Salesforce adopts Claude as default engine, while a survey denies effects on productivity

The industry news of August 28 in the language model sector is an integration between two companies that until now had a more limited commercial relationship. Salesforce and Anthropic have presented “Claudeforce,” an expansion of their collaboration: Claude is integrated inside Salesforce products and, in the opposite direction, Salesforce becomes accessible from within Claude (AI Agent Store).

The technically relevant point is that Claude is being adopted as the default reasoning model in Salesforce’s suite of agentic products, that is, in the systems that carry out tasks autonomously on behalf of the user (AI to ROI). Anthropic has also unified memory between Claude in conversation mode and Claude Cowork, so that context built in one environment remains available in the other (NeuralBuddies).

Why “default” is the word that matters

In the language model economy, the default position is worth more than any technical comparison. A company using a customer management platform rarely chooses which model runs beneath its automated processes: it uses whatever the vendor has set. Becoming the starting setting inside a widely used corporate suite means entering a number of workflows that no marketing campaign aimed at individual users could ever reach.

This is also why operations of this kind should be read as distribution agreements before being read as product announcements. However, the available material does not include the elements that would allow its weight to be measured: not indicated are the economic value of the deal, its duration, whether there is exclusivity, or which Salesforce products are involved and since when.

The data point going in the opposite direction

On the same day, a National Bureau of Economic Research survey on three years of AI adoption in companies reports that more than 90% of executives interviewed record no effects on their company’s employment, and 89% report no effect on company productivity. The same summary notes that the job cuts announced by companies have not slowed down.

The news currently comes from a single source (NeuralBuddies, which reports the NBER survey); no independent confirmation is available in the material at this newsroom’s disposal. The sample size, the method used to select the companies, the exact survey period and the wording of the questions are not known: these are the pieces of information that would determine how much weight to give the result, and without them the number should be treated as an indication, not as a measurement.

With this caveat, the juxtaposition of the two news items remains instructive. On one hand, corporate platforms are making the use of models the default behavior of their own products; on the other, the executives of those same companies, asked about three years of adoption, overwhelmingly state that they see no shift in either employment or productivity.

What we do not know

The full text of the agreement between Salesforce and Anthropic does not appear to have been published. It is not indicated whether Salesforce customers can replace the default model with another one, or under what conditions. On the survey side, the original NBER document is not accessible in the material consulted: until then, the reported percentages remain secondhand.