Market Pulse
Market Pulse
Where Inertia Goes When Agents Do the Checking
In December 2023, Lloyds paid 1.40% on an easy-access savings account; Metro Bank paid 5.22% on a similar one. Gaps that wide only last if many customers never compare. Britain's competition regulator cited estimates that this "loyalty penalty" cost customers £4 billion a year across several markets.
Singapore's central bank chief suggests customers' AI agents could move money for smaller differences in incentives. If so, the margin built on not looking could shrink, or it could move: to friction at the bank's door, or into whatever an agent counts as best.
Where Inertia Goes When Agents Do the Checking
In December 2023, Lloyds paid 1.40% on an easy-access savings account; Metro Bank paid 5.22% on a similar one. Gaps that wide only last if many customers never compare. Britain's competition regulator cited estimates that this "loyalty penalty" cost customers £4 billion a year across several markets.
Singapore's central bank chief suggests customers' AI agents could move money for smaller differences in incentives. If so, the margin built on not looking could shrink, or it could move: to friction at the bank's door, or into whatever an agent counts as best.
Research Summaries
Generative AI Without Guardrails Can Harm Learning (Bastani et al., PNAS, 2025)
Students mostly copied answers and did not notice their learning slipping. The authors argue that this learning is what lets people catch fallible AI output.
The harm came from the interface. A tutor that made students attempt problems first erased the exam penalty, though it produced no gain.
Research Summaries
Endoscopist Deskilling Risk After Exposure to AI in Colonoscopy (Budzyń et al., Lancet Gastroenterology & Hepatology, 2025)
A 2026 prospective trial in Endoscopy, covering 13 endoscopists and 5,013 procedures, found unassisted detection returned to baseline once AI was withdrawn.
The accompanying Lancet commentary warned that losses surface when the tool is absent, and named trainees and clinicians moving between systems.
Research Summaries
Generative AI at Work (Brynjolfsson, Li and Raymond, Quarterly Journal of Economics, 2025)
The largest gains went to the least experienced. Outage results suggest some of what the tool modelled stuck, which favours assisted juniors over no juniors.
Agents accepted only 35% of suggestions, so they were filtering constantly. Outages measured resolution speed, so whether they filtered well went unmeasured.
Research Summaries
AI Assistance and Unaided Follow-Up Performance on a Workplace Task (Cruces et al., NBER Working Paper 34851, 2026)
Usage intensity was similar across groups, yet 64% of submissions were pure copy-paste, and less-educated participants were ten points likelier to paste wholesale.
The paper is not yet peer-reviewed, and retention was tested after minutes. Less-educated users still trailed more-educated peers, so expertise kept mattering.
Short Read
An AI agent can only warn you about something after it looks, and how often it looks is a cost decision. Anyone who's run production alerting learned this years ago. Agents are learning it now.
Anthropic's helper checks your accounts every few hours, because checking constantly would burn through your plan. Google is building infrastructure so an idle, waiting agent costs almost nothing. Shopify launched agent shopping that can't tell the agent when an order changes, so the agent keeps asking.
When a company calls its agent "proactive," meaning it acts before you ask, somebody is paying for all that looking. Subscribers pay in usage limits, developers pay per status check, and the cloud provider pays to keep idle agents cheap.
A cheap sleeping agent still needs something to wake it up. If services sent a signal the moment a delivery ran late or a meeting moved, agents could sleep until something real happened. Whoever controls those signals may end up selling the expensive part, the noticing.
