What changed this week — and what it means

So what · This lands on the same desks as the AI-restructuring question. If any 2027 workforce plan touches roles held by H-1B or similar visa holders, model the rule as if final: a redundancy notice becomes a departure notice, portability becomes a race, and your relocation and severance policies may need a visa-specific track. Two things to do before November 10: pull a count of affected employees by visa type and expiry date, and decide with counsel whether to file a comment — trade groups will, and employer data on mobility costs carries weight. Pair this with last week's SB 951 90-day AI-layoff notice: for a California employer, the two rules together turn "who is affected" into the first question in any restructure.

So what · The HR-tech question isn't whether Paige is good; it's where she sits. Salesforce is positioning the employee help desk — the front door to HR — inside Slack, with HiBob as a likely system of record behind it. If your HRIS vendor also sells an HR agent (Workday, SAP, Oracle, UKG all do), you now have two candidates for the same tier-0 experience, and only one can own the conversation log. Decide the front door deliberately: which agent employees talk to, which system holds the transcript, and what happens when Paige needs to write to payroll. And read the long-horizon runtime as a preview — agents that remember across weeks are agents that accumulate employee context, which is exactly what your data-retention policy was written without.

So what · Four hundred agents is a catalog, not a roadmap. For a SuccessFactors customer, the useful question for your account team is which of those agents can write to Employee Central, what approval step sits in front of each, and whether they're licensed in your existing Joule entitlement or metered separately. The vendor's own numbers hand you the governance argument: if 97% of organizations say they're not ready, "we'll turn on the agents when we have the review process" is a defensible position, not a laggard one. And route the Patch Day note to whoever owns your SAP basis team — HR data is the payload.

So what · If your AI screening vendor promised a faster funnel and your time-to-hire is flat, you're in the majority — and the data says the bottleneck moved, not disappeared. AI clears the top of the funnel faster, and then the candidates aren't there. Two implications: measure time-to-hire by stage so you can see where the days went, and put the "62% hiring for changed roles" number in front of the L&D budget owner — a role that's changing is a role you can fill from inside faster than from a thin market. Our analysis of which skills the market is actually paying for.

So what · Put this next to the SAP and Salesforce items above and the picture is clear: vendors are shipping hundreds of agents into functions where nine in ten HR teams are still at "limited use." That's not a reason to wait; it's a reason to pick one workflow — recruiting screening or L&D content, where your peers already are — and operationalize it properly, with a named owner and a validation step, before adding a second. On retention: if turnover is a business risk at 57% of companies, HR's AI business case should be written in retention terms (manager time freed, faster internal moves) rather than headcount terms. It's the number the CFO is already watching.

So what · Inventory first: which notetakers are in use in recruiting, who turned them on, and do they announce themselves? Then adopt the most stringent jurisdiction's standard everywhere — all-party consent, spoken and logged at the top of every interview — because it's operationally simpler than a state-by-state matrix. Two contract questions for the vendor: is interview audio used to train models, and can we set retention to match our candidate-data policy? And note the connection to California's AB 1883 (still on Newsom's desk as of today): a transcript that's later run through sentiment or "engagement" scoring is emotion inference, and it's about to be regulated.