- Published
- Updated
- 9 min read

TL;DR: The big CRM and office platforms now sell prebuilt AI agents to small businesses, bundled into a plan, billed per outcome in credits, or added per seat. They do common work well and your specific work poorly. Check what you can change, how credits bill, what you keep if you leave, and who approves actions.
Update, October 2026: an earlier version of this note quoted market losses, a leaked company memo, project cost averages and a customer story that we could not trace to a primary source. Those are gone. What remains is checked against each vendor's own pages and established press.
Prebuilt AI agents are now part of the pitch from nearly every platform a small business already pays for. Your CRM offers one. Your office suite offers one. The message is the same each time: switch it on and the work gets done. I spent time with what Salesforce, HubSpot and Microsoft actually publish about their AI agents, because the details decide whether one helps your business or quietly runs someone else's playbook inside it.
None of this makes the products bad. Some of them are a fine choice for routine work. The point is to know what you are buying before the renewal date arrives.
What are the big platforms selling small businesses as AI agents?
Three patterns show up across the vendors. One bundles an agent into the plan you already have. One charges per result, paid in credits. One adds AI to each seat. Each model shapes what you can change and what you end up paying.
Salesforce: an agent included in the small business suites
In March 2026 Salesforce said "The Employee Agent is included in the Starter and Pro Suites," that it can "provide instant summaries, draft personalized communications, and log activities," and that teams can turn it on "with no setup or extra costs." (Salesforce (opens in a new tab))
The limits sit in the fine print. An admin review of the small business version found "you do not get access to Agent Builder or Prompt Builder" and "you cannot customize or add to them." (Salesforce Ben (opens in a new tab)) For summaries and email drafts, that is fine. For the way your shop quotes a job, it means the agent works the vendor's way or not at all.
HubSpot: AI agents billed per outcome, paid in credits
HubSpot moved two of its AI agents to outcome pricing on April 14, 2026. The customer agent went "from $1.00 per conversation to $0.50 per resolved conversation," and the prospecting agent now costs "$1.00 per lead recommended for outreach." Both are "available to Pro and Enterprise customers," and HubSpot says its customer agent "already resolves 65% of conversations." (HubSpot (opens in a new tab))
Paying for results sounds fair, and in many ways it is. The billing runs on HubSpot Credits, and the account limit "remains the overall limit for credit usage across all HubSpot tools." Credits "expire at the end of each usage period and do not roll over." If you bought extra credits before September 16, 2026 and hit the limit, the account "will automatically be upgraded," and "You can only downgrade your HubSpot Credits tier at the end of your contractual commitment term." (HubSpot Knowledge Base (opens in a new tab)) So a busy prospecting month can use up credits your support agent needed, and an upgrade can follow you to the end of the contract.
Microsoft: AI as a per-seat add-on
Microsoft sells Microsoft 365 Copilot Business as an add-on. Its pricing page lists it as "Originally starting from $21.00 now starting from $18.00" per user per month on a yearly plan, with the discount running from July 1 to December 31, 2026, and $25.20 per user per month billed monthly. (Microsoft (opens in a new tab)) Per seat is easy to budget. It also means a ten-person team pays for ten seats whether three people use it or ten.
What do the failure numbers say about AI agents?
The research is not kind to agent projects, but read it closely, because the reasons matter more than the headline.
Gartner's forecast for agent projects
Gartner predicted that "More than 40% of agentic artificial intelligence projects will be canceled by the end of 2027," citing "escalating costs and unclear business value." It also warned about "agent washing," which it described as "the rebranding of products such as AI assistants and chatbots without significant agentic capabilities," and estimated that "only about 130 of the thousands of agentic AI vendors are real." (Reuters (opens in a new tab))
That second point is the one I would hold onto. A product labeled as AI agents may be a chatbot with a new name. Ask what it can do without a person clicking each step.
Why the MIT finding does not mean "build it all yourself"
MIT's NANDA initiative reported that "about 5% of AI pilot programs achieve rapid revenue acceleration." Fortune's write-up traced the gap to a "learning gap" and poor integration with how companies work, not to model quality. The same report found that "Purchasing AI tools from specialized vendors and building partnerships succeed about 67% of the time," while internal builds succeed "only one-third as often." (Fortune (opens in a new tab))
An earlier version of this note used that study to argue for custom builds. That was a stretch. The finding points at fit and help, not at writing everything from scratch. The projects that worked had someone who understood both the tool and the business. A prebuilt agent with a good partner can beat a homemade one that nobody maintains.
Where do prebuilt AI agents fit, and where do they fall short?
I sort a business's work into two piles before anyone buys AI agents.
Commodity work versus the work that makes you money
Commodity work looks the same in every company: summarizing a call, drafting a follow-up email, logging an activity, answering "what are your hours." Prebuilt AI agents are built for exactly this, and the vendor has seen millions of examples. Buy it if the price is right.
