Quick answer
On 26 August 2025, Gartner predicted that 40% of enterprise applications will be integrated with task-specific AI agents by the end of 2026, up from less than 5% in 2025. In sales tools, the label has arrived: Salesforce, HubSpot, Outreach, Salesloft, Apollo, Gong and LinkedIn all sell "agents". On their own pages, checked on 15 October 2026, most of these agents research, draft, summarise and update CRM fields. Only Salesforce, HubSpot and Outreach describe agents that can email prospects without a person approving each message, and all three let you keep approval on. Apollo says its agents don't run unattended yet, and LinkedIn's Sales Assistant is still in beta. Judge each one by what it may do without a human, not by its name.
What Gartner predicted, and who it was for
Gartner's press release of 26 August 2025 says 40% of enterprise applications will be "integrated with task-specific AI agents by the end of 2026", up from less than 5% in 2025. Its best case has agentic AI driving about 30% of enterprise application software revenue by 2035, over $450 billion.
The 40% is stage two of a five-stage path:
- 2025: most enterprise apps get embedded AI assistants.
- 2026: up to 40% of apps include task-specific agents.
- 2027: a third of agentic AI implementations combine agents with different skills inside one app.
- 2028: networks of agents work across apps, and a third of user experiences move from native apps to agent front ends.
- 2029: at least half of knowledge workers learn to work with, govern or create agents for complex tasks.
Two things are easy to miss. The release is written for C-level leaders at software companies, who it says have a three to six month window to set their agentic product strategy. So it's a forecast about vendors shipping features, which explains why every sales tool you pay for now has an "agents" page. And Gartner's own numbers don't line up neatly: its June 2025 release forecast 33% of enterprise software applications including agentic AI by 2028. Read the 40% as a direction, not a count anyone will publish.
I set up cold email and LinkedIn outbound inside B2B teams, with 2000+ meetings booked for B2B clients, so my question is narrower than Gartner's: what do the agents in the tools a sales team already pays for actually do?
Assistant or agent: Gartner's own test
The same release draws the line. AI assistants simplify tasks but depend on human input and don't act on their own. Task-specific agents can carry out complex tasks end to end. Calling an assistant an agent is what Gartner calls "agentwashing", and it names it as the most common misconception.
In May 2026 Gartner added four autonomy levels, each with its own controls. They're a useful way to sort sales tool agents, because they describe what the agent may do, not what it's called. Source: Gartner, 26 May 2026.
| Gartner level | What the agent may do | Sales example | Controls Gartner says it needs |
|---|---|---|---|
| 1. Observe | Read data, show output only to the person who asked | Account summary, call brief | Scoped data access, sign-in, usage logs |
| 2. Advise | Draft and recommend; a person acts | Email drafts, suggested next steps | Accuracy testing and training people not to over-trust it |
| 3. Act with approval | Write data or send, after a person approves each action | Agent sends once a rep approves the email | Approval workflow with an audit trail |
| 4. Act autonomously | Act within guardrails; people review exceptions and results | Agent emails, follows up and replies on its own | Monitoring, rollback, circuit breakers and a named owner |
Gartner's warning in that release is that 40% of enterprises will demote or decommission autonomous agents by 2027 because of governance gaps found only after an incident. The more an agent may do alone, the more those gaps cost.
What the agents in 7 sales tools do today
I read each vendor's own product and help pages on 15 October 2026. This is what the vendors say their agents do, not something I measured.
| Tool | Agent | What it does | Emails prospects without approval? | Pricing |
|---|---|---|---|---|
| Salesforce | Agentforce Engagement (renamed from Agentforce Lead Nurturing) | First emails from CRM records, a follow-up nudge, answers from material you upload, a meeting link when someone is interested, opt-out handling | Yes, unless you switch on manual approval. It can send as the record owner. | Agentforce pricing (Flex Credits or add-ons), confirm with Salesforce; needs Enterprise, Performance or Unlimited with Sales and Einstein for Sales |
| HubSpot | Prospecting Agent | Watches buying signals, scores accounts against your ICP, sources contacts from data providers such as ZoomInfo and Apollo, drafts outreach in each rep's voice | Your choice: review every draft or switch to fully autonomous mode | $1 per lead it recommends outreach for, paid in HubSpot Credits; Starter, Professional and Enterprise |
| Outreach | Revenue, Research, Personalization, Deal and Meeting Prep agents, plus Omni | Finds high-intent accounts, sources contacts, writes email, call and LinkedIn content, suggests deal field updates, builds meeting briefs | Plays can run fully autonomous, sending as the seller. Admins set autonomy by segment. A Reply Agent is listed as coming soon. | Check current pricing |
| Salesloft | Agents for research, stakeholders, priorities, email, deals and coaching | Researches accounts and buyers, suggests stakeholders, orders daily tasks, writes cadence and play emails, summarises deals, flags coaching needs | The agents page describes no agent that runs outreach end to end | Check current pricing |
| Apollo | Agents | Research accounts, build lists, draft outreach, surface signals | No. Apollo's FAQ says reps approve before anything goes out and full autonomy is on its roadmap. | Check current pricing |
| Gong | AI Data Extractor, AI Briefer, AI Theme Spotter, AI Trainer, custom agents | Fills CRM fields from calls and emails, writes briefs, spots recurring themes, runs role-play practice | Doesn't contact prospects. Data Extractor writes to your CRM with no rep input. | Check current pricing |
| LinkedIn Sales Navigator | Sales Assistant | Finds leads, explains why they fit, suggests a warm path in, drafts intro requests and outreach | No, it drafts. LinkedIn says it's in beta with a select group of customers. | Part of Sales Navigator, not generally available |
Three things stand out.
