Customer support AI agent development companies

A customer support AI agent is a program that uses an AI model to answer a customer and to act on the answer. It reads the customer's message, looks up the order or the account in the company's systems, writes a reply, and passes the case to a staff member when the rules say so. A scripted chatbot follows a fixed set of questions and answers that a person wrote in advance. An agent chooses its next step as it goes, inside the limits the company sets. That freedom is why the company has to set clear limits on what a support agent may do.

This page lists the firms in this guide that say they build support agents. Reveneau and Fluxon come first, and the other nine follow in the guide's order, which gives no ranking. Every fact is marked with its source. A firm's own case study is its own account, and no outside source checked any support result on this page. Anthropic's partner directory, Anthropic's list of service firms that work with its Claude models, gives most firms a tier, set by Anthropic: Global Premier, Preferred or Select. The full list of AI agent development companies has the size and office of each firm.

What a support agent is allowed to do

OpenAI, which makes the GPT models, publishes "A practical guide to building agents" [1]. It says an agent needs a way to pass control to a person "when it can't complete a task", and adds, "In customer service, this means escalating the issue to a human agent." It then says, "Two primary triggers typically warrant human intervention." The first is "Exceeding failure thresholds", with the advice, "Set limits on agent retries or actions." Its example is an agent that "fails to understand customer intent after multiple attempts". The second is high-risk actions: "Actions that are sensitive, irreversible, or have high stakes should trigger human oversight until confidence in the agent's reliability grows. Examples include canceling user orders, authorizing large refunds, or making payments." That is OpenAI's own guidance, and it gives a buyer a clear question for every firm: which actions may your agent take alone, and which go to a person?

A support agent's answers matter as much as its actions. In Moffatt v. Air Canada, decided in February 2024, British Columbia's Civil Resolution Tribunal held Air Canada responsible for wrong fare information that its website chatbot gave a customer [2]. That bot answered questions and took no actions. How to test an AI agent before you buy covers the case in full.

Reveneau and Fluxon

Reveneau offers a service to build a company chat assistant and a service for AI development. By its own account, AI writes 100 percent of its code, and each change has to pass a large eval suite, a set of automated tests written from the project's specification, before it is released. Reveneau says Jev, the model that grades the suite, makes the suite run ten times faster than its previous grader. It says it uses AI in place of hiring more engineers and puts that saving into the prices clients pay, and that it takes responsibility for the whole project through production, the live system people use, and after release. Its offices are at 435 Utah Street in San Francisco and on Floor 33, 909 Mainland Street, in Vancouver. Anthropic's partner directory had no Reveneau entry when its 321 entries were searched on 20 October 2026.

Fluxon's AI development page lists "Agents & multi-agent systems" among the things it builds, and says they reduce the time and cost of complex work "while keeping people in control" [3]. A multi-agent system is one in which several agents pass work between them. No support agent case for a named client was found on its pages. Its own site shows Google, Zapier, Puzzle, Eppo and Braintrust as clients. Anthropic's directory had no Fluxon entry on 21 October 2026.

The nine firms at a glance

Firm	Channels it names	How cases reach a person, by its own pages	Its own product in the work
BotsCrew	Website chat, internal assistant for support staff	Switch "from automated flow to Live Chat"	None named
Galadrim	Chat, voice and several channels	Not described on the pages read	None named
Thinkia	Voice and chat	Decides "when AI should hand off to a human"	AI Contact Experience
Lazer Technologies	Support triage, sales, voice	Not described on the pages read	None named
Mactores	Customer support that takes questions in the customer's own words	"what humans must approve" set by its engineers	Aedeon, which it says "builds the agent infrastructure"
Yellow Tech	Phone, chat, WhatsApp, social media, email	"Exceptions go to human review" (document agents)	None named
Linnify	Conversational agents on product data	Not described on the pages read	ARC framework
Bounteous	Salesforce Agentforce Service	"human oversight" in its own platform	Agent Management Platform
First Line Software	Customer conversations, leads, support	Not described on the pages read	Jaime

The firms

BotsCrew. Its homepage case for S&B Filters describes an order-status assistant that works with NetSuite, a business system for orders and accounts, with one assistant for support staff and one for customers on the website [4]. The homepage says an earlier system that connected Claude straight to NetSuite "never reached production", and that BotsCrew "rebuilt the approach from scratch" [13]. It gives three results for that work: "98% faster first response time", "$140k annual cost savings" and "Up to 50% of requests fully automated". The case page lists the same figures, with the 50% given without "Up to", under the heading "Projected Business Impact", and says, "Estimates are based on pre-implementation ticket volume, average handling time reduction driven by AI-assisted responses, and typical support agent cost benchmarks." These are BotsCrew's own estimates, and no outside source checked them. Its customer service page says buyers can "Instantly switch from automated flow to Live Chat" and "Provide the context to your agents", where "agents" means the human support staff who take over the chat. Anthropic lists BotsCrew at the Select tier.

