Four releases this week, from four unrelated corners of business, tell the same story. People use AI constantly now. They are far slower to let it do anything on its own.
Shoppers ask AI for ideas, then pay themselves
Visa’s 2026 Holiday Spending Shift Survey, run by Morning Consult among 1,016 US adults in late September, found that 54% have used AI for at least one shopping task, up from 47% a year ago. The jobs they hand over are practical ones: comparing prices (61%), checking stock (59%), finding discounts (57%) and suggesting gifts (54%). And 41% wouldn’t tell the recipient that AI picked the present.
Payment is where it stops. Only 23% trust generative AI to complete a purchase for them, according to Visa’s separate Trust Index. Gen Z is further along on all of it. More than eight in ten are open to AI help, and 73% would let it suggest gifts. The same group was hit hardest by scams last season, 39% against nearly three in ten overall, and three-quarters of all shoppers have changed how they buy because of fraud.
Visa’s interest is plain. Its own data says 61% of consumers trust Visa to secure a payment made by AI, and the company wants to be the step between AI as a shopping assistant and AI as a shopping agent.
Optimization teams use GenAI daily, and a third build it in
Gurobi makes a mathematical optimization solver used for scheduling, routing and supply chain planning. For its sixth annual State of Mathematical Optimization report it surveyed 382 professionals outside academia. Use of GenAI or LLMs in their optimization work jumped to 79%, from 30% in 2025. Only 37% use those tools systematically in development.
Where respondents expect GenAI to help is telling: code generation and debugging (68%), model formulation (67%) and talking to optimization models in plain language (62%). In every one of those, the solver still makes the decision and a person still checks the model. Automation is growing around that core. 45% already run optimization automatically in production, and 90% combine machine learning with optimization on at least one project, up from 81% in each of the previous three years. The share solving multi-objective problems, where cost and service pull against each other, rose from 32% to 40%.
That split makes sense. An LLM drafting a formulation that a solver then verifies is an easy sell. An LLM making the scheduling call itself is not.
Accountants are told to pick one workflow
At CCH Connections 2026 in Orlando, Wolters Kluwer put executives from Microsoft, Salesforce and McKinsey on its main stage with consultant Sandra Wiley to talk about agents that move from recommending actions to taking them. The advice was modest. Start with one workflow, secure the data behind it, and decide in advance when an agent should stop and hand the call back to a professional. Microsoft’s Lara Rubbelke said building an agent is the easy part, and fitting it into real work with the right context, governance and human judgment is the hard one. The panel ended by asking every firm to set priorities for the next 12 months and pick one move for the next 90 days.
German companies want AI on their own terms
ISG’s 2026 Provider Lens report on AWS partners in Germany found enterprises still expanding their AWS use, with GenAI changing how they handle cloud migration and application modernization. They also want security, compliance and data sovereignty, and they favor providers with established German operations. It’s the corporate version of the shopper who asks AI for gift ideas and then pays by card. Use the tool and keep control of where the data lives and who answers for it.
Roughly two to one
Put the numbers side by side. Fifty-four percent of shoppers use AI for shopping, and 23% would let it pay. Seventy-nine percent of optimization professionals use GenAI, and 37% have built it into how they work. Accounting firms are told to start with a single workflow. German firms buy AI services from partners whose data stays in Germany. Across very different users, about twice as many people use AI as trust it to act.
The companies selling agents know this, and they are selling trust as the product. Visa offers its brand as the safe layer for agentic payments. Wolters Kluwer offers governance built into the tax software. ISG’s top-rated German partners offer sovereignty. The next stage of AI adoption will be sold as a receipt: proof of what the agent did and who signed off.