What the conference was
SaaStr AI Annual 2026 took place May 12–14 at the San Mateo County Event Center, in the San Francisco Bay Area. It calls itself the world’s largest B2B and AI event. The agenda page lists sessions such as “AI Agents in Production: Scaling Beyond the Hype” and “AI-Powered Support → Revenue Engine”, and it is mostly about sales and go-to-market teams. On the agenda page we read in October 2026, we did not find a session on field service, trades or operations.
That is not a criticism. It tells you whose problems the headline examples are written for. If you run crews, trucks or a service desk, you have to translate them.
The session with numbers
The clearest figures came from a Salesforce session on small business. Its description says Salesforce has closed 29,000 Agentforce deals in 15 months, and that annual recurring revenue from its AI agents jumped 330% year over year to over $500 million. These are the speaker’s figures as printed in the session description, not independent measurements.
The same description says the talk covers why small business became Salesforce’s fastest-growing segment, how Agentforce works in a 20-person company compared with a 2,000-person one, first-win use cases (sales-development automation, customer service, onboarding and routine workflows) and why data quality matters more than which model you choose.
Three ideas that carry over to an operations firm
- Give each agent a named job. “AI everywhere” is not a plan. A sales team’s first agent usually writes outreach; an operations firm’s first agent usually answers a customer or writes up finished work.
- Records first, agents second. The session’s own point about data quality is the one to keep. An agent that reads a client record with the timeline, the work, the money and the files can answer from it. An agent pointed at three half-updated spreadsheets cannot.
- Start with one simple, high-value use. The session recommends starting small to build trust. For most firms that means one agent doing one job, with a person checking the first few results.
| The job | A sales team’s example | An operations firm’s example in Operations Link |
|---|---|---|
| Answering a caller | Qualifying a new prospect | Vera answers a customer you already have about their own job, invoice or quote, after checking who is calling (a service) |
| Writing the paperwork | Drafting outreach | The invoice drafts itself when a field job is done, and you send it (with Field operations, a service) |
| Preparing the day | Meeting prep notes | Nova’s morning brief: what changed, what needs you, what is coming (fifteen days free, then a service) |
| Checking the work | Scoring a deal | A pay run goes through two approvals before the bank file is made (Payroll, a service; Ontario, Canada today) |
What these agents do not do
Operations Link is not a prospecting tool, and its agents are narrower than the sales-agent examples at the conference. Vera does not book appointments, take messages or qualify new callers; she is English only, phone numbers are Canadian today, and she is a service a company arranges with our team. Nova cannot approve bills, send invoices or payments, approve pay runs or change roles. If you need an agent that takes a first call from a stranger and books the visit, other vendors sell that, and our agent comparison lists several.
Questions to bring to any vendor
- Who does this agent talk to: a stranger, an existing customer, or my own staff?
- What can it finish alone, and what waits for a person?
- What happens when it is wrong, and who sees it?
- How is it paid for, and is that written down?
Our side-by-side pages cover the rest: see Salesforce, HubSpot and the full list.
Frequently asked questions
What did SaaStr AI Annual 2026 cover about agentic CRM?
The conference ran from May 12 to 14 in San Mateo and was mostly about sales and go-to-market teams. Its agenda included sessions on AI agents in production and AI-powered support, and a Salesforce session on selling agents to small business. We did not find a session on field service, trades or operations.
Did SaaStr propose a follow-the-agents framework?
We could not find a follow-the-agents framework on the SaaStr conference pages, so we do not attribute one to SaaStr. The phrase is our own shorthand for choosing software by what its agents do for your customers and your crews rather than by feature list.
What does the agent trend mean for operations-driven small businesses?
Pick one agent for one job, make sure it works from clean records, and check its first results yourself. The jobs that matter most to a firm with crews are usually answering existing customers, writing up finished work and laying out the day.
What is Vera, and which agent should an operations business start with?
Vera is the agent on your phone line and in your client portal. She answers customers you already have about their own jobs, invoices and quotes, after checking who is calling. She does not book appointments, take messages or qualify new callers. Vera suits existing customers who keep calling the office for status; if new callers are the gap, look at an agent from another vendor that takes new requests.
What is Nova, and how is she different from a chatbot?
Nova is the AI assistant you talk to inside the app. She writes the morning brief and answers questions from your own team documents. She works inside the app and is held to rules: she cannot approve bills, send invoices or payments, approve pay runs or change roles. She has fifteen days free when you start, then she is a service.
Related reading: the SaaStr AI Annual 2026 preview, self-driving CRM, a voice AI comparison chart and four AI-native CRMs compared.