This week's news: Copilot seats start billing by the task, the Fed raises rates with AI borrowing in the mix, the note-taker on your calls becomes a wiretap question, California puts a human back in employment software, and Google moves your search budget onto its own judgment.
Read the whole thing in three minutes, or go deeper where it touches your business. We talk about the market here, not about ourselves.
Every article comes with a working tool. It runs right here in the page, it is free, and we do not need your information to use it. No signup, no download, nothing to fill in first. That holds for every article in every issue, back issues included.
Send it when it hits. Every article here has its own link; the send button next to each one copies it, ready to text or email to the person who needs it. Back issues stay in the archive and their tools keep working, so nothing you forward goes stale.
Want a reminder when the next one is up? Leave an email and we send one note a week. That is the whole program. We will not flood your inbox.
Microsoft posted on September 16 that from November 2, new Microsoft 365 Copilot Business licenses bought through a reseller ship with pay-as-you-go billing switched on by default. The metered services named are Copilot Cowork, Work IQ APIs and GitHub Copilot Harness. The seat itself lists at $18 a user a month on annual terms. Cowork bills in credits at about a cent each, and a heavy task runs past $7.
The seat price used to be the ceiling on the line item. After November 2 it becomes the floor. At 150 seats, a quarter of your people running two heavy tasks a day is five figures a year that never went through a purchase decision. The default configuration allows every new pay-as-you-go agent, so what your staff can spend is set by whoever releases the next agent, not by you.
Do not sign a Copilot purchase or renewal through your IT provider after November 2 without written spending caps in place first. Do not assume the reseller configured limits; the default is on, and the setup is automatic. Do not approve metered agents company wide because a pilot group liked them. Cap by group, review the first invoice line by line, then widen.
Seat licensing had one useful property for an owner: the bill was a headcount times a price, and the controller could forecast it to the dollar. Metered agent work breaks that. Cost now tracks how many tasks people hand to the software and how heavy those tasks are, and neither number sits in any system your finance team reads. Microsoft does let an admin set a hard cap for the tenant, a group or a single user. The problem is that the cap is optional and the billing is not.
Treat the November 2 date as a contract term, not an IT setting. Before any new Copilot Business seats land, have your IT provider set a tenant ceiling and group caps, and confirm in writing who can release a new pay-as-you-go agent. Keep the ceiling low for the first ninety days and let actual usage argue for more. The trade is some friction for heavy users in exchange for knowing your number. At this size that trade is worth making every time.
On September 16 the Federal Reserve raised its target range a quarter point to 3.75 to 4 percent, unanimously, the first hike since July 2023. Wells Fargo moved prime from 6.75 to 7.00 percent effective September 17. Twelve of eighteen officials project another quarter point this year. Chair Kevin Warsh named AI hyperscalers raising funding among the forces pushing yields up, and the 10-year Treasury crossed 5 percent.
Every floating dollar on your revolver or term loan just got more expensive, and variable lines typically reset within one or two billing cycles. On $10 million drawn, a quarter point is $25,000 a year, and the projected next move doubles it. The projections also show no cuts through 2027, so the hurdle rate on every capital project, including your own automation spend, went up with it.
Do not wait for the October 27 to 28 meeting to call your banker about a swap or a fixed-rate tranche. Do not recheck covenants against last quarter's interest expense; run them at the new prime plus one more quarter point. Do not approve projects this quarter at a hurdle rate set when money was cheaper.
For two years the rate story was about inflation at the grocery store. This one has a new buyer in it. The largest AI companies are raising capital at a scale that competes with everyone else for the same dollars, and the Fed chair said so when he explained why long yields are rising. The equipment that buildout needs is also getting more expensive. You do not have to own a data center to pay for one. You only need a loan priced off prime.
The practical response is to decide how much floating exposure you want to carry into 2027, because the projections say rates stay here. Ask your lender for a quote to fix or swap part of the balance before October 27, and compare it against two more quarter points on what stays floating. Rerun interest coverage at the higher rate. If headroom gets thin, have that conversation with the bank now, while it is your idea.
On August 13 a federal judge in San Jose let wiretap, California privacy and Illinois biometric claims against Otter proceed, finding the vendor plausibly collects and uses call audio for its own purposes, including training. A class action against Granola under the same theory was filed July 30. A ruling on Fireflies is pending. Damages run $5,000 per California violation and up to $10,000 under the federal Wiretap Act.
If the vendor is a party to the call, then everyone on the line has to agree, not just your employee who invited the bot. Your sales team, recruiters and account managers are running these tools on calls with California and Illinois participants every day. The vendors' own terms put the notice duty on the user, which means the exposure sits with the company that deployed the tool, not the one that built it.
