Top AI agent development companies / Cost and method

Cost and method

How these AI agent development companies were chosen

Each of the fifteen firms after Reveneau and Fluxon passed the same tests: its own pages describe agents it builds for other companies, a city and country could be recorded for it, its site could be read on 21 October 2026, and it appears on no earlier Reveneau list. All fifteen are listed in Anthropic's partner directory. This page lists every outside source used, the facts that rest on a firm's own pages alone, the conflicts between sources and how each is shown, the claims left out, and the firms considered and dropped.

Published October 21, 2026. Editorial.

Key takeaways

Six of the fifteen firms have at least one fact from outside their own pages: a company register, the EU's VIES tax system or a news report. For the other nine, every fact apart from Anthropic's tier is the firm's own.
Simform, Roboyo AI, Atrium, Baresquare, TRY and five firms whose sites gave no readable text were considered and dropped, each for a stated reason.
Ranking badges, ratings, client counts and project counts on firm pages were left out of the entries.
No outside source confirmed any client a firm names, and no outside source checked any result a firm publishes.

This page explains how the seventeen firms in this guide were chosen, how each fact was checked, and what is missing. Every page used was read on 21 October 2026 and saved that day.

What counts as an agent here

An AI agent, in this guide, is software in which an AI language model takes steps by itself toward a goal. It looks things up, uses other systems, fills in forms, writes code and passes work to a person. A firm counts when a business can hire it to build agents of this kind. Anthropic, in its own guide "Building effective agents", calls a workflow a system that follows "predefined code paths" and an agent one in which models "dynamically direct their own processes and tool usage". Many firms build both, and the guide includes a firm when its pages describe either type of work done for clients.

The order

The order is fixed. Reveneau is first and Fluxon is second. Reveneau's entry states only what Reveneau says about itself, each statement marked as its own: that AI writes 100 percent of its code; that every change must pass a large eval suite, written from the specification, before release; that Jev grades the suite and makes it run ten times faster than its previous grader; that it uses AI in place of hiring more engineers and puts that saving into clients' prices; and that it takes responsibility for the whole project through production and after release. Its offices are at 435 Utah Street, San Francisco, and Floor 33, 909 Mainland Street, Vancouver. Anthropic's directory did not list it when its 321 entries were searched on 20 October 2026 [1].

Fluxon's entry uses its own AI development and enterprise pages, its careers page, its Stripe partner listing and Forbes's company page for it [8]. Its two staff figures, 200 from Forbes as of March 2026 and "300+ builders" from its own careers page, are given together. Its pages showed no agent case for a named client, and the entry says so.

The other fifteen follow in the order of the research notes, which gives no ranking. No firm gets a score, a rating or a "best" label, and no page compares two named firms on speed, price or quality.

The tests

Each of the fifteen had to pass every test.

It builds agents for other companies. Its own pages had to describe agent work it does for clients, through a service page, a method page or a case study. A firm whose pages centre on its own software product, with build work for hire unclear, was left out.
A city and country could be recorded. The office could come from the firm's pages, a register or the VIES system.
Its site could be read. A site that refused the request, was offline or gave no readable text was left out.
It is new to Reveneau's lists. No firm here appears in an earlier Reveneau guide to AI development or forward deployed engineering firms.

All fifteen also hold a tier in Anthropic's Claude Partner Network directory, which held 321 entries on 21 October 2026: 256 at Select, 40 at Preferred, 16 at Global Premier and 9 with no tier [1]. Anthropic sets the tier. The firm writes its own description and picks its own office regions, so those are treated as the firm's own words.

How each source is worded

Source type	What it shows	Used in this guide for	How the sentence says it
The firm's own pages	What the firm says about itself	Services, cases, results, offices, staff counts	"says", "by its own account", "its page lists"
A company register	That a company exists, with its legal form, seat and dates	Galadrim (France), HAVIT (Czech Republic)	The register is named
The EU's VIES system	Whether a VAT number is valid, and sometimes the name and address	Yellow Tech, Thinkia, Diverger	VIES is named
A news report	What a journalist reported, often from the company	BotsCrew (AIN.UA)	The site and date are named, with whom it relied on
An outside database	Facts a third party records	Forbes's page for Fluxon	Forbes is named, with its date
Anthropic's directory	Whether Anthropic lists a firm, and at which tier	Every firm	The tier, with the date read
A survey, study or ruling	Context about agents	KPMG, tau-bench, NIST, Moffatt v. Air Canada	The source, its date and its conditions

The outside checks

French company register: Galadrim, SIREN 817451313, created 17 December 2015, registered office 3 rue des Acacias in Paris, staff band of 50 to 99 salaried employees for 2024 [2].
Czech ARES register: HAVIT, s.r.o., company number 25612697, created 1 October 1997, at Budějovická 778/3a in Prague 4 [3].
EU VIES system: Yellow Tech's Italian VAT number returned the name YELLOW TECH S.R.L. and a Milan address [4]. Thinkia's and Diverger's Spanish VAT numbers are valid, and Spain returns no name or address [5] [6].
AIN.UA, a Ukrainian technology news site, on 11 February 2025: CourtAvenue's majority stake in BotsCrew, terms not disclosed, and 60 staff. AIN.UA says the founders announced the deal to it [7].
Forbes company page for Fluxon, figures as of March 2026 [8].

