Salesforce Bets Big on “Agentic Enterprise” as It Arms Itself with a $50bn Buyback

A new software cycle, and a new balance sheet play

From the 60th floor of Salesforce Tower, Marc Benioff cast the latest swoon in software valuations as just another “SaaS pocalypse” – and a buying opportunity. The numbers he and new operating and finance chief Robin Washington laid out for fiscal 2026 suggest a company using a cyclical downturn in sentiment to re‑gear both its technology stack and its balance sheet.

On the headline figures, Salesforce delivered one of its cleanest years in recent memory. Revenue for the year rose 10% to $41.5bn (9% in constant currency), with the fourth quarter accelerating to 12% growth. Subscription and support revenue grew “slightly above” 10% in both nominal and constant‑currency terms. Current remaining performance obligations – the nearest proxy for contracted, near‑term revenue – climbed 16% to $35.1bn, aided by a four‑point contribution from Informatica. Total RPO, a longer‑dated measure of backlog, breached $72bn, up 14%.

Benioff, who has never been shy about grandiose framing, called it “one of the best performances in software ever.” What gave his rhetoric more bite was the way Washington paired it with a reshaped capital allocation agenda. After returning more than $14bn to shareholders in FY26 – 99% of free cash flow – Salesforce is now layering leverage onto what has been, until now, a notably conservative balance sheet. The board has boosted the quarterly dividend by 5.8% to $0.44 a share and, more dramatically, lifted the share repurchase authorization to $50bn.

The subtext is simple: at roughly $16bn‑plus of expected free cash flow in FY27, management believes buying Salesforce stock at today’s multiples is as compelling as any large‑scale acquisition. Washington stressed that this does not preclude further M&A; it just imposes a tighter formula on deals and acknowledges that earlier cycles – Slack and Tableau in particular – were expensive in share count. This time, debt will be part of the toolkit to retire equity.

For investors used to fretting about the durability of double‑digit growth, the move is a declaration of confidence that the next leg of expansion will be driven not by another megadeal, but by the company’s own reinvention around AI agents.

The “agentic” pivot: from tokens to work

If the balance sheet message was blunt, the product message was more architectural. Benioff spent much of his shortened presentation sketching a four‑layer “operating system for the agentic enterprise”: commodity large language models at the base; a harmonisation and federation layer branded Data 360 (now buttressed by Informatica); the familiar suite of Salesforce apps above that; and a new Agentforce layer that orchestrates AI agents, increasingly surfaced through Slack.

His claim is that a new market has opened up alongside traditional SaaS: agents as a service, with “humans and agents working together” across sales, service, marketing and internal productivity. The numbers, at least in their first year, are striking. Agentforce and Data 360 ARR now total $2.9bn, including $1.1bn of Informatica Cloud ARR and roughly $800m from Agentforce itself – up 169% year on year. New bookings on Salesforce’s most premium AI bundles – Agentforce One Edition and Agentforce for Apps (A4X) – nearly tripled quarter‑on‑quarter. More than 60% of Agentforce and Data 360 bookings in the quarter came from existing customers expanding their commitments.

The company has also started to quantify AI usage in an unusually granular way. It says it has now consumed over 19 trillion tokens from model providers such as OpenAI, Anthropic, Mistral and Meta’s Llama – a proxy for the raw intelligence flowing through its stack. But Benioff and product chief Patrick Stokes clearly see that metric as too abstract for investors trying to link AI to revenue.

Enter a new disclosure: Agentic Work Units, or AWUs. Defined as discreet “units of AI work” – a record update, a triggered workflow, a decision, a call to Salesforce’s “multi‑component” orchestration service – AWUs are meant to capture the conversion of tokens into enterprise‑grade transactions. To date, agents on the Salesforce platform have generated 2.4bn of these, including 771m in the fourth quarter alone.

Investors should not yet read AWUs as a direct revenue driver. Washington was careful to say that in the near term, AI is broadly neutral for gross margins: token prices are expected to fall as models commoditise, while Salesforce engineers are working on ways to make agents more deterministic and less compute‑hungry. But the ratio of tokens to AWUs is already becoming an internal management tool. Customers with high AWUs per token are deemed to be extracting more value; those with low ratios become targets for customer success teams to tweak implementations.

The bigger monetisation story, as Chief Revenue Officer Miguel Milano outlined, is more prosaic. Salesforce says it is now extracting AI value in three overlapping ways: upgrading its vast installed base on to premium SKUs with embedded Agentforce entitlements; rolling out core apps to new seats that are suddenly justifiable because AI‑boosted ROI is higher; and selling “fuel” in the form of Flex Credits for customer‑facing agents. In the fourth quarter, roughly half of Agentforce bookings came from these credits and half from higher‑tier SKUs.

The early traction is weighted heavily toward the company’s largest accounts: all of the top 10 deals in Q4 included Agentforce, data, sales, service, platform and analytics; Informatica landed in six of those ten, a swift validation of the acquisition. New wins over $1m grew 26% year on year; those over $10m were up 33%. Salesforce closed a 10‑year IDIQ contract with the U.S. Army, with a $5.6bn ceiling, and a string of notable competitive takeaways, including life sciences customers leaving Veeva and IT service management clients defecting from ServiceNow to Salesforce’s new ITSM product.

