From Aryan Vatsa | Product & Market Analysis
Gartner's 40% AI Agent Forecast Tested on 50 Enterprise Apps: 26% to 96%
On this page
In August 2025 Gartner forecast that 40% of enterprise applications would carry task-specific AI agents by 2026, up from less than 5%. With 50 days of the year left, we scored 50 widely used enterprise apps against their vendors' own pages. The Gartner 40 percent enterprise apps AI agents test comes out at 56% if the agent must act. It falls to 26% if the vendor must also say generally available, and rises to 96% if a label is enough.
Key takeaways
- The answer depends almost entirely on the definition. Of 50 leading enterprise apps, 48 market an embedded agent, 38 ship one to paying customers, 28 ship one that takes actions and 13 ship one their vendor calls generally available.
- At the top of the market, Gartner's 40% is already cleared. Under Gartner's own definition, an agent that acts rather than answers, 28 of the 50 apps qualified on 11 November 2026.
- Service and CRM lead, while finance and security lag. All 5 CRM apps and 5 of 6 service desks ship an acting agent, against 3 of 8 finance apps and 1 of 4 security platforms.
- Gartner published two incompatible versions of this forecast nine weeks apart. In June 2025 it predicted 33% of enterprise software apps would include agentic AI by 2028. In August it said 40% by 2026.
Did Gartner's 40% prediction come true? For the 50 most widely used enterprise apps, yes, under Gartner's own definition. As of 11 November 2026, 56% ship a task-specific agent that takes actions. Only 26% ship one the vendor explicitly calls generally available. Across all enterprise software, including the long tail, nobody has published a count.
What Gartner forecast about AI agents in enterprise apps
The forecast is more specific than the headline suggests. Gartner's release of 26 August 2025 said that "forty percent of enterprise applications will be integrated with task-specific AI agents by 2026, up from less than 5% today." It did not say "up to". It was a point forecast with a date attached, which is what makes it testable.
The figure is stage two of a five-stage path. Stage one was AI assistants embedded in almost every enterprise application by the end of 2025. Later stages run to collaborative agents inside one application by 2027 and agent ecosystems across applications by 2028. Gartner also projected agentic AI at about 30% of enterprise application software revenue by 2035, above $450 billion. That claim is a decade out. The 40% figure expires in seven weeks.
Assistants are not agents, by Gartner's own definition
The most useful sentence in the release is a warning. Gartner wrote that AI assistants "depend on human input and do not operate independently", and that "the most common misconception is referring to these AI assistants as agents." It blamed the confusion on agentwashing.
That gives us a definition to test against. A chat panel that answers questions about a record is an assistant. A component that takes a task, works through steps and changes something in the system is an agent. We think that line is the right one, and we hold vendors to it below. Gartner's other famous chart is covered in three years of Hype Cycle agent placements.
How we built the 50-app sample and the scoring rubric
Gartner never published the applications behind its "less than 5%" baseline. So any test has to define its own denominator, and that choice decides much of the answer. We took widely deployed systems of record and systems of work in 8 categories: productivity, content, CRM and marketing, service and ITSM, finance and procurement, HR, developer and data, and security.
That gives 50 apps from 39 vendors. Some vendors appear twice because they sell separate products to separate buyers, such as Sales Cloud and Service Cloud, or Jira and Confluence. Each app was scored on what its own vendor documented, on pages opened between 9 and 11 November 2026.
Four definitions, from loosest to strictest
Each app gets four yes or no scores. Every tier is a subset of the one above it, so counts can only fall as the definition tightens.
| Tier | Test | What it lets through |
|---|---|---|
| T1 Labelled | The vendor markets an embedded feature as an agent, at any status | Announcements, betas and renamed chat assistants. |
| T2 Shipping | T1, and paying customers can use it now | Production features that may only answer or draft. |
| T3 Acts | T2, and the vendor documents it creating or updating records, running workflows or opening pull requests | Gartner's own definition. Approval gates on high-impact steps are allowed. |
| T4 Acts, GA stated | T3, and the vendor's page says generally available | Only agents with an explicit production commitment. |
What counted as evidence
Only primary vendor material counted: press releases, documentation, release notes and newsrooms. Trade press could point us to a page but never scored an app alone. Where a vendor site blocked access and no vendor-issued copy existed, the app scored down rather than up. The test covers agents inside the application, not platforms that govern other agents, which is why Okta scores zero on every tier.