For the other nine firms, Lazer Technologies, Mactores, Linnify, Awiant, Cay Digital, Zemoso Labs, AscentCore, Bounteous and First Line Software, no outside source was found for any fact apart from Anthropic's tier. Each of their entries says so once.

Where sources disagree

Galadrim's size. The French register gives 50 to 99 salaried staff for 2024. Galadrim's agents page says "170+ experts" [9]. Both are given, each with its source. The register counts salaried staff in one year, and the reason for the gap was not checked.
Bounteous' size. Its Anthropic partner page says "a 5,000-employee business", and its directory text says "more than 5,000 team members worldwide" [11]. Both are its own, and both are given.
Linnify's main office. Its AI page labels the Cluj-Napoca address "Linnify HQ", and its about page says it has offices in Cluj-Napoca and Austin, Texas [12]. The guide lists Cluj-Napoca first and names both.
Mactores' company name. Its site footer reads "Mactores Inc.", and Anthropic's directory lists "Mactores Cognition Inc". Washington's company register was not checked.
First Line Software's countries. Its contact page gives an office in Belgrade, and its about page lists the countries it delivers from without Serbia [13]. Both are stated.
Thinkia's trading name. Thinkia's legal notice names SEARCH ENGINE BUSINESS S.L. as the owner of its site, and VIES confirms only that the tax number is valid.

Sources that need a label

Zemoso Labs' homepage links a VentureBeat article dated 24 April 2025 [10]. 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 and takes no facts from it. HAVIT, a firm of 30 people by its own count, picks six office regions in Anthropic's directory; those regions are its own choice, and the guide lists only its Prague seat as an office. Three firms run client agents on a product they own: Cay Digital on Agent Mission Control, First Line Software with Jaime, and Thinkia with AI Contact Experience. Mactores says its own platform, Aedeon, "builds the agent infrastructure". Their entries say so, because a buyer may then pay for a product as well as for the build. BotsCrew's S&B Filters case page lists its figures under "Projected Business Impact" and calls them estimates, and every page that uses those figures says so.

What could not be checked

The Ontario, Washington, Texas, Virginia, Romanian and Swedish company registers were not checked, and Florida's refused the request. So the legal companies of Lazer Technologies, Mactores, Zemoso Labs, AscentCore, Linnify, Awiant and Cay Digital are their own statements.
Linnify prints no registration number on the pages read.
The partner lists of agent platforms were tried. LangChain's, CrewAI's and Decagon's partner pages answered "page not found", Sierra's partner page lists no firms, and Relevance AI's directory needs a script to load. So no agent platform's list confirms any firm here.
The Salesforce and AWS partner listings for Bounteous, Mactores and Awiant were not opened.
McKinsey's State of AI page did not answer within the time allowed, so no McKinsey figure is used.
A Gartner press release of 25 June 2025 on agentic AI projects was read for the research, but its page refuses automated requests, so the link checker used for this guide cannot confirm it, and it is not cited.
No outside source confirmed any client named by any firm, and none checked any published result.

What was left out

Ranking badges, such as a "#1" badge from a review site on BotsCrew's homepage, were left out, because this guide ranks no one. Satisfaction ratings out of five were left out for the same reason. Counts of clients, projects and agents in production on firm pages were left out of the entries too, because a count of past work is hard to check, and none was checked. Speed and savings figures are given only where an entry needs them, always quoted and marked as the firm's own, with every "up to" and "about" kept.

Firms considered and dropped

Firm	Why it was dropped
Simform	No office city or address on its about, contact or privacy pages
Roboyo AI	Its sites answered "temporarily offline" on two tries
Atrium	Its site refused two different requests
Kiso Technology, Casper Studios, Vedara, altermAInd, Chaos Gears	Their sites gave no readable text
Baresquare	Its pages read as a product company, which this guide leaves out
TRY (Oslo)	A marketing agency whose AI pages show no client agent work
Proxet, Devsloop, Perficient, Genzeon, Neurealm, TecAce, Workiflow, Empat, SPD Technology, InfoBeans	Listed in Anthropic's directory with agent wording, and not researched in the time available
Caylent, Provectus, Clearscale, Quantiphi, SFEIR, adesso, Forgd, Godel, Rosetree, Theodo	Already in earlier Reveneau research, so not repeated

Facts that will change

Directory tiers, staff counts, office lists, prices and model names can change at any time. HAVIT's workshop prices, Mactores' fixed-fee wording and every tier in this guide were read on 21 October 2026, and each is dated in the sentence that uses it. The tau-bench figures are for the agents and models tested in mid-2024, and NIST's for the upgraded Claude 3.5 Sonnet in the AgentDojo tests published in January 2025. KPMG's survey is for the third quarter of 2026. For the list itself, return to AI agent development companies. For a form to send to the firms you choose, print the AI agent project checklist. To ask Reveneau about a project, contact Reveneau.

Common questions

Why does every firm in this guide need a recorded city and a readable website?

