Copilot Gets More Expensive July 1, and Three Other AI Moves That Reach Your Back Office

This week the AI news reached the back office. Microsoft changed what you pay for Copilot and what it does inside Excel. OpenAI changed what ChatGPT remembers about your people. Here are the four stories that touch a distributor, and what to do about each.

Microsoft raises Copilot prices and puts agents in Excel

What happened: Microsoft 365 Copilot Business stays at $18 per user per month through June 30. On July 1 it goes to $21. Agent Mode is now fully available in Excel across web, Windows, and Mac, and it can run on either OpenAI or Anthropic models. Custom agents built in Copilot Studio bill on consumption credits, not a flat seat price.

What it means for you: If you run Copilot across your office staff, do the math before June 30 and lock current pricing if it favors you. The bigger story is Agent Mode. Your team already lives in spreadsheets for quoting, margin work, and inventory. An agent that builds those sheets, repairs broken formulas, and audits the numbers saves real hours, as long as the data going in is clean. Watch the consumption credits on custom agents. Open-ended usage runs up a bill fast, so set a cap before you pilot anything.

Sources: Microsoft 365 pricing updates, Agent Mode in Excel is now generally available (Microsoft)

ChatGPT now remembers what your team types

What happened: On June 4, OpenAI rolled out Dreaming V3, the biggest change to ChatGPT memory since the feature launched. A background process reads across years of past chats, synthesizes them, and keeps a running profile of each user. OpenAI cut the compute about 5x and is bringing memory to the free tier for the first time.

What it means for you: This is a data security question, and in our industry that is question number one. Anything a rep types into a personal ChatGPT account now persists and becomes part of a profile. Supplier pricing, customer formulas, SDS details, margin notes. None of that belongs in a consumer chatbot’s long-term memory. If you have not told your team what goes in the box and what stays out, this is the week to do it. Free-tier memory means more of your people are using it, not fewer.

Source: AI Update, June 5, 2026 (MarketingProfs)

ERP agents are starting to act, not just suggest

What happened: SAP is rolling AI agents across order management, warehousing, maintenance, and service through Q2 and Q3. These agents read the ERP, the warehouse system, and the transport system, then trigger actions without a person clicking each step. Industry surveys now put AI use among manufacturers at 94%, with this year’s shift moving from pilots to live operations.

What it means for you: Whether you run SAP or a smaller ERP, every vendor is heading this way. An agent that reads your order and inventory data and acts on it is only as good as that data. A distributor with clean, current records gets value. A distributor with three versions of the same customer and stale lot quantities gets faster mistakes. Fix the records before you turn on the agent.

Sources: SAP at Hannover Messe 2026 (ERP Today), Manufacturers move from AI pilots to operations (Digital Commerce 360)

Anthropic ships its most capable model, and it reads your paperwork

What happened: On June 9 Anthropic released Claude Fable 5, its most capable model for public use. Two things in it matter for an operator. It reads diagrams, charts, and tables buried inside PDFs, which every earlier model handled poorly. It also runs as an agent for days at a time, planning a job in stages and checking its own work. It is free for Pro, Max, Team, and Enterprise users through June 22, then billed by usage.

What it means for you: Your business runs on documents that lock data in formats software has always choked on. SDS sheets, certificates of analysis, supplier spec sheets, price lists sent as flat PDFs. A model that pulls the numbers out of a table inside a PDF is the first version of this technology that touches that pile directly. Try it on one real task this month. Hand it a stack of supplier COAs and ask it to put the assay values in one sheet. If it holds up on your documents, you have found a use that pays for itself. One number from the same week tells you the technology is here to stay. Anthropic reported that more than 80% of the code merged into its own production systems in May was written by Claude.

Sources: Anthropic releases Claude Fable 5 (TechCrunch), Anthropic says 80% of its production code is authored by Claude (VentureBeat)

One more number from this week. Industry reports count roughly 600,000 open jobs across US supply chain and manufacturing, with skilled people retiring faster than they can be replaced. This is what these tools are good for. They capture what your veterans know before they walk out the door, and they free up the people you have for the work only people can do.

Source: Manufacturers move from AI pilots to operations (Digital Commerce 360)

If you have started letting an AI agent touch your ERP or your quoting, I want to hear what broke first. That is usually where the lesson is.

Pete