Key Highlights

● Salesforce unveiled Agentforce Coworker, job-ready agents and Koa, its first in-house reasoning model, at Dreamforce.

● Second-quarter fiscal 2027 revenue was about $11.3 billion, up 11%, in an August 26 release.

● Agentforce annual recurring revenue topped $1.5 billion, up more than 240% year over year.

● Raised full-year revenue guidance of $46.1 billion to $46.4 billion includes pending acquisitions.

Salesforce (NYSE:CRM) has spent the past year trying to show that its artificial intelligence agents can move from demonstrations to daily work inside large companies. At Dreamforce, held over three days at the Moscone Center in San Francisco, the company laid out its latest answer in a recap of 21 announcements covering new agents, a reasoning model, voice features and tools for measuring how well agents behave.

The conference arrived a month after the software maker reported what it called a record second quarter of fiscal 2027. Revenue grew in the low double digits, while the company's own agent product line grew at a much faster pace from a smaller base. That contrast, steady growth in the core business and rapid growth in a newer one, frames much of the discussion around the company.

What are the headline Dreamforce announcements from Salesforce?

According to Salesforce's recap, the central theme was a package the company calls AIforce, which is meant to bring its software into the places where people already work: Claude, Slack and Salesforce itself. Claudeforce, the version tied to Anthropic's assistant, is in beta, Slackforce is due to reach general availability in October, and Agentforce Coworker is already generally available.

Agentforce Coworker is described as an autonomous teammate that can answer questions drawn from customer relationship data and Slack conversations and can plan and carry out multi-step tasks, including coordinating other agents. The company also introduced what it calls job-ready agents, pre-built for specific roles. One, named Piper, handles inbound sales leads around the clock. Another, named Fin, resolves service issues across many channels using its own models. A further sales agent called Hunter is scheduled for November.

The recap names partners including NVIDIA (NASDAQ:NVDA), Anthropic, Amazon Web Services, Microsoft Azure and Google, the last in connection with orchestrating third-party agents. It also references technology from companies Salesforce has acquired or partnered with, including Qualified, Fin and Regrello.

Taken as a whole, the announcements read as an effort to move from a single assistant to a family of agents with defined jobs, which is a more concrete pitch than the general pledges of earlier years.

How does Koa, the Salesforce reasoning model, fit into the Agentforce plan?

Koa is described by Salesforce as its first reasoning model built for customer relationship tasks inside Agentforce. The company said it was trained on synthetic data, built on NVIDIA's Nemotron family, and that Salesforce owns the model weights and hosts the model in its own infrastructure.

Those details matter to corporate customers who are cautious about where their data travels. Hosting a model within the vendor's own environment can simplify conversations about privacy and governance, although it does not remove the need for customers to test results. Koa is currently in pilot, with general availability expected in winter.

The model also reflects a strategic choice. Salesforce is offering access to outside models such as Claude while also developing its own, which gives it a way to tune cost and behavior for specific business tasks. The company did not publish side-by-side performance data in the recap.

An accompanying feature, AI Skills, lets teams write reusable governed instructions that any agent can run, either conversationally or through a builder. That feature is in pilot with general availability planned for October. It is a small item in the list, but it speaks to how the company expects customers to keep agents consistent across departments.

Why is Salesforce emphasizing testing and monitoring of Agentforce agents?

A notable share of the announcements concerned measurement rather than new agents. Salesforce introduced voice and conversation testing in beta that grades multi-turn exchanges, voice observability that flags audio quality problems, and custom scorers that let teams define their own evaluation criteria. An experimentation feature for comparing agent variants enters beta in October.

Agent health monitoring is generally available and tracks more than 16 built-in metrics along with custom ones. An Agent Optimizer, also due in October, is meant to build and refine other agents and to surface recurring failure patterns found in production sessions.

The emphasis on observability responds to a common concern among businesses trying agents: results can vary from one session to the next, and without measurement it is hard to say whether a change helped. By offering its own tools for grading and comparing agents, Salesforce is positioning reliability as part of the product rather than an afterthought.

There is also a competitive logic. Customers that rely on built-in testing may find it harder to move agents to a rival platform. Whether that benefit materializes depends on how deeply the tools are adopted, which the company has not yet quantified.

What did Salesforce report for its record second quarter of fiscal 2027?

In its release dated August 26, covering the quarter that ended July 31, Salesforce reported revenue of about $11.3 billion, up 11% both as reported and in constant currency. The total included a contribution of about $456 million from Informatica. Subscription and support revenue was about $10.8 billion, up 12%.