A buyer needs to know where a firm works, and a fact has to be read before it can be stated. So a firm was left out when no office city could be recorded, as with Simform, or when its site refused the request, was offline or gave no readable text. Each firm also had to describe agent work on its own pages.

Why do all fifteen firms after Fluxon hold a tier in Anthropic's directory?

The candidates came from Anthropic's partner directory: the research opened the entries whose text describes agent work, then checked each firm against the guide's tests. Anthropic sets each tier, and the firm writes its own description and office regions. Thirteen hold the Select tier, and Mactores and Bounteous hold the Preferred tier. Reveneau and Fluxon are not listed.

Which firms were dropped because their websites could not be read?

Roboyo AI's sites answered that they were temporarily offline on two tries. Atrium's site refused two different requests. Kiso Technology, Casper Studios, Vedara, altermAInd and Chaos Gears gave no readable text. None of them was assessed, and each could be added if its site can be read on a later check.

Why was Simform left out of this guide?

Simform's pages describe agent work, including agentic AI and agentic software delivery. But no office city or address appeared on its about, contact or privacy pages when they were read on 21 October 2026, and this guide needs a city and country for every firm it lists. So Simform was dropped for that reason alone, and the guide makes no judgement about its work.

Why does no firm from Reveneau's earlier research appear in this guide?

The guide set out to show a new set of firms. Caylent, Provectus, Clearscale, Quantiphi, SFEIR, adesso, Forgd, Godel, Rosetree and Theodo appeared in earlier Reveneau research, listed or dropped, so they were not repeated, and every firm after Fluxon here is new. The guides to AI development companies in the US and in Europe cover a different set of firms.

Did any agent platform's partner list confirm a firm in this guide?

None. LangChain's partners page returned an error, as did CrewAI's and Decagon's partner pages. Sierra's partner page lists no firms, and Relevance AI's directory needs a script to load. So the one partner list used is Anthropic's directory. Each firm's work with other agent platforms is described from the firm's own pages.

How far apart are Galadrim's own staff count and the French register's band?

The French company register puts Galadrim in the band of 50 to 99 salaried employees for 2024. Galadrim's own agents page says 170+ experts, which is above the top of that band. The register counts salaried staff in one year, and the firm's figure has no date. The reason for the gap was not checked, so both figures are given.

Why is the VentureBeat article about Zemoso Labs not used as press coverage?

Zemoso's homepage links a VentureBeat article dated 24 April 2025 about the firm. The article page carries the label Contributor Content, and its title says it reports what Zemoso Labs' chief executive says. So the article is treated as the firm's own words, and this guide uses no facts from it. The facts in Zemoso's entry come from Zemoso's own pages and Anthropic's tier.

Why are ranking badges and client counts missing from the firm entries?

This guide ranks no firm, so ranking badges from review sites and satisfaction ratings out of five were left out. Counts of clients, past projects and agents in production were left out too, because a count of past work is hard to check, and none was checked. Speed and savings figures appear only where needed, quoted and marked as the firm's own.

For which firms is the legal company name the firm's own statement?

Lazer Technologies, Mactores, Zemoso Labs, AscentCore, Linnify, Awiant and Cay Digital. Their company registers were not checked, and Florida's register refused the request for Cay Digital. Galadrim and HAVIT were matched in government registers, Yellow Tech through the VIES tax system, and Thinkia and Diverger through VIES for their tax numbers only. Ask every firm which company will sign.

Why is there no Gartner figure in this guide?

A Gartner press release of 25 June 2025 on agentic AI projects was read during the research. Its page refuses automated requests, so the link checker used for every source in this guide could not confirm the address, and the guide does not cite it. A Gartner figure on that page is also a forecast, so it would be marked as one if used.

References

More in Cost and method

What AI agent development costs

Few of the firms in this guide print a price. Of the seventeen in this guide, HAVIT is the one that prints amounts on the pages read: "From €3 900" and "From €4 900", both "excl. VAT", for two one-day workshops. Mactores promises "Fixed date. Fixed fee." and prints no amount. Cay Digital prices its training per group of people and prints no number. This page quotes each one exactly, explains what changes the cost of building and running an agent, and lists what to ask for in a written quote. Every page was read on 21 October 2026.

How to test an AI agent before you buy

A demo shows an agent working once. Before you sign with a firm that builds AI agents, ask to see the agent tested on your own cases, many times, against hidden instructions, and on the actions that must go to a person. This page explains five tests and where each idea comes from: the tau-bench study of agents in 2024, NIST's tests of hidden instructions in 2025, the Moffatt v. Air Canada decision about a chatbot's wrong answer, and the guidance Anthropic and OpenAI publish about their own models. Every source was read on 21 October 2026.

AI agent project checklist

This checklist is a form to fill in before you contact any firm that builds AI agents, and to send to each firm on your shortlist. It asks you to describe the task, the systems the agent will connect to, the actions that must wait for a person, your own test cases, your data, and your budget. It then lists the answers to ask each firm for, so you can compare them side by side. Each question links to a fact in this guide, read on 21 October 2026, that shows why it matters.

Previous: How to test an AI agent before you buy Next: AI agent project checklist