Top AI agent development companies

An AI agent is a computer program that uses an AI language model, a model that reads and writes text, to finish a task in several steps by itself. A support agent might read a customer's message, look up the order, write a reply and pass the case to a person when the rules say so. A back-office agent might read an invoice, check it against a price list and send it on for approval. A coding agent might read a bug report, change the code and run the tests. The steps are chosen by the model as it goes, inside limits that people set.

This guide lists seventeen firms that a company can hire to build agents of this kind for it. Reveneau is first and Fluxon is second. The other fifteen follow in a fixed order with no ranking, and no firm gets a score. Every page used was read on 21 October 2026, apart from one search of Anthropic's directory on 20 October, and each sentence says where its fact comes from, because a firm's own website, a government register and a court decision deserve different levels of trust.

What makes a program an agent

Anthropic, the company that makes the Claude models, makes a distinction in its own guide, "Building effective agents", published on 19 December 2024 [2]. It calls a workflow a system in which models and tools follow "predefined code paths", steps that a programmer wrote in advance. It calls an agent a system "where LLMs dynamically direct their own processes and tool usage". An LLM, or large language model, is the type of AI model behind tools such as Claude and ChatGPT. Anthropic uses "agentic systems" as one name for both workflows and agents. Its guide also says, "When building applications with LLMs, we recommend finding the simplest solution possible, and only increasing complexity when needed. This might mean not building agentic systems at all." That is advice from a company that sells the models many agents run on. A note at the top of the page, read on 21 October 2026, says much of the tooling it describes "has changed since December 2024".

The difference matters when you buy. A firm that builds a workflow writes every step, so the steps can be checked one by one. A firm that builds an agent lets the model pick its next step, so the firm has to test what the agent does across many runs. How to test an AI agent before you buy explains what that testing looks like.

What one survey of executives found

KPMG US, the US arm of the accounting and advisory firm KPMG, publishes a quarterly survey of executives about AI. Its AI Quarterly Pulse Survey for the third quarter of 2026, published in September 2026, says "the percentage of surveyed executives reporting deployment of AI agents increased from 53% to 62% since last quarter" [3]. The same page describes the same figure a second way: "62% of organizations are currently building, testing, or deploying AI agents, up from 53% last quarter." The first wording counts executives who report deployment, and the second counts organisations building, testing or deploying, so a reader should note both. The page also gives a figure for multi-agent systems, meaning systems in which several agents pass work between them, in two ways: "25% are developing or deploying multi-agent systems" and "25% of organizations have deployed multi-agent systems". These figures describe the executives KPMG surveyed and their organisations. The page read gives no sample size and does not describe who was asked, so the figures do not show how many companies in general use agents. KPMG sells AI services itself, so it is a third party to the firms in this guide, though it has its own interest in AI.

Agents also fail in ways a demo does not show. Shunyu Yao and three other researchers built a test called tau-bench, in which an agent talks with a simulated user, a user played by a language model, and has to follow a set of business rules while it uses the tools it is given. In their paper, first posted on 17 June 2024, the authors report that "even state-of-the-art function calling agents (like gpt-4o) succeed on <50% of the tasks" [4]. GPT-4o is a model made by OpenAI. In the retail part of the test, the share of tasks those agents solved in all eight of eight tries was "<25%". Those are the authors' own figures for the agents and models they tested in mid-2024, and newer models will score differently. The point for a buyer stays the same: ask to see the same test run many times.

The list at a glance

Anthropic's Claude Partner Network directory lists the companies Anthropic names as service partners and gives most of them one of three tiers: Global Premier, Preferred or Select. Anthropic sets the tier. Each partner writes its own one-line description and picks its own office regions. The directory held 321 entries when it was read on 21 October 2026: 256 at Select, 40 at Preferred, 16 at Global Premier and 9 with no tier [1].

