Every CRM now ships an “agent.” Not a copilot that drafts an email when you ask — an agent that’s supposed to handle a support ticket end to end, qualify a lead overnight, or update records without a human babysitting it. That’s the pitch, anyway.
The hard part isn’t deciding whether you want AI in your CRM in 2026. You already have it, whether you turned it on or not. The hard part is figuring out which platform’s agents are actually load-bearing, what they cost once real usage kicks in, and whether the thing fits the stack your company already runs on. Picking a CRM for its AI agents is a different decision than picking a CRM, and most comparisons still treat them as the same question.
I’ve pulled together what the five major platforms — Salesforce Agentforce, HubSpot Breeze, Microsoft Copilot/Dynamics 365, Zoho Zia, and the value tier underneath all of them — actually charge and actually do, with current 2026 pricing. The short version: the right answer depends almost entirely on how much you want the agent to do on its own and what software your team already lives in.
The shift nobody opted into: copilots became agents
Through 2024 and most of 2025, CRM “AI” meant suggestion engines. Summarize this thread. Draft a reply. Score this lead. A human stayed in the loop on every action.
2026 is the year that loop got cut. Salesforce says Agentforce is now handling billions of agent interactions a month across its customer base. HubSpot claims its Breeze Customer Agent resolves 65% of conversations on its own and cuts resolution time by 39% across roughly 8,000 customers. Those numbers come from vendor marketing, so discount them accordingly — but the direction is real. The agent closes the ticket, not just drafts the reply.
That changes the buying math in two ways. First, pricing moved from “per seat” toward “per thing the agent did” — per conversation, per resolved ticket, per action, per credit. Second, governance suddenly matters a lot more, because an autonomous agent touching customer records is a much bigger liability than an autocomplete box. Keep both in mind as we go.
It also means the old comparison axis — “which CRM has the best AI” — is close to useless now. They all demo well. The question that separates them is consumption pricing you can actually forecast and a permission model you’d trust on live data. A flashy agent with an unpredictable bill and weak audit trail is worse than a plainer one you can reason about.
Salesforce Agentforce: the most capable, and the hardest to budget
If your company is already deep in Salesforce — Sales Cloud, Service Cloud, Data Cloud — Agentforce is the most powerful option here, full stop. Agent Builder lets you define multi-step agents grounded in your own data, Einstein handles the reasoning, and Data Cloud gives those agents the unified customer context that makes their answers good instead of generic. Nobody else has that depth of platform underneath the agent.
The catch is pricing, which Salesforce has rewritten more than once and which remains genuinely hard to forecast. As of 2026 there are two models, and you can’t mix them in one org:
- Conversations: roughly $2 per conversation, where a conversation is a 24-hour interaction session between an agent and a user.
- Flex Credits: $500 for 100,000 credits. A standard Agentforce action burns 20 credits (about $0.10), and a Voice action burns 30.
The crossover is easy to reason about: at 20 actions per conversation you’ve spent the same $2 either way. Heavy, multi-step conversations favor the flat $2 conversation price. Light interactions favor credits. There’s also a seat-based path — Agentforce bundled into Sales/Service editions starting around $125/user/month for unlimited usage by licensed users, up to an Agentforce 1 Edition near $550/user/month that includes a million Flex Credits and Data Cloud credits per year.
Here’s my honest read. Agentforce is the right call when you already run Salesforce as your system of record and you have the implementation muscle — internal admins or a partner — to configure it well. It is the wrong call if you’re hoping to switch it on and see value next week, or if predictable monthly cost is a hard requirement. The Data Cloud consumption fees in particular have a way of showing up larger than the demo implied.
HubSpot Breeze: the one that’s easiest to actually adopt
Breeze is HubSpot’s whole-suite play: agents woven into the CRM you already use, available on Pro and Enterprise tiers, with a 28-day free trial on the flagship agents. For SMB and mid-market teams, this is the fastest path from “we should try CRM agents” to “the agent resolved a ticket today.”
The interesting move in 2026 is pricing. Starting April 14, HubSpot shifted its two headline agents to outcome-based billing:
- Customer Agent: $0.50 per resolved conversation — down from $1.00 per conversation. You pay when it actually closes the ticket, not when it merely replies.
- Prospecting Agent: $1.00 per lead recommended for outreach, replacing the old per-contact-enrolled monthly charge.
Under the hood it’s credits — $10 per 1,000, with a resolved conversation costing 50 credits. Paying only on resolution is a genuinely buyer-friendly structure, and at $0.50 a resolution it’s among the cheapest per-outcome rates in this whole comparison. It also quietly aligns HubSpot’s incentives with yours: they make money when the agent works, not when it stalls.
Breeze won’t match Agentforce’s ceiling for complex, deeply customized enterprise workflows, and the agents are most at home inside HubSpot’s own ecosystem rather than stitched across a sprawling external stack. But for teams that want capable agents without a six-month implementation, it’s the one I’d point most people at first.