Perhaps more telling for the medium term, the pipeline for these comprehensive bundles is feeding back into Salesforce’s long‑term model. The company has raised its FY30 revenue target from $60bn to $63bn, implying an 11% CAGR from FY26, and reiterated its ambition to reach a Rule of 50 – where revenue growth plus non‑GAAP operating margin equals 50 – by the end of the decade. In FY26, it expanded non‑GAAP margins by 60 basis points; in FY27 it plans a more modest 20‑basis‑point improvement to 34.3% as it invests in infrastructure, quota‑carrying headcount and customer success staff to support Agentforce adoption.

Customers as proof points – and fault lines

In a departure from the usual analyst‑heavy format, Salesforce gave much of its presentation over to three customers – SharkNinja, Wyndham Hotels & Resorts and SaaStr – painting on‑the‑ground portraits of how agents are being deployed.

For SharkNinja, a fast‑moving consumer appliance maker launching 25 products a year, the headline claim was speed. In eight weeks, ahead of the holiday season, the company and Salesforce built a guided shopping agent to sit across research, buying and troubleshooting. In just one quarter, agents participated in 250,000 consumer engagements, taking routine queries away from human agents and helping to lower service costs while improving response times.

Wyndham’s picture was more sweeping, and more operational. Chief executive Geoff Ballotti said the chain now has more than 5,000 Agentforce deployments across its 8,300‑plus hotels, anchored by a “Guest 360” data foundation built with MuleSoft and Data 360. Agents have encyclopaedic knowledge of guest history and preferences; voice agents are delivering a 200‑basis‑point uplift in direct booking conversion versus third‑party channels, shaving millions from labour costs in front offices and adding millions in incremental ancillary revenues. Guest satisfaction scores, he said, are 400 basis points higher than before.

The smallest of the three, SaaStr, was there to make a different point: that this is not just a large‑enterprise story. Founder Jason Lemkin described shrinking his go‑to‑market team from 15 humans to two‑and‑a‑half and “20 agents,” then rebuilding sales processes around Agentforce. In a business of modest scale, agents have already closed $2.7m of revenue with a further $3.5m in pipeline, Lemkin said – largely by re‑engaging leads that human reps never had time to pursue. A similar pattern, he added, is playing out at PayPal, which is using Agentforce to rescue abandoned merchant onboarding flows that were previously uneconomical to staff.

The through‑line is that Salesforce’s competitive advantage lies less in its own models than in its ability to bind models to structured data, workflow and an engagement surface. Slack, which Benioff noted now carries around 1bn messages a day versus 500m on X (Twitter), sits at the heart of that narrative. Slackbot – an AI assistant that can read Slack messages, Salesforce records and third‑party systems such as Google and Microsoft Teams – is fast becoming the company’s showcase. Benioff recounted using it to brief Workday’s chief executive in realtime on every deal and integration between the two firms.

The architecture, Stokes argued, turns Slack into the “system of engagement” for an agentic enterprise, sitting above a “system of context” (Data 360), “system of work” (the core CRM and service clouds) and “system of agency” (Agentforce itself). Anthropic, in its own developer event this week, underlined that view by featuring Slack at the start and end of every enterprise demo – a collaboration that Salesforce has sweetened with roughly $330m of equity investment, almost 1% of Anthropic’s cap table.

Yet this emergent stack also highlights where Salesforce is still on the back foot. Washington acknowledged that Marketing and Commerce Cloud remain under pressure and that Tableau continues to lag expectations, with on‑premises revenue timing a persistent drag. Those pockets of weakness – and the modest beat on organic CRPO growth, which came in around 9%, merely in line with guidance – are fuelling a recurring investor concern: can Salesforce drive an Agentforce‑led boom without watching older parts of its portfolio stagnate?

Benioff’s answer is to emphasise the company’s subscription heritage as a buffer. The installed base, with attrition steady at roughly 8%, gives him forward visibility and a platform for upsell. Milano’s sales organisation, he argues, is now properly tooled up: ramped account executive capacity is up 15–17% year on year; AI enterprise licence agreements (AELAs) are becoming the default SKU in large accounts, with more than 120 sold in Q4 alone and eight embedded in the top ten deals.

The sales motion, in other words, is shifting from discrete AI pilots to multi‑year, multi‑cloud commitments where Agentforce is less a line item and more an overlay. That is why Salesforce is now reconsidering how it breaks out revenue by “cloud” from FY27 onward; by the next reporting cycle, the product lines may be reframed explicitly around this agentic architecture.

For now, the macro story and the micro story are pulling in opposite directions. The market is treating software broadly – and Salesforce stock specifically – as if the boom times are over. Benioff is betting that the combination of rising AI‑driven backlog, a still‑disciplined margin trajectory and an unusually muscular buyback will convince investors that what looked like late‑cycle SaaS may actually be the early innings of a new software layer.