Firm	Kinds of agent, from its own pages	Main office, and who says so	Size, and who says so	Anthropic tier
Reveneau	Company chat assistant, workflow automation (own service pages)	San Francisco and Vancouver (own site)	Not stated	Not listed (searched 20 October 2026)
Fluxon	"Agents & multi-agent systems" (own AI page)	San Francisco (Stripe partner listing, address from Fluxon)	200 employees (Forbes, as of March 2026); "300+ builders" (own careers page)	Not listed
BotsCrew	Customer support, internal assistants	Lviv, Ukraine, and five US addresses (own footer); AIN.UA reports the team works in Ukraine	60 people (AIN.UA, February 2025)	Select
Galadrim	Customer chat, internal agents, documents	Paris (French company register)	50 to 99 salaried staff in 2024 (French register); "170+ experts" (own page)	Select
Thinkia	Contact centre, internal processes, coding	Las Rozas, Madrid (own legal notice)	Not stated on the pages read	Select
Lazer Technologies	Customer-facing, internal, engineering work	Toronto (own privacy policy)	"200+" engineers, designers and product managers (own about page)	Select
Mactores	Customer support, claims	Bellevue, Washington (own about page)	"200+ AWS-certified engineers" (own about page)	Preferred
Yellow Tech	Phone, customer chat, documents	Milan (EU VIES VAT system)	Not stated on the pages read	Select
Linnify	Documents, internal systems	Cluj-Napoca, Romania (own AI page)	"30+ Creators" (own about page)	Select
Awiant	Invoice checks and other internal work	Stockholm (own contact page)	35 people (own about page)	Select
Cay Digital	Finance and operations, coding	Jupiter, Florida (own homepage)	Not stated on the pages read	Select
Zemoso Labs	Internal work, coding	Plano, Texas, in the Dallas area (own footer)	"600+ people" (own about page)	Select
AscentCore	Coding	Herndon, Virginia (own contact page)	Not stated on the pages read	Select
Diverger	Moving old software, internal work	Alcobendas, Madrid (own legal notice)	Not stated on the pages read	Select
HAVIT	Coding, internal systems	Prague (Czech ARES register)	"30 HUMANS" (own homepage)	Select
Bounteous	Sales and service, internal, coding	Frisco, Texas, and Boston (own US page)	"5,000-employee" and "more than 5,000 team members" (own pages)	Preferred
First Line Software	Leads and support, invoicing and accounts, coding	Cambridge, Massachusetts (own contact page)	"500 + Engineers" (own about page)	Select

How to read each entry

Each entry gives the facts first. Fits then names the buyer the firm suits, judged from what the firm or a register says, and Pick someone else if names a case where a different firm would suit you more. Words such as "says", "by its own account" and "its page lists" mark a fact from the firm's own website. When a sentence names a register, a court, a news site or a survey, the fact comes from that source. A register is an official public list of companies kept by a government body. VIES is the European Commission's system for checking that an EU sales tax number is valid. How these AI agent development companies were chosen gives every check and every firm that was left out.

The firms

1. Reveneau

Reveneau has offices at 435 Utah Street in San Francisco and on Floor 33, 909 Mainland Street, in Vancouver, Canada. By its own account, AI writes 100 percent of its code. Reveneau says each change has to pass a large eval suite before it is released. An eval suite is a set of automated tests that check what the software says and does, and Reveneau says its suite is written from the specification, the document that sets out what the software must do. It says Jev, the model it uses to grade the suite, makes the suite run ten times faster than its previous grader did. Reveneau says it uses AI in place of hiring more engineers and puts that saving into the prices its clients pay. It says it takes responsibility for the whole project through production, the live system that people use, and after release. Reveneau offers services to build a company chat assistant and workflow automation. Anthropic's partner directory had no Reveneau entry when its 321 entries were searched on 20 October 2026 [1].

Fits: a company that wants agents and the software around them written by AI and tested against the specification before each release. You can contact Reveneau to describe the agent you need.

