Sales edition

On 18 September, Bloomberg published an investigation into the air strike on 28 February on a school in Minab, Iran, which killed 123 children. Palantir's Maven Smart System helped compile the target list. That system uses Anthropic's Claude to rank targets semi-autonomously. Military staff at US Central Command assumed Maven would flag outdated or conflicting intelligence. That did not happen. Work that normally took hours was done by the system in minutes. Palantir says there is no evidence the software malfunctioned, and that the company is not responsible for the underlying data. It did add new checks meant to catch conflicting intelligence. This is not sales news, but it does affect your work. It is the clearest example so far of what goes wrong when people trust an AI recommendation without checking it. A buyer looking at an AI feature in your product will soon want to know who is still keeping watch. Make sure you can answer that question before it is asked. The key word is "ranking": the system does not invent anything new, it puts existing data in order of importance. Such a list always looks convincing, even when the underlying data is old or contradictory. Because it moves fast, the natural brake disappears too: work that took hours left room to doubt and double-check by phone, in a few minutes that room is gone. You see the same pattern in your own pipeline: a lead score or an AI-suggested quote looks just as certain as a list a person has checked for hours. The line sits at anything that goes out under your name: a price, a promise, a claim about what the product can do. If you trust an agent there, you cannot fix a mistake with a follow-up email. At the same time, doing everything by hand is not the answer either, since then the tool delivers nothing. Let AI handle the groundwork and the ordering, but keep a person responsible for every number and every promise. Be honest with your customer about where that line sits in your product, because that question is coming.

Source →