01 / Agents
HubSpot now shows you every AI agent running in your business - on one screen
HubSpot switched on Agent Hub and Agent Builder in public beta on 23 July, available to every Professional and Enterprise customer. Agent Hub is a single screen listing every AI agent running across marketing, sales and service, whether each one is actually turned on, and what it is producing. Agent Builder is the part that matters more: a canvas where you describe a job in plain language and it assembles the triggers, workflow and agent behaviour using the customer data already sitting in your CRM.
HubSpot's own example is Ignite Reading, a virtual literacy tutoring provider. They built an agent that finds and parses each school district's academic calendar using the domain and school year already stored on the deal record. Fifteen to twenty minutes of manual lookup per district became seconds - more than 350 hours a year. Custom agents consume HubSpot Credits when they act, so this is not free at scale.
Owner takeaway: If you are on HubSpot Professional or Enterprise, open Agent Hub before you build anything new and list what is already running. Owners are routinely surprised to find a prospecting agent chasing an account while a service agent works an open complaint on the same account.
Original source: HubSpot - Meet Agent Hub and Agent Builder
02 / Microsoft 365
Copilot Chat in Outlook now reads your whole mailbox - and you do not need a Copilot licence
This one is buried in Microsoft's release notes and deserves more attention than it got. Copilot Chat inside Outlook has gone from reasoning over a single email thread to reasoning across a user's entire inbox, calendar, meetings and the other Microsoft 365 data they can already reach. Microsoft's note is explicit that this happens without requiring a Microsoft 365 Copilot licence.
For the many SEQ businesses who looked at per-seat Copilot pricing and quietly passed, that is a real capability arriving on the seats you already pay for. Staff can ask what a client said about a variation last month, or what they have committed to this fortnight, and get a usable answer.
It also means the amount of company data Copilot reads just widened for those users, gradually, without anyone signing a purchase order. Copilot surfaces only what a person could already open - which is exactly why loose mailbox and SharePoint permissions stop being theoretical the week this reaches you.
Owner takeaway: Before your team finds this on their own, ask whoever manages your Microsoft tenant to confirm two things: who has Copilot Chat enabled, and whether your SharePoint and shared-mailbox permissions still match who should actually see what.
Original source: Microsoft - Release Notes for Microsoft 365 Copilot
03 / Australian regulation
From 10 December, your privacy policy has to explain your automated decisions
From 10 December 2026, Australian businesses covered by the Privacy Act must disclose in their privacy policy where they use personal information in automated decision making that could significantly affect a person's rights or interests - both the kinds of personal information used and the kinds of decisions made. The obligation came in through the Privacy and Other Legislation Amendment Act 2024. The OAIC's consultation on its guidance closed on 15 June and the finished guidance is expected before the December start.
The test has three parts: a computer program makes, or does something substantially and directly related to making, the decision; that decision could reasonably be expected to significantly affect someone's rights or interests; and personal information is used in getting there.
Read that again and notice what is missing. Nowhere does it say the program has to be AI. A scoring spreadsheet or a rules-based screening tool can be caught. "We don't use AI" is not the answer to this question.
Owner takeaway: Write down every place a system decides something about a person in your business - application screening, credit or payment terms, tenancy, rostering, insurance triage. Anywhere a human is not genuinely making the call, that entry belongs in your privacy policy by December.
Original source: OAIC - Consultation on Guidance for Transparency in Automated Decision Making
04 / Automation plumbing
Your line-of-business systems can now talk to Copilot without copying your data
Microsoft's 15 July release notes contain two entries that matter to anyone thinking about connecting AI to their own systems. Agents built on the Model Context Protocol are now reachable from the Copilot pane inside Word, Excel, PowerPoint and Outlook, currently on the web. And admins can deploy and manage Federated Copilot Connectors - MCP-based connections to outside data sources, including custom line-of-business applications - from the Connectors tab in the Microsoft 365 admin centre.
The detail worth reading twice: Microsoft says these connectors use user-level authentication at runtime, which removes the need for organisational indexing and preserves the permissions already set on the source system.
That knocks out the standard objection to plugging a job-management or accounting platform into Copilot, which was that you ended up with a second copy of your data somewhere you could not see or govern. It is now cheap to trial one connection properly.
Owner takeaway: Ask your IT provider one question this week - which of our systems has an MCP server available today? If the answer includes your job management, accounting or ticketing platform, that is your first low-risk pilot.
Original source: Microsoft - Release Notes for Microsoft 365 Copilot
What I would do first
There is a pattern across all four. Gartner put worldwide spending on AI models and platforms at $64 billion for 2026, up 63.4% on last year, and its analyst was blunt about where the pressure now sits: budgets under scrutiny, focus on usage efficiency, cost control and measurable outcomes. Meanwhile Arctera's governance survey of 500 professionals in regulated industries found 55% had the policies and training in place but only 19% had the logging and retention needed to prove what their AI actually did.
Policies are easy. Evidence is not. And the December privacy obligation is an evidence requirement dressed up as a paperwork one.
So the first practical test would be a single page, written this week, that lists every automated or AI-assisted step already running in your business - who owns it, what data it touches, and whether anyone would notice if it went wrong. Not a strategy. A list. Most owners cannot produce one, and every item above gets easier once you can.
Do not automate the mess. Map it first.
Running a business that needs someone accountable for technology decisions, not just automations? See the fractional CIO seat.