Beyond AI Skills: What Does Your Team Need to Learn When AI Can Do More of the Work?

Conceptual illustration of AI-supported work and team learning.

At Dreamforce on September 15, 2026, Salesforce introduced AIforce, extending access to its business data, workflows, and rules through AI interfaces. In the announcement, Marc Benioff described the shift simply: “AI is creating an interface revolution.” Salesforce’s announcement

It is an interesting moment. Software is becoming easier to work with through conversation, and AI agents are being given more ability to act.

The question I keep coming back to is what an organization needs to learn as those possibilities expand.

If an agent can qualify a lead, recommend a next step, or trigger a workflow, the team still needs to agree on what a good decision looks like. It needs a way to handle exceptions and learn from the consequences.

My view: as AI makes execution easier, an organization’s ability to define, evaluate, and improve the work becomes more valuable.

A concrete example: qualifying a sales lead

One example published alongside the Dreamforce announcements makes this practical.

Salesforce describes Siemens using two AI agents for inbound sales development. An engagement agent contacts incoming leads. Interested prospects move to a qualification agent, which gathers information and routes suitable opportunities to a seller. Salesforce reports that Siemens now engages all of these inbound leads across 132 countries. Salesforce’s Siemens account

That is a reported improvement in coverage. It does not, by itself, tell us what happened to opportunity quality, conversion, or seller productivity.

The example interests me because it makes the workflow visible: engage, qualify, hand off. Each step carries a business judgment that somebody has to define and maintain.

For another company considering a similar approach, those judgments could become the hardest part of the project.

The decisions behind the automation

Consider a hypothetical team introducing AI lead qualification.

Marketing considers a lead qualified when the company matches its target profile and the contact shows interest. Sales wants evidence of an active buying project. Customer success knows that one apparently attractive segment often needs unusually intensive onboarding.

All three perspectives contain useful information. The agent needs a working definition that reconciles them.

A faster handoff will carry whatever definition the organization gives it into more decisions. If the definition is weak, speed can make the weakness more consequential.

That creates practical questions:

These questions connect technology to the way the organization works. Answering them requires people to surface disagreements, examine real cases, and accept responsibility for improving the process.

How this has changed my thinking

At Demand Metric, much of our past week has involved thinking about how to make expertise easier to apply through research, frameworks, assessments, and practical tools.

A useful framework can make a complex problem easier to understand. An assessment can help a team identify capability gaps and decide where to focus its resources. Research can provide the evidence to make those decisions with greater confidence.

What has become clearer to me is how much value depends on what happens after people get access to those resources.

Jesse Hopps’s emphasis on ownership, coaching, and learning through real work has helped sharpen my thinking here. People need opportunities to apply a method, question it, and develop the judgment to use it independently.

My research and measurement perspective brings another question: what evidence would show that they are getting better?

Together, those ideas have moved my attention toward organizational capability: a team’s repeatable ability to make decisions, execute, and adapt.

AI makes this especially relevant. The work people need to learn changes when software can perform more of the steps.

Mindset, skillset, and toolset become very practical

The lead-qualification example gives these three terms a concrete meaning.

The dimensions depend on one another. Better data helps when people know how to interpret it. Analytical skill helps when people are willing to question their assumptions. A revised process helps when the systems can carry it into daily work.

This also gives coaching a specific purpose. An experienced coach can help a team examine a difficult handoff, identify the assumption behind it, and decide what to test next. The team develops judgment by working through a problem it actually owns.

Measure whether the system is learning

For the hypothetical qualification project, I would want to look beyond the number of messages sent or meetings booked.

I would track whether sellers accept the opportunities, why they reject them, whether accepted opportunities progress, and how much rework the handoff creates. Those measures need a baseline and consistent definitions; even then, changes in demand, lead mix, or seller behavior may affect the results.

I would also look at the team itself. Can it explain the qualification rules? Recognize an exception? Investigate a disappointing result? Improve the process without waiting for someone else to diagnose every problem?

A small, well-scoped project gives people a place to develop those abilities. They can test an approach, review actual handoffs, adjust the rules, and carry what they learn into the next cycle.

Research and peer learning support that process. They introduce evidence and alternative explanations. The team’s own experience then provides something concrete to compare, question, and improve.

That is how knowledge begins to accumulate as capability.

The Dreamforce announcements give us plenty to consider about what technology can do next. The question I would bring into the next leadership meeting is more immediate:

As AI takes on more of this workflow, what must our team get better at—and how will we know it is learning?