Cost to Build an AI Agent in 2026: Builder, Platform or Custom, With Real Numbers
The three ways a 10 to 100 person company gets an AI agent in 2026, what each costs to build and to run for 2,000 agent runs a month, and the one complexity that justifies a custom build.
Jun 11, 2025
An AI agent for a 40-person company costs between about $600 a year and $100,000 to build in 2026, depending on which of three routes you take: an agent builder subscription (Lindy from $49.99 per user per month, read 19 October 2026), a platform agent on a suite you already pay for (Microsoft Copilot Studio at $200 a month per 25,000 credits, ChatGPT Business at $20 per seat per month billed annually), or a custom agent built on OpenAI or Anthropic models, which KUMO prices at $20,000 to $50,000 for one workflow and $50,000 to $100,000 when it works across several systems. The running cost is the part most quotes leave out: at today's model prices, 2,000 agent runs a month cost between $3 and $60 in model fees, which means the engineering, not the model, is the number that matters.
Key facts (updated 19 October 2026)
- Agent builder: Lindy Team $49.99 a month per active user for 5,000 credits, top-up $10 per 1,000 credits; an everyday task uses 2 to 250 credits, deep work 250 to 1,000
- Platform agent: Microsoft Copilot Studio $200 a month per pack of 25,000 Copilot Credits, or pay as you go, Azure subscription required; ChatGPT Business $20 per seat per month billed annually ($25 monthly) with workspace agents; Claude Team $20 per seat per month billed annually ($25 monthly)
- Custom agent build: $20,000 to $50,000 for one workflow and one integration set; $50,000 to $100,000 for multi-step agents across several systems or with a portal
- Model fees at 2,000 runs a month and 5,000 tokens per run: about $3 a month on Claude Haiku 5.5, about $60 on Claude Sonnet 5.5 (Anthropic prices read 19 October 2026)
- After launch: hosting (a few hundred dollars a month) and ongoing engineering of $5,000 to $10,000 per month if you have no engineers
Who this is for
An owner or operations lead at a company with 10 to 100 staff and no engineers, who has one process that eats hours every day: triaging inbound requests, chasing approvals, pulling data from three systems to answer one question, preparing quotes, reconciling invoices. You want to know whether a subscription will do it, whether the suite you already pay for will do it, or whether you need something built, and what each costs before anyone sends you a proposal.
What each route costs for 2,000 agent runs a month
Prices read 19 October 2026 from the vendors' pricing pages unless dated otherwise; billing basis as stated.
| Route | What you pay | Year one for a 40-person company | What you get |
|---|---|---|---|
| Lindy Team (agent builder) | $49.99 a month per active user, 5,000 credits each, pooled; $10 per 1,000 extra credits | $600 for one operator seat; $6,000 if ten people use it | Scheduled routines, inbox and calendar work, 1,500+ integrations, approvals built in, model choice per task |
| Relevance AI (agent builder) | Enterprise pricing only, not published | Quote | Multi-agent workforces, evaluations, SSO and audit logs; aimed above the 40-person company |
| Microsoft Copilot Studio (platform) | $200 a month per 25,000 Copilot Credits, or pay as you go; Azure subscription required | $2,400 for one pack a month, plus Microsoft 365 you already pay for | Agents with 1,500+ connectors, multi-agent orchestration, evaluation, governance in Entra and Purview |
| ChatGPT Business (platform) | $20 per seat per month billed annually, $25 monthly; Premium seat $100 | $9,600 for 40 standard seats | Workspace agents, connectors to Google Workspace, Slack, Microsoft 365, no training on your data |
| Claude Team (platform) | $20 per seat per month billed annually, $25 monthly; Premium seat $100 | $9,600 for 40 standard seats | Connectors, skills, enterprise search, up to $500 a month in pooled API credits |
| Custom agent, one workflow (KUMO) | $20,000 to $50,000 once | $20,000 to $50,000 plus model fees and hosting | Your rules, your systems, approvals and audit log, tests on real data, handover of code and prompts |
| Custom agent, several systems or portal (KUMO) | $50,000 to $100,000 once | $50,000 to $100,000 plus model fees and hosting | Multi-step agent across CRM, finance and documents with a human review queue |
Want the number for your own scope? Send a short brief: what takes your team the most manual time, which systems are involved, roughly how many people touch it. A senior engineer replies within two working days with a band and the questions we would ask on a call. No call needed. Or get a range in a minute from the AI development cost calculator.
The builders are cheap and fast and right for personal and small-team work. The platform route is nearly free if you already pay for Microsoft 365 or ChatGPT Business and your process fits what their agents can do. The custom route costs ten to fifty times more up front and is only right when the process does not fit the other two, which is the next section.
What the model costs to run
Model fees are billed per million tokens (MTok). Anthropic's published API prices on 19 October 2026: Claude Haiku 5.5 at $0.10 input and $0.50 output per MTok for prompts up to 100K tokens; Claude Sonnet 5.5 at $2 input and $10 output; Claude Opus 5.5 at $4 input and $20 output. OpenAI's per-token API prices were not captured on the same day and are omitted here; they sit in the same order of magnitude.
