← Blog
Field note

Build vs Buy AI Agent 2026: Agentforce $2/Chat vs Custom Cost

Thien Nguyen · Jul 8, 2026

Buy Salesforce Agentforce or Microsoft Copilot Studio when the task is generic and your data already lives in that platform: list prices start around $2 per Agentforce conversation and $200 for 25,000 Copilot Studio credits a month. Build a custom AI agent when the workflow is your advantage, touches regulated data, or runs at a volume where per-action fees outgrow a one-time build.

Updated 13 November 2026: we re-checked Agentforce and Copilot Studio pricing on salesforce.com and microsoft.com and added a worked cost comparison. Since Microsoft shipped Agent Framework 1.0 in April 2026 and every major SaaS vendor now bundles an "agent builder", the question has flipped. It is no longer "should we use AI agents?" but "should we buy one off the shelf or build our own?". Below is how a senior engineering team actually makes that call, with the cost math and the breaking points that vendor demos usually skip.

Looking for AI coding agents that write your software, such as Cursor, Claude Code or the GitHub Copilot cloud agent? That is a different decision, covered in our AI dev agent platforms comparison. This post is about agents you deploy inside your business.

What does "buying" an AI agent mean in 2026?

Buying means you rent an agent that runs inside a platform you don't own: Salesforce Agentforce for CRM workflows, Microsoft Copilot Studio for Microsoft 365 and Teams, or a range of vertical agents (support, sales, coding). You configure prompts, connect the permitted data sources, and go live in a few days.

The trade-off you accept: the agent can only reason over the data the platform exposes, its actions are limited by the vendor's connectors, and you pay per seat, per conversation or per action. When the platform already holds your system of record, that is a reasonable trade. When important data or logic lives outside it, the gaps show up fast.

When buying an agent makes sense

  • Commodity tasks that look the same at every company: meeting summaries, drafting replies, CRM updates.
  • Use cases that sit entirely within one vendor's data (Agentforce on Salesforce, Copilot on SharePoint/Teams).
  • The team needs an agent running this quarter and can live with the platform's guardrails.

What does "building" an AI agent mean?

Building means you own the orchestration: the model choice, the retrieval layer, the tool/function calls, memory and, most importantly, authorization and audit logging for every action. In 2026 you rarely build from zero; you assemble on frameworks like Microsoft Agent Framework, Claude Agent SDK, CrewAI or LangGraph (compared in our AI agent frameworks guide). The engineering value isn't in the prompt. It's in the integrations, the safety guardrails, and the evaluation harness around it.

That is also where the real cost sits. On real projects, writing the "agent" is only 25–35% of the work. Integrating with your systems, plus QA and safety testing, makes up the remaining 40–60%. Any comparison that ignores this is comparing a subscription fee with a proof of concept, not with a production system.

When should you build a custom AI agent instead of buying Agentforce or Copilot Studio?

Build when the workflow is a competitive advantage, touches sensitive data (HIPAA, PCI), needs integrations the platform can't reach, or runs at a volume where per-conversation fees exceed the cost of owning it. Buy when the workflow is generic, low-risk, and lives in a platform you already pay for. Use the table below as a first-pass filter.

CriterionLeans "Buy"Leans "Build"
WorkflowGeneric, horizontalCore / differentiating
Where the data livesAlready in the vendor's cloudSpread across your own systems
ComplianceStandard, little PHI/PIIHIPAA / PCI / proprietary data
Time to valueDays, the top priorityWeeks, acceptable in exchange for control
VolumeLow or unpredictableHigh and steady (thousands of conversations a month)
Ownership & lock-inVendor dependency is acceptableThe agent itself is IP you must own
Cost shapePer seat/conversation/action, foreverHigh upfront investment, low marginal cost

How much does Agentforce cost per conversation or action in 2026?

On Salesforce's Agentforce pricing page (checked 13 November 2026), customer-facing conversations cost $2 each, and Flex Credits cost $500 per 100,000 credits, with a standard agent action using 20 credits, which works out to $0.10 per action. Per-user add-ons for Sales, Service, Field Service and Operations cost $125 per user per month.