The result: 26% to 96% of the 50 apps carry AI agents
The same 50 apps, scored four ways, produce a spread of 70 percentage points. At the loosest reading, 48 apps carry an agent. At the strictest, 13 do. Gartner's 40% line falls between tiers 3 and 4.
Our verdict is that the forecast has been met at the top of the market, on the definition Gartner itself gave. We would not stretch that verdict further. The 26% figure matters as much, because it shows how many acting agents still ship under phrases like "available now" or "rolling out" rather than a plain production commitment.
The scored list: AI agents in 50 enterprise applications
Status and behaviour paraphrase the linked vendor page, dated by publication.
| App | Agent | Status | What it does | T1 | T2 | T3 | T4 | Source |
|---|---|---|---|---|---|---|---|---|
| Microsoft 365 | Agent Mode | GA | Multi-step edits in documents. | Y | Y | Y | Y | Microsoft, Apr 2026. |
| Google Workspace | Workspace Studio | GA | Runs flows across Gmail and Drive. | Y | Y | Y | Y | Google, Dec 2025. |
| Slack | Slackbot | GA | Answers, drafts, schedules. | Y | Y | N | N | Salesforce, Jan 2026. |
| Zoom Workplace | AI Companion workflows | Beta | Follow-up tasks. | Y | N | N | N | Zoom, Dec 2025. |
| Notion | Notion Agent | Launched | Edits many pages at once. | Y | Y | Y | N | Notion, Sep 2025. |
| Asana | AI Teammates | All paid customers | Writes across connected apps. | Y | Y | Y | N | Asana, Oct 2026. |
| monday.com | monday agents | Builder in beta | Drafts campaigns, closes tickets. | Y | N | N | N | monday.com, May 2026. |
| Smartsheet | Agentic Workflows | Available today | Runs workflows alone. | Y | Y | Y | N | Smartsheet, Nov 2026. |
| Miro | Sidekicks | No status stated | Builds boards. | Y | N | N | N | Miro, Oct 2026. |
| Confluence | Rovo agents | On paid plans | Creates and edits pages. | Y | Y | Y | N | Atlassian, Aug 2026. |
| Box | Box Agent | Top tiers | Write-back not live. | Y | Y | N | N | Box, 2026. |
| Dropbox | AI actions | 70,000+ teams | Executes approved plans. | Y | Y | Y | N | Dropbox, Nov 2026. |
| DocuSign IAM | Iris agents | GA, North America | Requests approvals, triggers steps. | Y | Y | Y | Y | DocuSign, Aug 2026. |
| Adobe Acrobat | Productivity agent | Available | Generates content only. | Y | Y | N | N | Adobe, May 2026. |
| Sales Cloud | Agentforce SDR | GA | Engages prospects alone. | Y | Y | Y | Y | Salesforce, Aug 2024. |
| HubSpot | Breeze Customer Agent | Available globally | Resolves tickets. | Y | Y | Y | N | HubSpot, Apr 2025. |
| Dynamics 365 Sales | Sales Qualification Agent | GA | Researches leads, starts outreach. | Y | Y | Y | Y | Microsoft, Dec 2025. |
| Zoho CRM | Zia Agents | Paid plans | Generates quotes. | Y | Y | Y | N | Zoho, undated. |
| Adobe Experience Platform | Agent Orchestrator | GA | Creates customer journeys. | Y | Y | Y | Y | Adobe, Sep 2025. |
| Service Cloud | Agentforce Help Agent | GA | Manages cases, updates orders. | Y | Y | Y | Y | Salesforce, Jun 2026. |
| ServiceNow ITSM | L1 Service Desk Specialist | Available now | Resolves cases end to end. | Y | Y | Y | N | ServiceNow, May 2026. |
| Zendesk | AI agents | GA | Acts in backend systems. | Y | Y | Y | Y | Zendesk, May 2026. |