Pick someone else if: your rules require a partner listed in Anthropic's directory, or you want a firm with an office in Europe. Reveneau's two offices are in San Francisco and Vancouver.

2. Fluxon

Fluxon appears in Stripe's partner directory with an address at 236 West Portal Ave in San Francisco, and Fluxon supplied that address itself [9]. Its 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" [5]. Its enterprise page tells buyers to "Connect trusted data and business logic so agents can reason accurately and act reliably." [6] The AI page also says Fluxon builds "alongside Anthropic, Google, and OpenAI"; none of the three was asked about it. No agent case for a named client was found on its pages. The AI case titles there include "Turning complex documents into usable data" and "Making AI quality measurable at every stage". Forbes's company page, with figures as of March 2026, gives 200 employees [7], and Fluxon's careers page says "300+ builders" [8]. Its own site shows Google, Zapier, Puzzle, Eppo and Braintrust as clients, and no outside source read for this guide confirms any of them. Anthropic's directory had no Fluxon entry on 21 October 2026 [1].

Fits: a company that wants agents connected to its own data and business rules, with people kept in charge of the work.

Pick someone else if: you want to read a finished agent project for a named client before you call.

3. BotsCrew

BotsCrew gives an address at 223V, B.Khmelnytskoho str, in Lviv, Ukraine, in its homepage footer, beside five US addresses, among them 548 Market St #39969 in San Francisco [10]. AIN.UA, a Ukrainian technology news site, reported on 11 February 2025 that CourtAvenue, a US company, had "acquired a majority stake" in BotsCrew, that "Financial details of the transaction were not disclosed", and that "BotsCrew currently employs 60 people" [11]. AIN.UA based its report on what BotsCrew's founders told it. BotsCrew's homepage calls it "an AI transformation company". Its case study 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. The homepage gives three results for that case: "98% faster first response time", "$140k annual cost savings" and "Up to 50% of requests fully automated". The case page lists the same figures under the heading "Projected Business Impact" [34] 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." So they are BotsCrew's own estimates, and no outside source checked them. The homepage says it picks among "Claude, GPT, Llama, or open-weight" models for each use. Open-weight models are models whose files anyone can download and run. Anthropic lists BotsCrew at the Select tier [1].

Fits: a mid-sized company that wants a support or order-status assistant connected to its own order and customer systems.

Pick someone else if: you need a large team in North America. In February 2025, AIN.UA put the headcount at 60 and reported that the team would keep working in Ukraine.

4. Galadrim

Galadrim is listed in the French company register under SIREN 817451313, the identity number France gives each company, with its registered office at 3 rue des Acacias in Paris and a creation date of 17 December 2015 [13]. The register puts its staff in the band of 50 to 99 salaried employees for 2024. Galadrim's own agents page gives a larger figure, "170+ experts" [12]. The two figures come from different sources, and the gap was not explained on the pages read. Galadrim's legal notice names offices in Paris, Lyon and Nantes. Its agents page says it "designs and develops custom AI applications and agents that automate your operations". It describes Athos, built for a client it names as Effy, as "an agent system combining OCR and LLM" that sorts files and checks them against regulations. OCR is software that reads text from scanned pages. For a client it names as Mercialys, it gives "4 operational AI agents in 6 months", its own figure. Galadrim describes its own test method: "We build a set of reference cases with your business experts and score each version against it". Anthropic lists Galadrim at the Select tier [1].

Fits: a company in France that wants agents built into its own systems and tested against cases its own staff write.

Pick someone else if: you need a team outside France. Every office Galadrim names is in Paris, Lyon or Nantes.