Do not assume the meeting platform's recording banner covers a third-party bot. Do not let staff choose their own note-taker on free or personal plans. Do not keep a tool that trains on your audio because it is cheaper than the enterprise tier that does not. Pick one approved tool, get the training terms in writing, and script the consent line.
California requires every party to a confidential conversation to agree before it is recorded. For years the working assumption was that the host's click satisfied that, because the host was the one recording. The Otter ruling moves the question to the vendor. If the software keeps the audio and uses it to improve its own models, it is arguably a separate listener, and the other people on the call never agreed to it. Illinois adds a separate exposure because a voiceprint can count as biometric data.
The fix is cheap relative to the damages math. Standardize on one note-taker, on a plan whose contract says your audio is not used for training. Have the bot announce itself, and have your people say one consent sentence at the top of any call with an outside party. For recruiting calls and anything involving California or Illinois participants, default to no recording unless the person agrees on the record. Then check the free accounts, because that is where the risk actually lives.
Governor Newsom has until September 30 to act on SB 947, which bars relying solely on automated systems to discipline or fire workers, with $500 per violation and a private right of action from July 1, 2027. Separately, California's automated decision rules take effect January 1, 2027 for businesses above $26.6 million in revenue, covering hiring, pay, promotion and termination. Colorado's replacement AI law starts the same day.
Much of this readership crosses the California revenue threshold. From January 1, any tool that screens applicants, allocates shifts, sets pay or flags discipline needs a pre-use notice, an opt-out or real human review, and four years of kept inputs and outputs. If SB 947 is signed, workers can sue over firings the software drove. Your vendors will not carry that for you.
Do not wait for the veto decision to start the inventory; the January 1 rules apply either way. Do not accept an applicant tracking or scheduling renewal without asking what the vendor logs and for how long. Do not describe a manager clicking approve on every recommendation as human review; Colorado asks for meaningful review, and a rubber stamp is not that.
Most owners at this size could not list every place software makes or shapes an employment decision in their business. The applicant tracking system ranks candidates. The scheduling tool assigns hours. The performance platform flags people. Each of those is now a regulated decision point in California from January 1, and in Colorado too, with notice, explanation and record-keeping duties attached. Newsom vetoed a broader version of SB 947 last October, so a second veto is possible. The January rules do not depend on him.
Start with a register: every tool, what decision it touches, which states it covers, and who reviews the output. That list tells you where to add notices, where human review is real and where it is a rubber stamp, and which vendors need to show you their logs. Budget for counsel to review the notices, not to build the inventory; your HR lead can do that in a week. The expensive version is doing it after the first demand letter.
Since September 1, Google has been converting campaigns that use campaign-level broad match or automatically created assets into AI Max, aiming to finish by the end of September. Search term matching switches on by default, and for automatically created assets campaigns, so does AI-written ad text. Google cites 7 percent more conversions at similar cost. An independent study of more than 250 campaigns found cost per acquisition up a median 16 percent.
Your ads are now showing on searches you did not choose, and the report that tells you whether that worked comes from the company selling the clicks. At $50,000 a month in search spend, a 16 percent rise in acquisition cost is roughly $96,000 a year for the same customers. The switch happened in your account whether or not anyone at your agency mentioned it.
Do not assume your agency opted you out; the migration runs in place and the defaults favor more matching. Do not judge AI Max on Google's conversion column alone; compare cost per acquisition against your own CRM for the same weeks. Do not let AI-written ad text run on regulated claims or pricing without someone approving it.
The mechanism is simple. Search term matching lets Google's model decide which queries are close enough to your keywords to buy. More matching means more impressions and more spend, and the headline 7 percent lift is Google's own measurement on its own data. Independent readings disagree: higher conversion value in some retail accounts, but at a higher cost per acquisition, and one study found invalid traffic doubling on AI Max over nine months. None of that means turn it off. It means someone other than Google has to check.
Ask your agency for a list of every campaign that was converted and the date it happened. Pull cost per acquisition from your own CRM for the four weeks before and after. Where cost rose and qualified leads did not, turn search term matching off for that campaign, or ring-fence it with brand and URL exclusions. Keep it where it earned its place. The decision should rest on numbers you own.
The Briefing arrives weekly: what moved, what it changes, what not to do, and a working tool for each. Every issue gives the tools away, free and built for the week they cover. No product tours, no victory laps. If it stops earning the read, unsubscribe.
This is back issue No. 010. Every tool in it still runs. Read the current issue · Open the archive
If something in here lands on a problem you are already working, and you would rather not work it alone, write to hello@goudegroup.com and tell us what you are exploring. You get a reply from a person, not a sequence.