| Jira Service Management | Rovo | Included, Rovo Service unconfirmed | Action not verified. | Y | Y | N | N | Atlassian, Oct 2025. |
| Freshdesk | Freddy AI Agent | Sold, no GA date | Processes refunds. | Y | Y | Y | N | Freshworks, undated. |
| Intercom | Fin | Paid product | Updates external systems. | Y | Y | Y | N | Intercom, undated. |
| SAP S/4HANA | Joule finance agents | Phased GA | Transfers funds. | Y | Y | Y | Y | SAP, Oct 2026. |
| Oracle Fusion ERP | Payables Agent | Shipped, 26A to 26D | Creates invoices. | Y | Y | Y | N | Oracle docs, undated. |
| Workday Financials | Accounting Agent | Early adopters, 2027 | Prepares work for approval. | Y | N | N | N | Workday, Oct 2026. |
| NetSuite | SuiteAgents | In NetSuite Next | Write access not stated. | Y | Y | N | N | NetSuite, Oct 2026. |
| Dynamics 365 Finance | Reconciliation agent | Preview | Suggests matches. | Y | N | N | N | Microsoft, Sep 2026. |
| QuickBooks Online | Payments Agent | Rolled out, US | Creates and sends invoices. | Y | Y | Y | N | Intuit, Jul 2025. |
| Coupa | Navi agents | GA | Execution not stated. | Y | Y | N | N | Coupa, Jun 2026. |
| SAP Concur | Expense Validation Agent | GA | Guides, does not act. | Y | Y | N | N | SAP, Apr 2026. |
| Workday HCM | Talent Acquisition Agent | Early adopters | Outreach needs approval. | Y | N | N | N | Workday, Oct 2026. |
| SAP SuccessFactors | Joule HR agents | Phased GA | Creates and edits positions. | Y | Y | Y | Y | SAP, Oct 2026. |
| Oracle Fusion HCM | HR agents | Shipped, 26B | Edits employee records. | Y | Y | Y | N | Oracle docs, undated. |
| ADP Workforce Now | ADP Assist | Chat assistant | Guided answers. | N | N | N | N | ADP, undated. |
| UKG Pro | Bryte Voice | Unverified | Voice actions claimed. | Y | N | N | N | UKG release copy, Aug 2026. |
| Dayforce | Dayforce AI Agents | Select agents live | Takes action in HCM. | Y | Y | Y | N | Dayforce, Oct 2026. |
| GitHub | Copilot coding agent | GA | Opens pull requests. | Y | Y | Y | Y | GitHub, Sep 2025. |
| GitLab | Duo Agent Platform | GA | Creates merge requests. | Y | Y | Y | Y | GitLab, Jan 2026. |
| Jira | Rovo agents | GA | Assigns issues, updates code. | Y | Y | Y | Y | Atlassian, May 2026. |
| Snowflake | Cortex Agents | GA | Answers, custom tools only. | Y | Y | N | N | Snowflake, Nov 2025. |
| Databricks | Supervisor Agent | GA | Calls tools, reads data. | Y | Y | N | N | Databricks, Feb 2026. |
| Datadog | Bits AI SRE | GA | Investigates, does not fix. | Y | Y | N | N | Datadog, Dec 2025. |
| Figma | Figma agent | Open beta | Edits designs. | Y | N | N | N | Figma, Jun 2026. |
| CrowdStrike Falcon | Charlotte AI agents | GA unconfirmed | Builds workflows. | Y | N | N | N | CrowdStrike, Sep 2025. |
| Cortex XSIAM | Cortex AgentiX | Available today | Contains hosts, with approval. | Y | Y | Y | N | Palo Alto Networks, Oct 2025. |
| Okta | Okta for AI Agents | Governs others | No embedded agent. | N | N | N | N | Okta, Sep 2026. |
| Splunk ES | Agentic SOC | No status label | Drafts playbooks. | Y | N | N | N | Splunk, Sep 2026. |
Totals: T1 48, T2 38, T3 28, T4 13. "Available now" and "rolling out" count as shipping. Phased general availability counts as GA stated. A page that would not load, or stated no status, scored the app down.
