Agentforce review, October 2026

By Giorgio Bonan. Updated October 21, 2026, published October 21, 2026.

Agentforce is a capable agent platform with a hard cost model. It does real work when the CRM data is clean and the task can be finished end to end; it disappoints when it is bought to fix a messy pipeline or a forecast nobody trusts. This review sticks to figures with a source and a date, and says which ones come from Salesforce.

The verdict, by buyer

What Agentforce does well

• Acts inside Salesforce's rules. Agents run with the platform's permissions, sharing and audit trail, which keeps security reviews familiar.
• Covers many jobs on one platform. Service, sales, field service and custom agents share one builder, one billing pool and one monitoring view (the Digital Wallet).
• Has a broad model choice. The Atlas reasoning engine can use Claude, the default in Vibes and Agentforce Coworker, and Salesforce is piloting its own CRM model, Koa.
• Outcome-based pricing exists for support. For the Help Agent, Salesforce Ben reports that Salesforce bills only resolved inquiries, with a defined rule: at least two turns, no explicitly negative final feedback, and no escalation to a human.

Where it costs or stalls

• Cost predictability. Four models coexist: Flex Credits at $500 per 100,000 (20 credits per action), $2 per conversation, $125 per user per month, and editions such as Max from $550 per user per month. Agent calls through MCP and APIs will add a meter from November 2026, at rates not yet published. Our Agentforce pricing estimator shows how fast credits add up.
• Edition gate. "AI can be added to Core and above," states Salesforce's Sales Cloud pricing page. Teams on Starter Suite or Pro Suite start by upgrading the CRM.
• Data dependence. An agent that updates opportunities from call notes, or summarizes a pipeline, reproduces what is recorded. Stale stages and close dates become confident answers.
• Analyst caution. Gartner's note of 4 August 2026 advised heads of applications to be cautious before activating Agentforce Coworker, citing its effect on Salesforce spend and data access.

What Agentforce does for sales, specifically

Most Agentforce reviews are written from the service side. For sales teams, Salesforce's own material describes these capabilities:

The first line is the most useful for managers: suggestions arrive where the pipeline is reviewed. It still depends on reps having logged the notes and emails it reads.

Cost scenarios for a sales team

The heavy-use line shows why the add-on exists: past about 60 actions per seller per working day, credits cost more than the flat fee. The light-use line shows why credits make sense for most sales teams.

Who should wait

• Teams below Core. The upgrade comes first, and it is the larger cost.
• Teams with low logging discipline. Agents that suggest updates from activity have little to work with when activity is not recorded.
• Teams that need a price for the budget committee this quarter. Agent calls through MCP and APIs will be metered at rates Salesforce has not yet published.
• Teams that want one thing: an honest forecast. Pipeline Inspection, opportunity scoring and a weekly review routine come first, see Einstein vs Agentforce.

What changed in 2026

The numbers, and who reports them

The gap between the first and the fourth line is the honest summary of Agentforce in 2026: strong revenue growth, and a much smaller share of customers running agents in production.

Pros and cons at a glance

Security and data questions

Agentforce agents act with the permissions they are given, and Salesforce plans agent registration so that each agent gets its own identity with narrower permissions than a human user. Before going live, answer four questions:

1. Which records can the agent read? The same as the user it acts for, unless it has its own identity.
2. Which records can it change, and is there an approval step? Write actions without approval are the main source of surprises.
3. What leaves Salesforce? With Claude or another external model in the loop, check the trust and retention terms that apply to prompts.
4. Who reads the logs? Agent actions are logged; someone has to look at them.

An implementation timeline that works

Teams that start with "deploy Agentforce" rather than one use case tend to end the quarter with agents configured and nothing measured.

Agentforce for small sales teams

For a team of 5 to 20 sellers, three facts frame the decision. AI is added from the Core edition upward, so a team on Starter Suite or Pro Suite starts with a CRM upgrade. Core includes 500,000 Flex Credits a year, enough for a few actions per seller per day. And the $125 add-on costs $1,250 a month for 10 sellers, a large line for a small team. The practical route is to use what Core includes, pilot one agent on credits, and buy the add-on only for sellers who use it heavily.

What reviewers on both sides agree on

Whether they come from Salesforce partners, analysts or competing vendors, assessments of Agentforce in 2026 tend to agree on three points. The platform is credible and moving fast, with new models and surfaces every quarter. The pricing is complex, and buyers struggle to predict a monthly bill. And results depend on the data and the process around the agents more than on the agents themselves. The disagreements are about timing: partners argue for starting now, analysts for caution on spend and data access.

The bottom line

Agentforce is a platform bet, not a quick fix. It pays off for organizations ready to measure one use case at a time, on clean data, with someone owning the cost. For a sales team whose goal is a forecast it can defend next quarter, the cheaper first steps are in its own edition.

Our position, stated

TruthPipe sells a fixed-price pipeline integrity tool that competes with one narrow use of Agentforce: checking the pipeline before forecast calls. We have kept that use in its own line of the tables and given every Agentforce figure its source, so you can check them without taking our word for it.

Five questions before signing

1. Which use case will we measure first, and what number proves it worked?
2. Which pricing model fits our measured usage, credits or the per-user add-on, and when do we review the choice?
3. What do we already pay for? Credits included in our edition, Foundations, Einstein features.
4. What will agent calls through MCP cost once Salesforce publishes the rates, and do we have the 30-day notice in our contract?
5. Who owns the agents after go-live: configuration, permissions, Digital Wallet, and the decision to switch an agent off.

Reading the verdict honestly

Agentforce's revenue numbers say that large customers are buying. The only published estimate of production use, 8% of eligible customers, says that most have not yet deployed agents broadly. Both can be true: platform deals are signed before agents are built. For a buyer, the useful reading is that the platform is real and the deployment work is yours. Budget for that work, not only for the license.

What we could not verify

• Independent accuracy benchmarks for Agentforce sales agents. Koa's figures come from Salesforce's own CRM Bench.
• A typical cost per rep per month in production. Salesforce publishes unit prices, not observed consumption.
• The rates for agent calls through MCP, announced for November 2026.

Agentforce or another tool, by sales job

What to measure in the first 30 days

• Cost per active seller per working day, from the Digital Wallet.
• Share of agent suggestions accepted, for Pipeline Management: below one in three, the suggestions are noise for your team.
• Time saved on the task the pilot targets, measured by asking the same five sellers before and after.
• Data changed by agents and reverted by people. Every revert is a wrong action.

How to run a fair pilot

1. Pick one task an agent can finish without a human, such as logging a call summary to the opportunity.
2. Run it on one team for four weeks. Read the Digital Wallet weekly.
3. Divide credits drawn by active users and working days: that is your real cost per rep per day.
4. Compare with the $125 per-user add-on: above about 60 actions per rep per day, the add-on is cheaper at list price.
5. Measure the outcome you bought it for (hours saved, records completed), not the number of agent runs.

If the outcome you need is a forecast you can defend, start with what your edition already includes; Einstein vs Agentforce lists it by edition.