Microsoft Copilot and Dynamics 365: pick this if you live in Microsoft 365
The case for Microsoft isn’t that its CRM agents are the most advanced. It’s gravity. If your sellers run on Teams, your data sits in Dataverse, and your IT team governs everything through the Power Platform and Entra, then Copilot agents land inside a world your company already secures and administers. That governance story is the real product.
Pricing in 2026 runs through Copilot Studio credits rather than a flat per-agent fee. A pack is $200/month for 25,000 Copilot Credits, with pay-as-you-go available so you settle on actual consumption. Overage meters at $0.01 per standard message and $0.02 per generative message, and the expensive operations — Dataverse connections, third-party connectors, complex Power Fx — bill at multiples of the standard rate. That last detail matters: a “message” in a real agent often triggers several billable actions, so model your costs on the workflow, not the message count.
One welcome change: Microsoft Copilot for the Dynamics 365 app families (Sales, Service, Finance, Supply Chain) is included in the corresponding Dynamics SKU as of 2026, with no $30/user uplift stacked on top. If you’re already paying for Dynamics, a meaningful chunk of the Copilot experience comes with it.
Choose Microsoft when the center of gravity is Microsoft. Choose almost anything else if it isn’t — the Copilot Studio pricing model rewards teams who already know how to govern Power Platform consumption, and punishes those who don’t.
Zoho Zia: the value play that undercuts everyone on agent cost
Zoho did the thing none of the others did: it made the agents themselves free. Zia Agents — built, deployed, and managed in Zia Agent Studio — carry no per-agent or per-action fee. You bring your own LLM API key, connect the model, and pay your model vendor directly for token usage. Zoho takes nothing on the agent layer.
For Zoho CRM specifically, the bundle lands around $40/user/month including the AI features. Compare that to Agentforce’s per-action credits or Microsoft’s per-message metering and the budgeting difference is stark — your only variable cost is the model usage you already understand, billed by OpenAI or Anthropic or whoever you pick, not a credit currency you have to convert in your head.
The trade-offs are what you’d expect. Zoho’s platform depth and ecosystem don’t reach Salesforce’s, the agent tooling is younger, and “bring your own model” means you own the prompt engineering and quality tuning that a more managed platform handles for you. But for a cost-conscious SMB, or any team that finds the credit-based pricing of the big players genuinely hard to plan around, Zia is the most predictable bill in this comparison by a wide margin. Predictable cost is a feature, and it’s underrated.
A note on the pricing models, because they’re not comparable
One thing that trips up every buyer here: these five don’t bill on the same unit, so a side-by-side number is mostly a mirage. Salesforce sells conversations or credits. HubSpot sells resolved outcomes. Microsoft sells messages that fan out into multiple billable actions. Zoho sells you nothing on the agent and routes you to your model vendor.
The only way to compare them honestly is to take one real workflow — say, a tier-1 support resolution that touches three records and sends two replies — and price that same workflow on each platform. On HubSpot it’s a flat $0.50 if it resolves. On Salesforce it’s roughly five actions, so about $0.50 in credits, or $2 flat under the conversation model. On Microsoft it depends entirely on how many of those record touches hit Dataverse at multiplied rates. On Zoho it’s whatever your model spent on tokens. Run that exercise with your own numbers before you sign anything; the headline rates won’t survive contact with your actual usage.
So which one, actually
Strip away the vendor decks and the choice mostly collapses to two questions: what stack are you already on, and how autonomous do you actually want the agent to be.
Already on Salesforce, with admins or a partner to configure it? Agentforce, and accept that the bill takes work to forecast. Its ceiling is the highest here and Data Cloud grounding is the real differentiator — just don’t expect predictable pricing or fast time-to-value.
SMB or mid-market wanting results without a long implementation? HubSpot Breeze. The outcome-based $0.50-per-resolution pricing is buyer-friendly, the trial is real, and adoption is the fastest of the five.
Your company runs on Microsoft 365 and IT governs through Power Platform? Microsoft Copilot with Dynamics 365. The governance and Teams integration outweigh any single-feature gap, and a lot of it ships inside your existing Dynamics license — but learn the Copilot Studio credit model before you commit, because messages aren’t the same as billable actions.
Cost predictability is the priority, or you want to own your model choice? Zoho Zia. Free agents plus your own LLM key is the cheapest and most plannable structure on offer, with the understood trade-off of a thinner platform and more DIY tuning.
There’s a quieter point underneath all of this. The autonomy that makes these agents worth buying is the same autonomy that makes them risky. An agent that resolves 65% of tickets unsupervised is also an agent that can confidently resolve them wrong at scale. Before you turn one loose on live customer records, the question worth more than any pricing table is how each platform lets you scope, audit, and roll back what the agent did. Run a narrow pilot — one workflow, real data, eyes on every action — and watch what it actually does before you widen the gate.