AI Agent Deployment Takes Minutes After the Setup Work Is Done

AI agent deployment takes minutes on platforms like Salesforce’s Agentforce Coworker because the agent copies the data, permissions, and workflows you already have when you switch it on. The work that decides whether it’s useful happens before that switch, so start your time-to-value clock on the day that prep begins.

At Dreamforce 2026, Salesforce told the room its new AI agent , Agentforce Coworker, could be “up and running in literally minutes.” A few minutes later, on the same stage, Stephanie Sadowski explained how Accenture got there. Sadowski leads Accenture’s Salesforce business group, so the consulting firm is both a Salesforce customer and a seller of Salesforce implementation work. Accenture switched Coworker on for 16,000 people overnight. Before that, it had been building AI agents since 2024 and “getting all the data right, getting the governance right.” Then: “Once we had done that, we could turn coworker on” (1. Salesforce, 2026).

Both statements are true. Salesforce was timing the switch. Accenture was describing the work that made the switch worth turning on, and that work never shows up in the minutes.

What do the minutes in AI agent deployment measure?

Salesforce’s own product page explains the speed. It says the agent “instantly inherits your Salesforce setup, maps your custom workflows and metadata, and respects your governance and sharing rules” (2. Salesforce, 2026).

Read that at face value. Turning the agent on builds nothing. It reads what your Salesforce account (your “org,” in Salesforce terms) already contains: the customer records, the rules about who can see which records, the automated workflows, and any agents you’ve already built. The switch takes minutes because all of that existed before anyone clicked it. And whatever condition it’s in is the condition the agent works from.

Salesforce’s admin team puts it more directly. Its launch post for admins says “the most important work you can do right now happens before you ever open Setup” (3. Salesforce Admins, 2026). The prerequisites come first, too. Before the setup screen does anything useful, the org needs Salesforce’s customer data platform (Data 360) running, an admin with the right permissions, and a license for every user (4. Singh, 2026).

The minutes are real, and they measure the last step of a much longer project.

Where did Salesforce’s featured customers spend their time?

Look at the customers Salesforce put in front of the Dreamforce audience. Where the public record shows a timeline, the fast rollout sat on top of earlier work.

Accenture was one. The Adecco Group, the staffing company, was another. It said its Coworker pilot “showed strong adoption in a matter of days” (5. The Adecco Group, 2026). The same announcement notes that Adecco had been running AI agents in its recruiting operations since April 2025. The pilot was fast because the agent arrived in a company that had already spent more than a year putting agents to work.

Fulton Bank, the Pennsylvania bank Salesforce featured in its Agentforce keynote, went from first hearing about Agentforce to running it in 1 hour, and had 3,000 employees using it within 6 weeks (6. Salesforce, 2026). That’s the strongest case for speed in the whole event, and it’s also the one where the timeline is least clear. The demo that followed showed Fulton’s Coworker handing a sales-coaching request to “an already built and specialized opportunity agent” (6. Salesforce, 2026). Nothing said on stage covers when that agent was built, or what Fulton had connected before its hour started. The hour is the only number anyone published, and it measures the switch.

The best counterexample comes from a practitioner. Andrew Russo, VP of business systems at BACA Systems, which makes robotic cutting machines for stone fabricators, called Coworker “the easiest thing I’ve ever setup for our entire Salesforce org” (2. Salesforce, 2026). I believe him.

Look at what he said it did, though. In a post quoted by Divaker Singh, a Salesforce consultant at the consulting firm Girikon, Russo wrote that Coworker could “pull together and navigate our complex sales and ERP data” (4. Singh, 2026). ERP data comes from the finance and operations system, which lives outside Salesforce. Coworker can only read it if that system is already connected, and wiring a finance system into a customer data platform takes far longer than a few minutes. Russo’s setup was easy because his org was ready.

What happens when the org isn’t ready?

An org that isn’t ready hands its problems straight to the agent. Robin Leonard, a partner at Xenai Digital, a Salesforce consulting firm in Sydney, described a financial services client that wanted autonomous customer service agents by the end of the quarter (7. Leonard, 2026). Then his team opened the org. They found “forty-odd record types hanging off the Account object,” case routing held together by “three generations of half-documented Flows,” and knowledge articles last updated during the pandemic. In plain terms, the same kind of customer was set up dozens of different ways, and nobody could fully explain how support cases got assigned. Salesforce hit the same problem on its own help site , where conflicting and outdated pages fed its own agent wrong answers.

That agent would have switched on in minutes. It would have inherited every one of those problems. Leonard’s verdict: “You get wrong answers, delivered faster, with more confidence, to more people, in more places” (7. Leonard, 2026).

Nothing about the switch would have warned anyone. The setup screen checks licenses and permissions, and that’s where its checks end. Whether Sales, Service, and Marketing agree on what a customer record means is a question only your own team can answer.

Where should the time-to-value clock start?

The budget is where this costs you. Vendors quote the setup time, and it’s easy for that number to become the start date in your business case . The data cleanup, permission rework, outside-system connections, and first agent builds then land in no budget line at all. So the payback period in the plan comes out shorter than the one you’ll live through, and the difference shows up when results are compared against the plan.

Start the clock on the day the prep begins. Count the prep as project cost. It makes the business case look worse on paper, and a few vendor reference numbers will look less impressive next to yours. The trade-off is worth taking, because the longer number is the one that still holds a year later.

Then, alongside the questions to answer about your own team first , add this question to your next vendor meeting: “When did your reference customer start the prep, and what did they already have connected?” If the answer is 2 years of agent work and a fully configured data platform, that’s your real timeline, minus whatever you’ve already done.