Something shifted this week, and I want to name it clearly.
On May 5, Anthropic announced that Claude agents are now in production at JPMorgan Chase, Goldman Sachs, Citi, AIG, and Visa. Not pilots. Not proofs of concept. Production deployments — agents running real workflows inside the largest financial institutions in the world. PwC followed on May 14 with a partnership to embed Claude agents into enterprise client work across finance and life sciences.
At roughly the same time, Salesforce's Agentforce reached the point where it is not just suggesting pipeline updates — it is making them autonomously, watching for stale stages, pulling external signals, and updating records without a human in the loop.
I have been selling AI products since before most people knew what an LLM was. I have watched three distinct waves of enterprise AI adoption. The first was “let's explore.” The second was “let's pilot.” The third — which is where we are right now — is “let's run this in production.” And the third wave changes everything about what sales leadership actually means.
What just changed
The distinction that matters here is not between AI tools and AI agents, even though that framing is everywhere right now. The distinction is between AI that assists a human decision and AI that executes a multi-step workflow autonomously.
For the last two years, the enterprise AI story was mostly the first kind. AI that drafts your outreach. AI that summarizes a call. AI that suggests next steps. These are legitimately useful, and I use them every day. But they are still fundamentally about making a human faster at something a human was already doing.
What Anthropic and Salesforce are shipping now is the second kind. Agents that watch a portfolio of deals in real time, identify the ones that have gone quiet, pull Gong transcripts and email threading data to understand why, update the CRM accordingly, and flag the rep when human judgment is actually required. The human is still in the loop — but the loop is much bigger. The agent handles the 80 percent that was just process; the human handles the 20 percent that actually requires relationship intelligence.
That is a fundamentally different operating model for a sales team.
What this means for how you manage
If you are running a sales team right now and you are not actively thinking about which parts of your process should be handed to an agent, you are making the same mistake companies made in 2010 when they thought of email marketing as “the thing marketing uses” instead of the primary customer retention channel.
The practical question is not “how do I buy an AI product.” The practical question is: what are the three most time-consuming things my reps do that do not require their specific relationship context? Those are the candidates for agent automation. Qualification follow-up, CRM hygiene, competitive research, initial outreach personalization — these are all in play. Sellers using AI are already 3.7x more likely to hit quota. The next delta is the ones whose entire workflow is agent-augmented versus the ones still using AI one prompt at a time.
The deeper question, and the one I find more interesting, is: if agents handle the top-of-funnel process work, what does that do to your hiring model? The entry-level sales job has been under pressure for two years. If an agent can run 1,000+ qualified touches a day at a cost per contact an order of magnitude lower than a human rep, the SDR motion looks different. Not gone — but different. The humans who thrive are the ones who can manage agents, interpret what the agent is surfacing, and engage at the moments where relationship credibility actually matters.
This is not science fiction. Anthropic is running it at Goldman Sachs right now.
The window is still open — but it's closing
What I keep telling the founders and sales leaders I work with is that we are in the narrowest window of competitive advantage from agentic AI, and most organizations are not moving fast enough through it.
The companies that implement agentic workflows in the next 12 months will have a structural cost and speed advantage over the ones that wait until it becomes obvious. By the time it becomes obvious, the vendors will have raised their prices, the talent that knows how to implement this will be expensive and scarce, and the organizational habits needed to work alongside agents will take 18 months to build.
The Anthropic–Wall Street story this week is not just a press release. It is a signal about where the enterprise AI adoption curve is. The largest, most compliance-heavy, most risk-averse institutions in the world just put AI agents into production. If JPMorgan can do it, there is no credible excuse for your RevOps team not to at least run a proof of concept this quarter.
The concrete takeaway: pick one sales workflow this week — just one — where the output is predictable and the judgment required is low. Map the steps. Then ask whether an agent could run those steps with a human reviewing the exceptions. Qualification outreach. Competitive battlecard research. Deal hygiene notifications. If the answer is yes, you have just found your pilot. Start there, measure what changes, and build from that foundation.
The demo phase is over. The organizations that treat this month as the starting gun will look very different in 24 months from the ones that are still running webinars about AI readiness.