Worked example: an agent that reads a request, looks up two systems and drafts an action uses about 4,000 input tokens and 1,000 output tokens per run. At 2,000 runs a month that is 8 million input tokens and 2 million output tokens.
- Haiku 5.5: 8 × $0.10 + 2 × $0.50 = $1.80 a month
- Sonnet 5.5: 8 × $2 + 2 × $10 = $36 a month
- Opus 5.5: 8 × $4 + 2 × $20 = $72 a month
Prompt caching cuts the input side further when the same instructions repeat. The point: even on the most expensive model, the model fee for a working agent at this volume is less than one hour of engineering a month. Agent cost is an engineering cost, not a model cost, until you reach hundreds of thousands of runs.
The one complexity that justifies custom
A builder or platform agent is the right answer when the task lives inside one or two tools and the rules are simple: summarise, draft, look up, schedule, file. Configuration is enough when the platform's connectors reach your systems and its approval step is the only approval you need.
A custom agent is justified by one of four things, and nothing else: approval rules that differ by record (per owner, per customer, per amount); actions across three or more systems where the order and the rollback matter; a review queue with audit trail because money or client data moves; or a cost per run that grows with volume on a subscription. If none of those is true of your process, do not commission a build; configure the platform and keep the money.
What the build includes and what sits on top
Included in the $20,000 to $50,000: discovery against your real tickets or requests, the agent, integrations with your systems, the approval and review step, an evaluation set built from last quarter's cases, tests, handover of code, prompts, eval sets and runbooks.
On top: model fees (see above), hosting (typically a few hundred dollars a month), the platform plan if your systems only expose their API on a higher tier, and ongoing engineering at $5,000 to $10,000 per month if you have no engineers to own prompt and model updates when vendors ship new versions.
Proof, labelled
CampaignHQ is KUMO's own workflow product, in production since 2022; it is where KUMO's agent patterns run first, so this proof is direct for the agent and approval mechanics. KUMO's founders built Volopay's first production version as its founding engineers, a finance product where every approval and every payment had to be traceable; that proof is adjacent here, relevant to the approval and audit parts of an agent. KUMO is a software engineering studio that builds custom AI agents, workflow apps and portals for 10 to 100 person companies, and a Registered Partner in the OpenAI Partner Network.
Questions buyers ask
How much does it cost to build an AI agent for a 40-person company in 2026? $600 a year on an agent builder seat, $2,400 a year in Copilot Studio credits on top of Microsoft 365, or $20,000 to $50,000 once for a custom agent that follows your rules across your systems. The right number depends on whether your process fits a builder, which the four tests above decide.
What does an AI agent cost to run per month once it is built? Model fees of roughly $2 to $72 a month at 2,000 runs depending on the model, hosting of a few hundred dollars, and ongoing engineering of $5,000 to $10,000 per month if nobody in-house owns it. The subscription routes replace the first two with the plan fee.
Is Copilot Studio or ChatGPT Business cheaper than a custom AI agent? Yes, by a wide margin in year one: $2,400 in credits or $9,600 in seats against $20,000 to $50,000. They stop being cheaper when the process needs per-record approval rules, actions across three or more systems, or an audit trail, because then the work moves to people around the platform.
How much of the AI agent cost is the model and how much is the engineering? At 2,000 runs a month the model is under $100 a month; the engineering is the whole build price. Model cost only becomes a line item at hundreds of thousands of runs, which is when a custom agent on a cheaper model beats per-run subscription pricing.
Can I start with a no-code agent builder and move to custom later? Yes, and it is the right order for most companies. What carries over is the process definition and the examples; what does not is the configuration, so keep a written record of the rules and the failure cases from day one.
What makes an AI agent cost more than $50,000? Several systems in one flow with ordering and rollback, a portal for customers or vendors, data residency requirements, or document processing with human review at volume. Each adds integrations and test cases; the band for that shape is $50,000 to $100,000.
Compare the platform routes in Copilot Studio vs ChatGPT Business vs a custom AI agent, see how KUMO scopes agents in AI workflow automation, run your own numbers in the AI development cost calculator, or send a short brief and get a scoped estimate in two working days, or book a 30-minute call.
The numbers in one file. The Build vs Buy Price Book, October 2026 has the published prices of 40 business tools read in October 2026, what they cost a 40-person company, and the four tests that say whether a custom build is worth it. Download the PDF (8 pages, free, no form).
In this series: AI agents for business
- Private AI Agents for Business: How They Work and What They Cost (2026)
- Best AI Agent Builders for a 40-Person Company in 2026: Prices, Ceilings and When to Build
- Private AI Agent Deployment: Cloud, VPC, or On-Premises?
- AI Agent Security Checklist for Production Teams in 2026
- AI Agent Pilot Plan: 30-60-90 Day Roadmap for Revenue-Stage Teams
- AI Agent Permissions Matrix for the First Production Workflow
The whole series, with the in-depth guide first: AI agents for business. For the build itself, see how KUMO delivers this.