The second pile is the work that makes your business different: how you price a custom job, which leads you chase first, how you handle the customer who calls every week. A prebuilt agent handles this the way an average business would, and your business is not average. That is where fit matters, and where an agent you cannot change becomes a ceiling.
Questions to ask before you sign
- What can we change? Instructions only, or the steps, tools and rules the agent follows?
- How does billing work when we are busy? Are credits shared across tools, do they expire monthly, and what happens when we run over?
- What do we keep if we leave? The data, yes. What about the instructions, the setup and the history of what the agent did?
- Who approves what the agent does? Can a person review a refund, a quote or an email to a customer before it goes out?
- What are we paying for that nobody uses? Count active users after the first month, not at signing.
If a sales rep cannot answer these in writing, that is an answer too.
How should AI agents connect to the tools you already run?

Most small businesses run a mix: accounting software, a CRM set up years ago, a scheduling tool, and a few spreadsheets that hold more of the operation than anyone admits. A vendor's agent sees the vendor's data well and the rest poorly. The useful question is how an agent reaches the other systems, and who controls that link.
Open standards keep the connections portable
The Model Context Protocol, or MCP, is "an open-source standard for connecting AI applications to external systems." Its own documentation says to "Think of MCP like a USB-C port for AI applications." (Model Context Protocol (opens in a new tab)) When AI agents reach your tools through an open standard, the connection outlives any one vendor. That is a practical guard against vendor lock-in, because switching the model or the platform does not mean rebuilding every link.
Give each agent only what it needs
OWASP, the security standards community, lists "Excessive Agency" as a top risk for AI applications. It traces the cause to "excessive functionality," "excessive permissions" and "excessive autonomy," and recommends you "Limit the permissions that LLM extensions are granted to other systems to the minimum necessary" and "require a human to approve high-impact actions." (OWASP GenAI Security Project (opens in a new tab)) A prebuilt agent that can read and write everything in your CRM by default is convenient. It is also more access than most tasks need.
What should a small business do about AI agents this quarter?
First, take the pressure off. The Census Bureau found that national business AI use was 19.8% in May 2026, and that use "didn't change significantly among firms with fewer than 20 employees." (U.S. Census Bureau (opens in a new tab)) Most small firms have not adopted AI agents yet. You are not behind for asking questions first.
Then take these steps in order:
- List the repetitive work. Write down five tasks your team repeats every week and how long each takes.
- Sort them. Mark each one as commodity work or work that makes you money.
- Try the included agent on the commodity pile. If your CRM or office suite already includes one, test it there before you buy anything new.
- Price the busy month, not the quiet one. For any credit or outcome pricing, estimate your heaviest month and read the overage terms.
- Plan the money pile separately. That work needs AI agents set up around your rules, with a person approving what matters and a way to change the setup when your process changes.
Good AI implementation starts with the workflow and ends with the tool. Buying first and fitting the business to the agent afterward is how projects land in Gartner's 40%.
Frequently asked questions
Are prebuilt AI agents worth it for a small business?
For common work like summaries, email drafts and logging activity, often yes, especially if your current plan already includes one. For the work that sets your business apart, check first whether you can change how the agent works.
How do outcome-priced AI agents bill?
HubSpot, for example, charges $0.50 per resolved conversation and $1.00 per recommended lead, paid in credits. Credits share one account limit across tools and expire monthly, so a heavy month in one area can affect another.
Can I customize the AI agent included in my CRM?
It depends on the plan. Salesforce's small business suites include an Employee Agent, but an admin review found those plans do not include the builder tools needed to change it. Ask the vendor in writing.
Why do so many AI agent projects fail?
Gartner points to rising costs and unclear business value. MIT's research points to a learning gap and poor integration. Model quality was rarely the problem; fit with the business was.
Should a small business build its own AI agents instead?
Not by default. MIT found purchased tools and partnerships succeeded far more often than internal builds. The better question is who sets the agent up around your process and keeps it running.
How do I avoid lock-in with AI agents?
Keep your data exportable, keep instructions and rules written down outside the vendor, and connect agents to your tools through open standards such as MCP where you can.
Where we fit
This is the work our AI implementation practice exists for: taking an idea out of the demo and into daily work, where input is messy, services fail, and a person checks the cases the model is unsure about. We pick the workflow, define what good output looks like, build with quality gates on every output, and keep a human path for anything uncertain. When a job needs several steps, our multi-agent workflows give each agent one task, the tools it needs and a stop condition, and our API integration work connects them to the tools you already pay for.
We run it for you on LTFI, where contacts, projects and AI staff sit in one place (opens in a new tab), so nothing is installed on your side. We do not build systems meant to replace people. We remove bottlenecks, and your people drive the work.
First conversation is free. No commitment.