5. Thinkia

Thinkia gives its address as C/ Chile, 8, Oficina 208-209, in Las Rozas, Madrid, in its legal notice, which names SEARCH ENGINE BUSINESS S.L. as the owner of the site [15]. The EU's VIES system confirms that the Spanish VAT number printed in that notice, B01764703, is valid; Spain's answer includes no name or address. Thinkia's homepage says, "We put AI agents to work running your business-critical processes" [14], and also, "Thinkia is a services company that happens to have products." Its products include AI Contact Experience, a contact-centre product it describes as one that "Resolves queries 24/7 across voice and chat" and "Decides when AI should hand off to a human and who should take the call." So some of Thinkia's agent work runs on products it owns. For coding work, it sells a method it calls Enterprise AI-SDLC. Its page describes writing a project's rules and earlier decisions into files that the coding agents read at each stage, so the agents keep that context from one phase to the next. No named client case was found on the pages read. Anthropic lists Thinkia at the Select tier [1].

Fits: a company in Spain or Latin America that wants a voice and chat agent with a clear handover to staff.

Pick someone else if: you want to read named client cases. None were found on the pages read.

6. Lazer Technologies

Lazer Technologies says in its privacy policy that it is the business name of 2621956 Ontario Inc., "an Ontario corporation with its registered office in Toronto, Canada", and names a US company, Lazer Technologies US, Inc. Its about page describes "a team of 200+ senior engineers, designers, and product managers" [16]. Its AI page lists "Customer-facing agents: support triage, sales qualification, voice automation", "Internal ops: routing, approvals, compliance drafting, document review" and "Engineering workflow automation (QA, CI/CD, release gating)". Support triage means sorting incoming requests by type and urgency. QA is quality testing, CI/CD is the system that builds, tests and releases code automatically, and release gating means a check that must pass before a release. Its clients page names Shopify, Coinbase, Hinge, PayPal and Decagon, among others [17]. For Hinge it says, "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. The page does not say what Lazer built for Decagon, a company that sells support agents. Outside Anthropic's tier, each fact in this entry is Lazer's own statement. Anthropic lists Lazer Technologies at the Select tier [1].

Fits: a North American company that wants senior engineers placed inside its own team to build customer or internal agents.

Pick someone else if: you want a fixed-scope project with published results. Its agent pages list skills and no measured results.

7. Mactores

Mactores gives Bellevue, Washington, as its headquarters, 2008 as its founding year and "200+ AWS-certified engineers" on its about page [19]. AWS, or Amazon Web Services, is Amazon's cloud computing service. Mactores calls itself "an AWS Premier Tier Services Partner"; Amazon's own partner list was not opened. Its AI agents page says it "transformed Safaricom's customer-support experience into an agentic platform" and "delivered a HIPAA-compliant agentic AI platform for KlearTrust" [18]. HIPAA is the US law on keeping health records private. Mactores says its own platform, Aedeon, "builds the agent infrastructure" while its engineers "design the agent behavior". Its commitment page heading reads "Fixed date. Fixed fee.", and its homepage says, "Mactores absorbs overage on delays inside our control", which means it says it pays the extra cost of delays it causes. No amount is printed. The homepage gives "12 weeks Median time to production" as its own figure for all its work, with no method shown. Outside Anthropic's tier, each fact in this entry is Mactores' own statement. Anthropic lists it, as Mactores Cognition Inc, at the Preferred tier [1].

Fits: a company that runs on Amazon Web Services and wants a support or claims agent built to a fixed date and a fixed fee.

Pick someone else if: your systems do not run on AWS. Mactores describes itself as an AWS firm.

8. Yellow Tech

Yellow Tech is confirmed by the EU's VIES system as YELLOW TECH S.R.L., with a valid Italian VAT number and an address at Via Carlo Boncompagni 3B in Milan [21]. Its homepage says, "We train employees to use AI and build AI systems that automate business processes." Its AI Factory page defines an agent this way: "An AI agent understands what's needed, finds where the information lives and acts inside company systems." [20] It lists voice agents that "answer the phone 24/7 and in several languages", customer agents for "chat, WhatsApp, social media and email", and document agents that read delivery notes, orders and invoices, check the data "against a confidence threshold" and write it into the company's ERP, the system that runs orders, stock and accounts. "Exceptions go to human review," the page says. Its case page for Nital says, "Yellow Tech built an agent with Nital's IT team that generates the product sheets." These are Yellow Tech's own statements. Anthropic lists Yellow Tech at the Select tier [1].