Current remaining performance obligation, a measure of contracted revenue due within a year, was about $33.5 billion, up 14%. Total remaining performance obligation was about $66.3 billion, up 11%.

On earnings, GAAP diluted earnings per share were $4.29, while non-GAAP diluted earnings per share were $5.90. GAAP operating margin was 20.5%, and non-GAAP operating margin was about 34%. Operating cash flow was about $1.3 billion and free cash flow about $1.1 billion.

Chair and chief executive Marc Benioff said the company had delivered one of its best quarters ever. Chief financial officer Robin Washington said artificial intelligence was amplifying the power, reach and value of the platform.

How large is the Agentforce business inside Salesforce's results?

Salesforce said Agentforce annual recurring revenue was above $1.5 billion, up more than 240% from a year earlier. It added that Agentforce and its data platform combined reached annual recurring revenue of nearly $3.9 billion, up more than 210%.

Context is useful here. Even at the higher figure, the combined business is a modest slice of a company with quarterly revenue of more than $11 billion, so the near-term path of total growth still depends largely on the established clouds for sales, service, marketing and commerce, along with the integration platform and Slack.

The pace of expansion in the agent line is what draws attention. When a product more than triples in a year, small changes in customer adoption can shift the company's overall growth rate within a few quarters. The reverse is also true: if pilots do not convert to paid deployments, the trajectory could flatten.

Annual recurring revenue is a company-defined metric, and Salesforce does not break out how much of the figure comes from new customers versus expansions of existing contracts in the release. That gap makes it hard to judge from the public numbers how broad the adoption is.

What guidance did Salesforce give for the third quarter and full fiscal year?

For the third quarter, Salesforce guided to revenue of about $11.42 billion to $11.5 billion, growth of 11% to 12%, with current remaining performance obligation growth of about 14%. That quarterly guidance excludes the pending Contentful and Fin acquisitions.

For the full fiscal year, the company raised its revenue range by $200 million to between about $46.1 billion and $46.4 billion, growth of 11% to 12%. That raise includes the two pending acquisitions, which are conditional on closing, so the organic portion of the increase is not separately stated in the release.

Full-year GAAP operating margin guidance is about 20%, and non-GAAP operating margin guidance is about 34%. GAAP diluted earnings per share are expected between about $10.21 and $10.25, with non-GAAP between about $16.67 and $16.71. Operating and free cash flow are expected to grow about 4% to 5%.

Cash flow growth well below revenue growth is an item to watch.

How is Salesforce distributing cash through its $25 billion accelerated program?

Salesforce said it paid about $364 million in dividends during the quarter. It also said a $25 billion accelerated share retirement program is expected to settle finally in October.

An accelerated program of this kind typically involves the company paying a bank up front and receiving shares over time, with a final settlement that adjusts the number of shares delivered based on the average price during the period. The final count will therefore only be known after settlement, and the effect on per-share earnings is reflected gradually.

The size of the program relative to the company's cash generation is a talking point. Free cash flow in the quarter was about $1.1 billion, a fraction of the program's total, so the program is large relative to a single quarter of cash generation; the company's filings describe how it is funded.

Pending acquisitions add another demand on capital. Balancing share retirement, dividends, deals and ongoing spending on artificial intelligence capacity is a challenge that the company will likely discuss at its next earnings call.

What are the open questions around Salesforce's agent rollout?

The first is timing. Several of the most ambitious items, including long-horizon agents that pursue goals over days or months with human approval checkpoints, are in pilot, with general availability slated for November. Koa is expected in winter, and voice for employee agents is in pilot on the desktop product.

The second is conversion. The company has not disclosed how many customers use agents in production, how much usage is paid versus trial, or how pricing will evolve as agents take on more work. Those answers will shape whether the strong annual recurring revenue growth continues.

The third is competition. Large software and cloud vendors, as well as model developers, are all building agent platforms. Salesforce's argument is that its data, governance tools and distribution through Slack give it an advantage, but rivals can offer similar claims.

The fourth is integration. With acquisitions such as Informatica already contributing revenue and Contentful and Fin pending, the company must combine products, teams and customers while launching new features at a fast pace. Execution on all of those fronts at once is a demanding task.

What should readers watch next for Salesforce?

Near-term dates include the October general availability of Slackforce, Agent Optimizer and AI Skills, the October settlement of the accelerated share retirement program, and the November arrival of the Hunter agent and long-horizon agents. The next quarterly earnings release, not yet dated, is the next scheduled occasion to see whether the agent business keeps growing at its recent rate.

The pending closings of Contentful and Fin are also worth tracking because they change the guidance baseline. Management commentary on pricing, customer counts and the contribution of the data platform would help clarify how much of the growth is broad-based.