Agentforce option (USD, list)PriceNotes
Conversations$2 per conversationCustomer-facing agents
Flex Credits$500 per 100,000 creditsStandard action 20 credits ($0.10); Agentforce Voice action 30 credits ($0.15)
Agentforce User License$5 per user per monthRequires Flex Credits; limited CRM objects
Agentforce add-ons (Sales, Service, Field Service, Operations)$125 per user per monthUnmetered employee usage
Agentforce Industries add-ons$150 per user per monthUnmetered employee usage
Max EditionFrom $550 per user per monthIncludes the add-on and 2.75M Flex Credits per org per year
Salesforce Foundations$0Agentforce Builder and Prompt Builder included

Salesforce offers three buying models: pre-purchase, pre-commit (billed monthly in arrears with a true-up) and pay-as-you-go. Unused Flex Credits do not roll over, and Salesforce's own cost examples exclude Data 360 credits and other consumption services. The page says prices are for information only and subject to change, so confirm them in your quote.

How much does Copilot Studio cost in 2026?

Copilot Studio costs $200 per month for a tenant-wide pack of 25,000 Copilot Credits (billed annually), or $0.01 per credit pay-as-you-go. A generative answer uses 2 credits, an agent action 5, and tenant graph grounding 10. Employee-facing use by people with a Microsoft 365 Copilot licence ($30 per user per month) is included, within fair-use limits.

Copilot Studio itemRateAt $0.01 per credit
Classic (authored) answer1 credit$0.01
Generative answer2 credits$0.02
Agent action (trigger, reasoning step, topic transition)5 credits$0.05
Tenant graph grounding10 credits$0.10
Reasoning model orchestration (premium)10 credits per 1,000 tokens$0.10 per 1,000 tokens
Agent flow actions13 credits per 100 actions$0.13 per 100

Sources: Microsoft's Copilot Studio billing rates (updated 3 November 2026) and Copilot Studio licensing guidance. A Copilot Credit Pre-Purchase Plan gives 5–20% off for one-year upfront commitments, and unused pack credits don't roll over. Customer-facing agents (your website, your customers) are not covered by employee Microsoft 365 Copilot licences, so they always consume credits.

How much does building vs. buying an AI agent cost?

Platform agents are billed per seat, per conversation or per action: cheap at first, but they compound with usage and never stop. A custom production agent is typically a five- to six-figure (USD) investment, followed by low marginal running costs (model tokens and hosting). The honest comparison is 3-year total cost of ownership, not the first month.

A worked example: 10,000 support conversations a month

Our arithmetic on the list prices above, before discounts, for a customer-facing support agent:

  • Agentforce, per conversation: 10,000 × $2 = $20,000 a month, or $720,000 over three years.
  • Agentforce, Flex Credits: if an average conversation triggers 3 standard actions, 10,000 × 3 × $0.10 = $3,000 a month ($108,000 over three years), plus any Data 360 credits.
  • Copilot Studio: with 4 generative answers and 2 agent actions per conversation, 18 credits × 10,000 = 180,000 credits, about $1,800 a month at pay-as-you-go. Switch orchestration to a reasoning model and Microsoft's own example bills 141 credits for a single follow-up message, which would multiply that figure several times.
  • Employee agent, 50 users: the Agentforce add-on at $125 per user is $6,250 a month ($225,000 over three years); Microsoft 365 Copilot at $30 per user is $1,500 a month ($54,000 over three years) if the agent stays inside Microsoft 365.

The pattern: platform pricing is efficient at low volume and for employees already licensed, and becomes the largest line item at high, steady customer-facing volume. That is where a custom build, whose running cost is mostly model tokens, usually wins over three years.

The numbers that decide it

  • Buy: low upfront investment, predictable unit cost that grows with adoption and ties your workflow to one vendor's roadmap and pricing policy.
  • Build: most of the cost is integration + QA/safety (40–60% of the project), not the model. Over three years, the build itself is often only 25–35% of TCO; the rest is running, monitoring and improving it.

See the line-by-line breakdown in our post on how much it costs to build an AI agent.