Fits: a company in Italy that wants its staff trained to use AI and working agents built by the same team.

Pick someone else if: you need a team outside Italy. Its directory text describes its work in Italian organisations.

9. Linnify

Linnify labels an address at Măcinului 34 in Cluj-Napoca, Romania, "Linnify HQ" on its AI page, and also lists Cologne, Germany, and Austin, Texas [22]. Its about page says it has offices in Cluj-Napoca and Austin, gives 2016 as its founding year and counts "30+ Creators". Its AI page offers "AI agents that extract, understand, and validate data from documents such as invoices, contracts, and forms", built with its own agent framework, which it calls ARC. It also says it integrates agents "directly into your existing systems, ERP, CRM, product platforms, and internal tools". A CRM is the system that holds customer records. The page states ">95% average accuracy across agents deployed in production" and gives no definition of accuracy and no test set. Outside Anthropic's tier, each fact in this entry is Linnify's own statement. Anthropic lists Linnify at the Select tier [1].

Fits: a growing company that wants document or data agents taken from a first trial to the live system by a small team.

Pick someone else if: you need a large team. Its about page counts 30 or more staff.

10. Awiant

Awiant names offices in Stockholm, Malmö and Karlstad, all in Sweden, on its contact page, and its about page says "35 people. One way of working." It says it is "Part of the Co-native group", which Awiant's own Co-native page calls "The Home of Nordic Cloud Specialists". Its case study for Volkswagen Financial Services describes agents for one category of supplier invoices, which must be matched against purchase orders and checked against price lists [23]. The case page says, "Handling time for these invoices went from up to 20 minutes to about five minutes each, based on the customer's own estimates." That figure is the customer's own estimate as Awiant reports it. The page says "invoices are approved by a person, and only invoices meeting predefined criteria will move toward autonomous handling as the solution matures", and that the system "runs on Anthropic's Claude models on Amazon Bedrock", built end to end with Claude Code. Amazon Bedrock is Amazon's service for running AI models, and Claude Code is Anthropic's coding agent. Outside Anthropic's tier, each fact in this entry is Awiant's own statement. Anthropic lists Awiant at the Select tier [1].

Fits: a Nordic company on Amazon Web Services that wants an agent for internal work such as invoices or claims, with a person approving each decision at first.

Pick someone else if: you need customer-facing agents or a team larger than 35 people.

11. Cay Digital

Cay Digital calls itself "an AI transformation consultancy in Jupiter, Florida" and gives no street address. Its agents page says, "We build agent workflows and run them on Agent Mission Control, where your people manage every agent, what it costs and what it returns." [24] Agent Mission Control is its own product, so client agents run on a product Cay Digital owns. The same page says, "Agents ask before they guess: a person approves, with an SLA on every request." An SLA is a promised response time. Its engineering page offers to "rebuild your engineering team's delivery around coding agents". The clients it describes on its homepage are unnamed, such as "a multinational manufacturer". The Florida company register refused the request made for this guide, so its legal company was not checked. Outside Anthropic's tier, each fact in this entry is Cay Digital's own statement. Anthropic lists Cay Digital at the Select tier [1].

Fits: a company on Microsoft tools that wants finance or operations agents with a person approving each exception.

Pick someone else if: you want named clients or a published team size. Neither appears on its pages.

