1 Custom build by Black Badger
Best for agents that write to your own systems under your own rules
An agent that only answers can be wrong cheaply. An agent that acts, by creating an order or changing a record, has to be right, permitted and logged every time. That is why custom is first on this list. Black Badger builds the agent against your actual systems, with your approval rules, a record of every action and a person in the loop where you want one. You own the code and the accounts.
Watch for. Start with a single action that is frequent and low risk. The written scope names which actions the agent may take and which always need a person.
- Price
- One-time build scoped in writing, no per-seat fee After launch you pay running costs such as hosting, database and email. There are no licenses, so adding a person adds no fee.
- Main limit
- Needs a scoping call first, and suits companies of 50 or more employees
What a custom build involves, and how to ask about one
2 Microsoft Copilot Studio
Best for agents that act across Microsoft 365 and Dataverse
Copilot Studio is the best product for agents that act, as long as the systems are Microsoft ones. It is sold for building agents, and credits are billed when an agent completes an action or response, which ties cost to work done. It ranks second because that billing follows volume, and Power Platform requests are capped at 250,000 per tenant every 24 hours on the standard subscription.
Watch for. Agents need an Azure subscription, and cost is hard to forecast before use because no per-credit rate is published for pay as you go.
- Price
- $200 per 25,000 Copilot Credits a month, or pay as you go Credit packs are tenant-wide and billed up front. Pay as you go has no upfront commitment and needs a linked Azure subscription. Credits are billed when an agent completes an action or response, and the number billed varies by usage. A pre-purchase plan offers up to 20 percent savings.
- Main limit
- Usage is billed per credit, so cost follows traffic. Quotas apply per environment.
3 Glean
Best for agents that work on content across many connected systems
Glean ranks third. Beyond search, it acts on content from the systems you connect. Its security page says users and agents are verified through your SSO provider before AI runs, with source-system permissions enforced on every write as well as every read. An agent needs both checks. It ranks third because what it can act on is set by the connectors, and no price is published.
Watch for. Ask which actions each connector supports in the systems you use, since that sets what an agent can do.
- Price
- Not published, quoted per customer Glean publishes no price. Its pages send buyers to a demo request.
- Main limit
- No public price or plan limits. Sold through a demo and a quote.
4 Retool
Best for developers who want AI steps inside tools they already built
Retool ranks fourth. A developer team that already connects tools to its databases and APIs can add AI to those tools, so an action runs through connections the team controls. Audit logs start on the Business plan. It ranks fourth because this is a builder for developers and not an agent product, and bring-your-own AI key is listed only for Enterprise.
Watch for. AI credits are pooled for the account and do not roll over, with 3,000 a month on Business.
- Price
- Builder $10, internal user $5 a month, billed yearly Team plan prices per user a month. Business is $50 per builder and $15 per internal user billed yearly. Monthly billing costs more. A builder is anyone who built or edited an app in the billing cycle, and everyone else enabled is an internal user. Enterprise is custom priced.
- Main limit
- SSO, SCIM, Git source control and granular roles are Enterprise only.
5 Claude Team and Enterprise
Best for staff-directed work through connectors, with an audit trail
Claude Team and Enterprise is fifth. Through connectors, staff can have the assistant work with company tools, and Enterprise adds audit logs, role-based access and a Compliance API for oversight. It ranks fifth on this question because Claude works inside its own apps and connectors. An agent embedded in your own systems means building on the API separately, which is a custom project.
Watch for. Enterprise has a 20 seat minimum, and usage is billed at API rates on top of the seat price.
- Price
- $20 a seat a month billed yearly, $25 monthly That is the Team standard seat, for teams of 2 to 150, with up to $500 a month in pooled API credits. A Premium seat is $100 a seat billed yearly, or $125 monthly. Enterprise is listed at $20 a seat a month billed yearly with a 20 seat minimum, plus usage billed at API rates.
- Main limit
- Team caps at 150 seats. Audit logs and SCIM are Enterprise only.
6 ChatGPT Business and Enterprise
Best for assistants staff use by hand, not agents inside systems
ChatGPT Business and Enterprise ranks sixth for this question. It gives every seat a general assistant, and custom GPTs let staff shape it. It is last on this list because of what the question asks. A custom GPT runs inside ChatGPT, and the product does not fit a company that needs an assistant tied tightly to its own systems and permissions.
Watch for. If the need is an agent inside your systems, treat ChatGPT seats as a separate purchase for general staff use.
- Price
- $20 a user a month billed yearly, $25 monthly That is the Business Standard seat, with a minimum of 2 users. A Premium seat is $100 a user a month billed yearly, or $125 monthly, with more usage. Enterprise is priced per user per month on annual terms through sales, with invoicing available.
- Main limit
- Business and Enterprise can buy extra credits when included use runs out.