
The pricing gap is the half worth sitting with. A supplier that will not price on outcomes is saying, politely, that it cannot predict the outcomes either, and the uncertainty then sits with whoever signed. That is a governance question wearing a commercial jacket. My own habit now is to read any per-seat or per-token quote as the floor of the number, and to ask what the vendor will put in writing about the result.
Advertisement
Fast Analytics on Live Data, Inside Postgres
When analytics on Postgres slows down, the usual fix is a warehouse and a pipeline to feed it. TimescaleDB keeps analytics in Postgres instead: continuous aggregates for live dashboards, compression up to 95%. No ETL, no lag. Start on Tiger Cloud with $1000 in credits.
3. Salesforce put a spending cap inside its agent platform
Salesforce added a FinOps dashboard, a Model Wallet and cost optimisation policies to Agent Fabric, together with Agent Evals and API Evals for scoring what the agents actually did.
Agent Fabric reached general availability on 29 April 2026. The writing around it this month was about control of a different kind: on 6 October Salesforce published a piece on giving AI autonomy without letting it run wild, and on 14 October Forrester argued that MuleSoft, not Agentforce, is the company's most strategic AI asset.
A wallet gets built when customers start asking what the thing cost them last month. That is the honest reading of a platform that went generally available in April and got a spend policy in October: the pilots were priced on enthusiasm. The useful part is not the dashboard, it is that the cap and the scoring shipped together, because cost and failure are the same incident seen from two desks. I would take the evals over the dashboard if I had to choose, since a cheap agent doing the wrong thing is still the expensive outcome.
Names to watch
Seen this week by Ins7ghts.
- Reduck · a browser an agent can drive
Observability. It announced the Reduck MCP platform, a browser automation service for AI agents, alongside $3 million in pre-seed funding on 22 October. - Exotool · plain language in, the same application out every time
Provenance. On 23 October it released a connector that puts Claude in front of its deterministic generation engine, covered by US Patent No. 12,461,721.
Companies that later raised money had been in the news about twice as often in the quarter before. Worth a second look in three months.
Advertisement
Most AI Tools Weren't Built for This
B2B customer issues move across teams and systems, not through a single chat window. A new Harvard Business Review Analytic Services briefing paper, sponsored by Front, breaks down where AI tools fall short in B2B service and what to ask before you invest.
Also today
- KT unveiled a platform that picks the AI model for you and manages what it costs. (4 min read)
- Oracle is bringing Google TPUs to Oracle Cloud Infrastructure. (4 min read)
- Narmi launched a read-only banking data connector for Microsoft 365 Copilot. (3 min read)
Reality CheckWhat survives the model you picked
My record list started as a BASIC listing on a 5¼-inch floppy. My first PC could not read that floppy and there was no way across, so I retyped a few hundred records into FoxPro by hand over a few evenings. Nobody budgeted those evenings. The list was mine, the platform was not, and the platform went first.
Two buyers paid real money for that difference this week, while the loudest claim in enterprise AI says the opposite: take the big general model and the industry-specific part handles itself.
Which part of your AI stack would still be yours if the model underneath it changed next quarter?
My call: by the end of March 2027, at least one more large bank or insurer buys an AI company outright for its data and its workflows rather than signing a model licence, and says so in the announcement.
If you have the real number for what switching AI on cost you after the signature, the part that was never in the licence, hit reply and tell me.
Yves
Tomorrow, The Argument: one position, argued with the best case against it.
Reach data and AI leaders in this inbox: book a placement.
Ins7ghts Daily is published by 7wData.