12. Zemoso Labs

Zemoso Labs links a map in its footer to 5151 Headquarters Dr #185 in Plano, Texas, in the Dallas area, and names Waterloo, Canada, and Hyderabad, India [25]. Its about page says "600+ people" and gives 2012 as its founding year. Its directory text says, "Zemoso builds production Claude agents and AI-native engineering teams for enterprises." Its FAWM method, short for Forward Agentic Work Mode, "reads your systems, SOPs and interview transcripts, maps the jobs your teams actually do, and scores each one before anyone builds an agent." SOPs are standard operating procedures, the written steps staff follow. For coding work, its FINE method has levels; at one of them, "Agents draft plans, code and tests. A person approves every phase." Its homepage links a VentureBeat article of 24 April 2025 about the firm. That article page carries the label "Contributor Content", and its title says it reports what Zemoso Labs' chief executive says, so this guide treats it as the firm's own words. Outside Anthropic's tier, each fact in this entry is Zemoso's own statement. Anthropic lists Zemoso Labs at the Select tier [1].

Fits: a large company that wants each job scored before any agent is built for it.

Pick someone else if: you want published agent results. None were found on the pages read.

13. AscentCore

AscentCore names offices at 205 Van Buren in Herndon, Virginia, and in Cluj-Napoca and Timișoara, Romania [26]. Its homepage says it works on two things: how a client's engineers release software, and how the client runs AI agents. It offers "Headless Claude Code agents in your CI pipelines" with "Human sign-off at PRD, implementation plan, merge and go/no-go". In plain words: coding agents that run without a screen inside the system that builds and tests code each time it changes, with a person approving the product plan, the build plan, each merge into the main code and the release decision. It also sells Insights HUB, its own platform for running agents inside the client's systems. The homepage gives "90% reduction in time to produce test suites" as its own figure, with no client or measurement shown. Outside Anthropic's tier, each fact in this entry is AscentCore's own statement. Anthropic lists AscentCore at the Select tier [1].

Fits: an engineering team that wants Claude Code run inside its own build system, with fixed points where a person signs off.

Pick someone else if: you want customer-facing agents. Its agent pages are about software work.

14. Diverger

Diverger gives its registered office at Avda Europa, 24, in Alcobendas, Madrid, in its legal notice, which names DIVERGER THINKING S.L.U. as the site owner. The EU's VIES system confirms that its Spanish VAT number is valid, and Spain's answer includes no name or address. Its homepage offers "Industrialized modernization of legacy software through autonomous agents" [27]. Legacy software means old systems that a company still runs. It says it moves systems written in COBOL, PL/SQL and VB6, three older programming languages, to modern designs "maintaining 100% functional parity", which is its own claim that the new system does everything the old one did. Its ZeroLegacy page says, "Agents execute the migration layer by layer while the Diverger team supervises, validates, and corrects." It also sets up Google's Gemini Enterprise with "agents connected to your data". Anthropic lists Diverger at the Select tier [1].

Fits: a company in Spain with old COBOL, PL/SQL or VB6 systems that it wants moved with agents doing the first version of the move and people checking it.

Pick someone else if: you want named clients or a stated team size. Neither appears on its pages.

15. HAVIT

HAVIT is listed in the Czech ARES register, the country's public register of businesses, as HAVIT, s.r.o., a Czech limited liability company, at Budějovická 778/3a in Prague 4, created on 1 October 1997 [28]. Its homepage says "30 HUMANS + N AGENTS". Its developer onboarding page offers agents that "safely read tickets, run tests, open PRs" inside a client's systems. A ticket is a written request for a piece of work, and a PR, or pull request, is a proposed code change that waits for review. It prints prices for two one-day workshops, "From €3 900 · excl. VAT" and "From €4 900 · excl. VAT". VAT is the sales tax added to most purchases in the EU. Its own blog describes work for Broker Trust, where product owners used Claude Code on the client's code and "Every change still goes through review by HAVIT's developers before it reaches production". Anthropic lists HAVIT at the Select tier [1]. The six office regions beside its directory entry are filters HAVIT chose for itself, and this guide lists only its Prague seat as an office.

Fits: a Czech or Slovak company that wants its developers set up with coding agents, with a published price for the first workshop.

Pick someone else if: you need a large delivery team. Its homepage counts 30 people.