- Few agents send on their own, and none has to. Salesforce, HubSpot and Outreach can run at Gartner's level 4 on prospect email. All three offer a human approval step, and Salesforce's is a setting you have to switch on.
- Pricing is moving to units of work. HubSpot charges per lead recommended. Salesforce's Agentforce pricing page meters actions in Flex Credits at $500 per 100,000, with a standard action using 20. That's easy to start and harder to forecast.
- The FAQ tells you more than the banner. Apollo's sales-leader page lists agents that run prospecting end to end, while its FAQ on the same page says they don't work unattended yet. Read the help pages before the hero section.
For what's actually live in Sales Navigator, see LinkedIn's AI Sales Assistant is still in beta. For how much of Outreach's agent story is new product and how much is a new label, see what to check before you buy the agentic label.
How to judge an agent inside a tool you already pay for
Before you switch one on, get answers to six questions. Most are in the agent's setup screen or help page.
- What level does it run at by default? Observe, advise, act with approval or act on its own. Then check who in your team can change it.
- What can it write to? Prospects' inboxes, CRM fields and reps' calendars carry different risks and need different owners.
- Whose mailbox does it send from? If it sends as your reps from your main domain, cold volume puts that domain's reputation at risk. That's why cold email usually runs on separate domains. Great copy means nothing if it lands in spam.
- What's the billing unit, and can you cap it? Per lead, per action or per credit. Set a usage alert before the first run.
- Can you see every action and why it was taken? You want a per-record log of what the agent sent or changed, not only a weekly total.
- How do you stop it? A pause switch, a way to undo CRM changes and thresholds that pause it automatically, which Gartner calls circuit breakers.
A ten-minute check. Open the agent's setup page in your tool and find three settings: the approval switch, the sending mailbox and the usage limit. If you can't find one of them, ask the vendor before you turn the agent on.
Built-in agent or a separate AI SDR
A built-in agent has one big advantage: it already sits on your CRM data and your reps' history. A separate AI SDR has its own: it can send from its own domains under its own limits. Pick by the job.
Pick the built-in agent if
- The job lives inside that tool: research, enrichment, drafts for reps, CRM updates or follow-up on inbound leads.
- Your CRM data is clean enough that you'd trust a new hire to work from it.
- You want one vendor and one bill, and the pricing unit makes sense for your volume.
Pick a separate AI SDR or your own sequencer if
- You send cold at volume and want it off your main domain with its own sending limits.
- You need email and LinkedIn in one sequence the built-in agent doesn't cover.
- The built-in agent sits at level 2 and the job needs more.
Either way, treat switching it on as a pilot: one segment, a control group, an end date and kill criteria agreed up front. I wrote that plan out in how to run an AI SDR pilot that avoids Gartner's 40% cancellation rate.
Why more agents won't mean more pipeline
Gartner's sales practice made the same point from the seller's side on 28 July 2026: by 2028, AI agents will outnumber sellers 10 to 1, yet fewer than 40% of sellers will say agents improved their productivity. Its analyst Dan Gottlieb put the reason simply: agents only work as well as the systems they run in, and fragmented systems get scaled fragmentation. I went through that forecast in what Gartner's AI predictions mean for outbound.
For a sales team that means the useful question isn't whether your tools are in Gartner's 40%. They probably are, at least in name. It's which single job you'd hand an agent, at which level, with whose mailbox and who's watching.
Key takeaways
- Gartner predicted on 26 August 2025 that 40% of enterprise apps will be integrated with task-specific AI agents by the end of 2026, up from less than 5% in 2025. It's a forecast about vendors shipping features.
- Gartner's own line: assistants depend on human input, agents carry out tasks end to end. Calling an assistant an agent is agentwashing.
- Of seven sales tools checked in October 2026, only Salesforce, HubSpot and Outreach describe agents that can email prospects without approval, and all three let you keep approval on.
- Judge any built-in agent by its default autonomy level, what it writes to, whose mailbox it sends from, its billing unit, its logs and its off switch.
- Switching on a built-in agent is still a pilot: one segment, a control group, an end date and kill criteria.
FAQ
What did Gartner predict about task-specific AI agents in enterprise apps?
In a press release dated 26 August 2025, Gartner predicted that 40% of enterprise applications will be integrated with task-specific AI agents by the end of 2026, up from less than 5% in 2025. It's stage two of five, leading to agent ecosystems across apps by 2028.
What is the difference between an AI assistant and an AI agent?
By Gartner's definition, an AI assistant depends on human input and doesn't act on its own, while a task-specific agent can carry out a complex task end to end. Gartner calls the habit of labelling assistants as agents "agentwashing".
Which sales tools have AI agents built in?
As of October 2026, Salesforce (Agentforce Engagement), HubSpot (Prospecting Agent), Outreach, Salesloft, Apollo, Gong and LinkedIn Sales Navigator (Sales Assistant, in beta) all describe agents on their own sites. Only Salesforce, HubSpot and Outreach describe agents that can email prospects without a person approving each message.
Can HubSpot's Prospecting Agent send emails automatically?
Yes, if you allow it. HubSpot says you can review and edit drafts each time you enroll contacts, then switch to a fully autonomous mode that sends without review. It costs $1 per lead the agent recommends outreach for, paid in HubSpot Credits.
Is LinkedIn's Sales Assistant available to all Sales Navigator users?
No. LinkedIn's own page says Sales Assistant is in beta with a select group of customers (checked October 2026). It finds leads, explains why they fit and drafts intro requests and outreach.