Where enterprise app AI agents are real and where they are a label
The aggregate hides a sharp split by category. Acting agents are close to universal where the outcome is a closed ticket or a qualified lead. They are scarce where the outcome is a ledger entry, a payroll record or a contained host.
Service and CRM: agents ship where the outcome is countable
All 5 CRM and marketing apps and 5 of 6 service desks ship an agent that acts. A resolved conversation or a qualified lead is a unit a vendor can bill, and pricing shows it. HubSpot moved its Customer Agent to $0.50 per resolved conversation from 14 April 2026. Intercom prices Fin at $0.99 per outcome. Those meters are compared in the breakdown of outcome-based AI pricing meters.
When a vendor can bill per completed task, it has every reason to ship an agent that completes tasks. We think that explains the CRM and service numbers better than any difference in technical difficulty.
Finance and HR: agents wait for a human signature
Finance is where the forecast meets resistance. Only 3 of 8 finance and procurement apps ship an agent that acts. Workday's Accounting Agent reaches early adopters in early 2027, and Microsoft still labels its Dynamics 365 Finance reconciliation agent a preview. Coupa says more than 20 Navi agents are generally available, but its release does not say they execute anything alone.
HR is similar. Workday's Talent Acquisition Agent is with early adopters, and candidate outreach still needs recruiter approval. ADP's Workforce Now page describes a chat assistant. We would not read this as a lag that closes in a quarter. Ledger postings and pay changes are exactly the actions auditors want a named human to own.
Security: vendors keep a human on the trigger
Security scores lowest, at 1 of 4. Palo Alto Networks' AgentiX can contain a host, but its agents "require human-in-the-loop approval for impactful actions". CrowdStrike's agents run "always under defender control". The category that knows most about attack surfaces is the slowest to let software act alone, and buyers should take the hint. The identity side is covered in the piece on agent identity and non-human access.
Gartner AI prediction accuracy: the definition does the work
A forecast that passes at 56% and fails at 26% on the same data is less a forecast about software than about vocabulary. That matters, because the 40% figure has been repeated in vendor decks for 14 months as though it described something countable.
Two Gartner forecasts, nine weeks apart
In late June 2025, Gartner said 33% of enterprise software applications would incorporate agentic AI features by 2028, up from less than 1%. On 26 August 2025, it said 40% would carry task-specific agents by 2026, up from less than 5%. Both cannot describe the same thing. The second has a higher share arriving two years sooner, from a baseline five times larger.
The likeliest reconciliation is that the two releases counted different things, and Gartner did not explain the gap. Its June release also estimated that only about 130 of the thousands of vendors marketing agentic AI were genuine. Project-level fallout is tracked in the ledger of canceled agentic AI projects. Our position: treat any share-of-apps forecast as unverifiable until the analyst publishes the denominator.
Buyer data points the same way. In Gartner's survey of 360 IT application leaders, reported in October 2025, 75% were piloting or deploying agents "in some form". Only 15% were considering, piloting or deploying fully autonomous ones, and just 19% highly trusted their vendor's hallucination protection. How such surveys are built is examined in the review of agent adoption survey methodology.
Where this test is weakest
This post makes a claim about 50 apps and should be read as nothing more. Three weaknesses deserve plain statement.
The top 50 is not Gartner's denominator
Gartner's forecast covers enterprise applications in general, including thousands of mid-market and vertical products with far smaller engineering budgets. Our 50 come from the largest vendors in each category, the most likely to have shipped agents. A 56% result at the top is consistent with a figure well under 40% across the whole market. We cannot measure that tail, and we would distrust anyone who claims to without publishing the list.