Galadrim. Its agents page lists "Conversational Agents and RAG" and "Voice and Multichannel Agents" among its offers [5]. RAG, short for retrieval-augmented generation, means the agent looks up the company's own documents before it answers. Its test method, by its own description, includes testing "the system on out-of-scope requests and circumvention attempts", meaning questions the agent should refuse and attempts to trick it. Anthropic lists Galadrim at the Select tier.

Thinkia. Its AI Contact Experience product "Resolves queries 24/7 across voice and chat" and "Decides when AI should hand off to a human and who should take the call. Context travels with the transfer." [6] In plain words, the staff member who takes the call sees what the customer already said. This is a product Thinkia owns, so part of what a buyer pays for may be a licence for that product. No named client case was found on its pages. Anthropic lists Thinkia at the Select tier.

Lazer Technologies. Its AI page lists "Customer-facing agents: support triage, sales qualification, voice automation" [7]. Support triage means sorting incoming requests by type and urgency, and sales qualification means asking a new contact the questions that show whether they are likely to buy. Its clients page says, of Hinge, "We helped them build a proof-of-concept for a support chatbot in their app." A proof of concept is a first version built to show that an idea works. Its clients page also lists Decagon, a company that sells support agents, and does not say what Lazer built for it. Outside Anthropic's Select tier, these facts come from Lazer's own pages.

Mactores. Its AI agents page says, "Mactores transformed Safaricom's customer-support experience into an agentic platform. Customers stopped navigating menus and started asking in their own words." [8] An agentic platform is a system built on agents. The same page says its "Forward-deployed engineers own the architecture decisions", including "what the agent is allowed to do, what humans must approve". It also says, "Aedeon builds the agent infrastructure. FDEs design the agent behavior." Aedeon is Mactores' own platform, and FDEs is its short name for its forward-deployed engineers. It promises a fixed date and a fixed fee and prints no amount. Outside Anthropic's Preferred tier, these facts come from Mactores' own pages.

Yellow Tech. Its AI Factory page lists voice agents that "answer the phone 24/7 and in several languages" and "Customer Agents Front office on chat, WhatsApp, social media and email." [9] Front office means the work that faces customers. It describes the difference in one line: "A chatbot answers the question and stops there." Its document agents send exceptions "to human review". Anthropic lists Yellow Tech at the Select tier.

Linnify. Its AI page quotes a client who credits a conversational agent with helping customers find their way through technical product data, and with showing the client where its product materials needed work [10]. The page names the speaker only as a marketing director. Linnify builds on its own agent framework, which it calls ARC, and says the client keeps "your data, your agents, and your IP". IP means intellectual property, the rights to what was built. Outside Anthropic's Select tier, these facts come from Linnify's own pages.

Bounteous. Its Salesforce partner page lists Agentforce Service, Salesforce's own agent product for customer service, among the products it works with. Its agentic page names its own "Agent Management Platform" with "built-in governance and human oversight" [11]. No published support result was found on the pages read. Outside Anthropic's Preferred tier, these facts come from Bounteous' own pages.

First Line Software. Its product Jaime "automates customer conversations, qualifies leads, and answers support questions" [12]. Jaime is a product First Line Software owns and sets up for each client, and its directory text names real estate, healthcare and hospitality as its industries. Outside Anthropic's Select tier, these facts come from First Line Software's own pages.

Three questions for every support agent firm

First, ask which actions the agent may take alone and which go to a person. Use OpenAI's examples, order cancellations, large refunds and payments, as a starting list, and write the answer into the contract.

Second, ask how a case reaches a person and what the person sees. Thinkia's product page describes passing the conversation history with the call. BotsCrew's describes a switch to live chat. Ask each firm to show this in a test with your own staff.

Third, ask for results on a support case like yours, with the measurement behind them. BotsCrew's "Up to 50% of requests fully automated" is an upper figure, in its own words, for one client, and its case page calls the figures estimates. Ask what share of requests the agent finished alone over a full month, and how many of those answers a person checked. What AI agent development costs covers what to ask about price, and the AI agent evals guide explains how such tests are built.