Comparing specific off-the-shelf agents? Our OpenAI Dots vs Meta Muse vs custom AI agent comparison prices each option at 5 and 50 users and at 100K tasks. If the agent sits inside a larger product, see AI app development cost in 2026 for build and token costs by architecture.

What are the hidden costs of buying an AI agent platform?

The biggest hidden costs are usage meters that grow with conversation length and model choice, data platform charges outside the agent price, and the integration work for systems the vendor's connectors don't reach. Licences are the visible part; these are what blow up the budget.

  • Meters that scale with design, not users. In Copilot Studio, reasoning models bill per 1,000 tokens, including the tokens the model uses to think, so long instructions, large knowledge results and long conversations cost more. In Agentforce, every action in a multi-step flow is metered.
  • Data charges outside the agent price. Salesforce's examples exclude Data 360 credits; Copilot Studio requires an Azure subscription for pay-as-you-go.
  • Use-it-or-lose-it capacity. Unused Flex Credits and Copilot Studio pack credits don't roll over.
  • Hard stops on overage. Copilot Studio blocks agents when a tenant on prepaid capacity reaches 125% of it, unless pay-as-you-go is linked; users then see an "agent is currently unavailable" message.
  • Integration and testing anyway. Connecting an EHR, billing system or internal API, and testing the agent's actions before it touches customer records, is engineering work whether you buy or build.
  • Admin and governance time. Someone has to own prompts, permissions, capacity allocation and monthly usage reviews.

Can you start with a bought agent and move to a custom one later?

Yes, and it is often the right sequence: use a platform agent to prove the workflow and measure real volume, then build once the numbers justify it. Plan the exit from day one, because prompts, flows and conversation logs built in a vendor's studio do not port to another stack automatically.

What makes the move cheap later:

  1. Keep business logic in your own APIs. If the bought agent calls your services for every action, a custom agent can call the same services later. Exposing them through an MCP server makes them reusable by any agent; see MCP server development.
  2. Export conversation logs and outcomes. They become the test set for the custom agent.
  3. Write down the evaluation criteria (accuracy, escalation rate, cost per resolved case) while you are on the platform, so you can prove the replacement is at least as good.
  4. Watch the contract. Annual packs and pre-purchase plans run for a year; time the switch to the renewal.

Which model fits you? Three scenarios

  1. Buy: a 40-person SaaS team wants to triage support tickets right inside Salesforce. The data is already there and the task is generic, so Agentforce delivers value within a week.
  2. Build: a healthcare startup needs an agent that reads PHI across the EHR and billing systems to prepare prior-authorization packets. Sensitive data, cross-system logic, auditability required: this is a build problem.
  3. Hybrid: most scale-ups buy Copilot for internal productivity and build exactly one customer-facing agent that creates their product advantage. Buy the commodity, build the moat.

Where do no-code and "vibe-coded" agents break?

They break exactly in the 40–60% that platforms and demos skip: reliable integrations, error handling, authorization for every action, and evaluation. A no-code agent that answers questions is easy; an agent that takes actions on customer accounts, correctly, every time, is a systems-engineering problem. That gap is why so many pilots never reach production; see why AI agent projects fail.

Before an agent gets write access to customer accounts, check it against our AI agent security checklist (24 controls). If it has to expose tools to Claude, ChatGPT or other assistants, MCP server development covers the cost, architecture and security of doing that properly.

What about HIPAA, PCI and regulated data?

If the agent touches PHI or cardholder data, the "buy" options narrow sharply: you need a BAA, minimum-necessary limits on what the agent can retrieve, encryption, and logs for every access decision, and not every platform or plan will sign or support that. Regulated workflows are usually where building wins, because the controls have to live in code you own. We cover this in depth in HIPAA-compliant AI agents.

For the evaluation side, meaning how to test an agent so the results hold up in an audit, see LLM testing for regulated industries.

If you've scrutinized the workflow and the answer is "build", that's what we do: senior engineers who own the integrations, compliance and evaluation, not just the prompt, at a fixed price per phase (from $10K) with you owning the code, as described in fixed-price AI agent development. See how we approach AI agent development, or tell us about your workflow and we'll give you a straight assessment of whether to build or buy.