16. Bounteous

Bounteous names offices at 6303 Cowboys Way in Frisco, Texas, and 10 High Street in Boston on its US page [30]. Its Anthropic partner page says it designs and deploys "AI agents, multi-agent systems, agentic workflows", and calls the firm "a 5,000-employee business that runs on Claude". Its directory text gives a second count, "more than 5,000 team members worldwide". Its Salesforce partner page lists Agentforce Sales, Agentforce Service and other Agentforce products, which are Salesforce's own agent tools. It also names its own Agent Management Platform "for designing, building, and operating AI agents at enterprise scale", and says, "As a Launch Preferred Partner, Bounteous hosts Claude Code Workshops around the world". No published agent result was found on the pages read. Outside Anthropic's tier, each fact in this entry is Bounteous' own statement. Anthropic lists Bounteous at the Preferred tier [1].

Fits: a large company on Salesforce that wants service or sales agents built in Agentforce, or Claude Code brought to its engineers.

Pick someone else if: you are a small company. Its pages are written for large enterprises.

17. First Line Software

First Line Software names offices in Cambridge, Massachusetts, Cirencester in the UK, Burwood in Australia, Berlin, The Hague, Bratislava, Prague, Podgorica and Belgrade on its contact page, and its about page counts "500 + Engineers across the US, EU, LATAM and Asia" [31], where LATAM means Latin America. Its directory text says it works for "enterprise clients in real estate, healthcare, and hospitality". Its product Jaime "automates customer conversations, qualifies leads, and answers support questions", so some of its agent work is a product it sets up for each client. For coding, it offers to embed with a client's engineers for 8 to 12 weeks and "install one agentic coding standard built on RACE Programming". Its Fooda case study is titled "How Fooda's Engineering Team Delivers 30% More Features Per Cycle Using Claude" [32]; the result line on that page puts a "~" sign before the 30%, and the work ran as a "6-month trial". Those figures are its own. Outside Anthropic's tier, each fact in this entry is First Line Software's own statement. Anthropic lists it at the Select tier [1].

Fits: a company in real estate, healthcare or hospitality that wants a lead and support agent, or a set program that teaches its engineers to work with coding agents.

Pick someone else if: your industry is outside the three its directory text names and you want a firm that knows yours.

Three kinds of agent

The firms above build agents for three kinds of work, and most build more than one. Customer support agents answer customers by chat, email or phone, look up orders and pass hard cases to staff. Back-office agents handle the internal work customers never see, such as invoices, claims, contracts and reports. Coding agents change code for a company's own engineers. Each kind has its own page: customer support AI agent development companies, back-office AI agent development companies and coding agent development companies. Ten of the fifteen name an office in Europe, and AI agent development companies in Europe lists them by country.

Price, tests and a checklist

HAVIT's two workshop prices are the only amounts printed on the pages read. Mactores promises a fixed date and a fixed fee without an amount, and Cay Digital prices its training per group of people without a number. What AI agent development costs sets out what each firm publishes and what to ask for in a quote.

A court case shows why testing matters even for a bot that only answers questions. In Moffatt v. Air Canada, decided on 14 February 2024, British Columbia's Civil Resolution Tribunal found that the chatbot on Air Canada's website "suggested Mr. Moffatt could apply for bereavement fares retroactively", which Air Canada did not allow [33]. A bereavement fare is a lower fare for travel after a death in the family. The tribunal ordered Air Canada to pay "a total of $812.02" and wrote, "It makes no difference whether the information comes from a static page or a chatbot." The bot in that case answered questions and took no actions. The tribunal still held Air Canada responsible for what it said. The test page explains how to check an agent before you sign, and the AI agent evals guide covers the subject in more detail.

Before you call any firm, fill in the AI agent project checklist and send the same copy to each one. To talk through an agent with the first firm on this list, contact Reveneau. For firms that build AI software of every kind, read the guides to AI development companies in the US and AI development companies in Europe.