Sample choice and vendor wording
The sample is editorial, and a different reasonable list could move the T3 count by several apps either way. Scoring rests on vendor wording, so an understated or blocked page scores low. UKG Pro, Miro, Splunk and CrowdStrike fall partly into that bucket. Several agents ship only in some regions or plan tiers, which flatters T2 and T3 for buyers outside them. And "acts" is a judgement: counting investigation as action would add Datadog and Databricks, taking T3 to 30 of 50, or 60%.
Frequently asked questions
What did Gartner predict about AI agents in enterprise apps?
In a press release dated 26 August 2025, Gartner predicted that 40% of enterprise applications would be integrated with task-specific AI agents by 2026, up from less than 5% at the time. It described this as stage two of five, after embedded AI assistants and before collaborative agents in 2027 and cross-application agent ecosystems in 2028. It also warned that calling AI assistants agents was the most common misconception.
Did Gartner's 40% AI agent prediction come true in 2026?
For the most widely used enterprise apps, yes. In a Zan Digital count of 50 leading apps on 11 November 2026, 56% shipped an embedded agent that takes actions, which is Gartner's own definition. Only 26% shipped one the vendor explicitly called generally available. Across all enterprise software, including smaller vendors, no published count exists, and the share is likely lower.
What is the difference between an AI assistant and an AI agent?
Gartner's definition is that AI assistants depend on human input and do not operate independently, while task-specific agents take a goal and complete it. In practice, an assistant answers questions or drafts text when prompted. An agent works through several steps and changes something, such as updating a record, sending an invoice or opening a pull request, often with approval required for high-impact actions.
Which enterprise apps have AI agents in 2026?
Among 50 leading apps scored on 11 November 2026, 13 have generally available agents that act. They are Microsoft 365, Google Workspace, Salesforce Sales Cloud and Service Cloud, Dynamics 365 Sales, Zendesk, SAP S/4HANA, SAP SuccessFactors, DocuSign, Adobe Experience Platform, GitHub, GitLab and Jira. Another 15 ship acting agents described as available now or rolling out, without the words generally available.
How accurate are Gartner AI predictions?
This one is only as accurate as its definition allows. Applied to 50 leading apps, the same data produces a share between 26% and 96% depending on what counts as an agent. Gartner also published a June 2025 forecast of 33% of enterprise software apps with agentic AI by 2028, which is hard to reconcile with August's 40% by 2026. Neither release named its sample of applications.
How can buyers tell whether a SaaS agent feature is real?
Ask four questions. Is the feature available to your plan and region today, or only announced? Does the vendor's documentation say it creates or changes records, or only answers and drafts? Does the vendor use the words generally available? And how is it priced? A per-resolution or per-action meter usually means the agent completes work, because a vendor cannot bill for outcomes it never delivers.
Where to start: score your own stack
Take your 10 largest software contracts and run them through the four-tier rubric above, using vendor documentation rather than the sales deck. It takes about an hour. You end up knowing which renewals buy a working agent and which buy a label.
Then, for every app that reaches T3, ask the vendor for last month's log of actions its agent took in your tenant. An agent that acts should leave a record you can read. If the vendor cannot produce one, price it as T2 at renewal. The AI agent procurement checklist has contract language to make that request stick.
References
- Gartner, Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026, 26 August 2025. Forecast, stages, revenue claim and agentwashing warning.
- Newswit, syndicated copy of the Gartner release, September 2025. Verbatim wording, as gartner.com refused automated access.
- IT Brief, Over 40% of agentic AI projects set for cancellation by 2027, 24 June 2025. The 33% by 2028 forecast and 130 vendors estimate.
- IT Brief UK, Survey finds slow adoption of autonomous AI agents, 1 October 2025. Gartner survey of 360 IT application leaders.
- HubSpot, Customer Agent and Prospecting Agent pricing, 13 April 2026. Per-resolution pricing.
- Intercom, Fin pricing, retrieved 11 November 2026. Per-outcome pricing.
- Vendor pages for all 50 apps, linked row by row in the scored table, retrieved 9 to 11 November 2026.
Weakest point of the source base: both Gartner releases were read through syndicated copies, and the 50 scores rest on vendor wording, which is marketing material even when primary. Re-score in February 2027.
Related reading