AI agents are coming for drudgery, not decisions

3 min read.

Author

Midas AI
August 12, 2025

CFOs at Salesforce and PayPal show why human plus agent workflows beat hands off automation

Finance leaders are moving past generic AI enthusiasm and into operational design. In a roundtable with reporters, finance chiefs from Salesforce and PayPal detailed how they are deploying agentic AI and what that means for skills, workflows, and budgets. Their message is consistent. Autonomous agents should shoulder routine work while humans focus on analysis and decisions. That approach is showing measurable gains and shaping how budgets get allocated.

What finance leaders are doing now

Salesforce defines this direction as digital labor, a hybrid workforce where humans and autonomous AI agents work together. The company is leaning into that with Agentforce for Sales, which handles routine tasks such as lead follow up and outreach around the clock. A Salesforce spokesperson said the agent recently analyzed more than 7,800 unworked leads to prioritize high intent prospects, personalize messaging and book over 50 meetings, generating $460,000 in net new pipeline in just three weeks. The tool’s positioning is clear. Let AI do the legwork so reps can sell.

The enterprise appetite for this model is growing. In a June survey of 261 global finance chiefs conducted by Salesforce and Morning Consult, about one third reported an aggressive approach to AI, 63 percent a moderate stance, and 4 percent a conservative one. Respondents said they allocate, on average, about one quarter of existing AI budgets specifically to AI agents, and 61 percent called AI agents and digital labor critical to competition. About one third also said AI requires them to have a riskier mindset around technology investments. Even among cautious stewards of capital, the signal is that agent spend is becoming a strategic line item, not a science project.

Inside the finance function, the use cases are pragmatic. PayPal’s finance chief Jamie Miller said the company is using AI agents for first drafts of internal content and executive level materials, including summarization of strategy. PayPal is not using AI to generate SEC filings such as 10 Ks, but Miller expects the company will work its way to some form of assistance on that. The team’s focus is on leverage where it counts today, with guardrails where the stakes are highest. Miller also emphasized what the workforce needs to thrive alongside AI, including curiosity, systems thinking, and self starting.

At Salesforce, finance chief Robin Washington described how AI is already changing her preparation for analyst interactions and competitive analysis. She said AI has reduced her earnings prep time to hours instead of days, and that the technology is being used to astutely understand the landscape. Washington framed the organizational shift plainly. Her boss says they are probably the last set of managers to just manage humans, and they will soon have AIs autonomously working side by side helping them make decisions. That does not diminish human responsibility. In her view, AI will reinforce the need for problem solving, emotional intelligence and higher level decision making. As she put it, AI brings the data directly to leaders and enhances their ability to do the art of the possible.

Design rules for a human plus agent model

For founders and operators building AI products, this is a design brief. Agentic AI should be engineered to remove toil and surface signal, not to take over judgment. Salesforce’s Agentforce for Sales is framed explicitly as a way to help reps spend more time selling by letting AI handle routine tasks. PayPal’s use of agents for internal drafting and summarization shows how to accelerate work without outsourcing accountability. The survey data reinforces that CFOs are already budgeting for agents and see them as critical to staying competitive, even as they acknowledge that adopting the technology will push them toward a riskier investment posture.

Designing for a hybrid model also aligns with how leaders plan to develop talent. If the aim is to amplify problem solving and systems thinking, then product choices should bring context and evidence to the forefront, keep humans in the loop on decisions, and make it easy to inspect what the agent did and why. That is the path to time savings like Washington’s, to pipeline lifts like Salesforce’s sales example, and to organizational confidence that AI is a force multiplier rather than a black box.

The takeaway is straightforward. The enterprises writing checks for AI are not chasing full automation. They are building digital labor where autonomous agents and humans work together to accomplish more than either could on their own. Founders who design for that reality will meet the moment.