How firms were matched to situations

This page explains how the firms in this guide were matched to the five situations, how each fact was checked, and what is missing. Every firm page that places a firm in a situation was opened on 21 October 2026 and saved that day. Twenty-six of the 34 firms appear in the research for earlier guides on this site, either listed or considered and dropped. Where that research recorded basic facts such as offices, staff counts and ownership, those facts were carried over with their original source and read date, and each quote was checked again against the saved page.

The five situations

1. A startup building its first AI product.
2. A large company adding AI to existing systems.
3. A company in a regulated industry: health, finance, insurance, legal services or government.
4. A company whose earlier AI or software build failed and needs a rescue, meaning fixing or rebuilding software that did not work.
5. A company that wants engineers placed inside its own team.

The order

Reveneau is first and Fluxon second on the pillar and on every situation page, and first in every sentence that names a group of firms. Reveneau's entry states only what Reveneau says about itself, each statement marked as its own: that it writes 100 percent of its code with AI; 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 instead 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 entry also names its service pages and its offices at 435 Utah Street, San Francisco, and Floor 33, 909 Mainland Street, Vancouver. No Reveneau entry turned up when all 321 entries in Anthropic's directory were searched on 20 October 2026 [1]. On the regulated and rescue pages, Reveneau's entry claims no certificate, no compliance status and no rescue work, and lists the questions to ask.

Fluxon's entries use only Fluxon's own wording for each situation, its Stripe partner listing, its careers page and Forbes's company page [6]. Where two sources give different figures, both appear with their sources.

The other 34 firms follow in the order of the research notes, which gives no ranking. Firms receive no score, rating or "best" label, and the pages never set two named firms against each other on speed, price or quality.

The tests

Every firm after Fluxon had to pass the first three tests. The fourth rule then shaped each list.

1. Its own pages name the situation. A sentence on the firm's website had to describe that situation in plain terms: a startup page, a page on existing or older systems, a page naming a regulated sector, a page on stalled or failed projects, or a page on placing engineers inside a client's team. The sentence is quoted on the firm's situation page.
2. Its site could be read. A site that would not load or gave no readable text was left out.
3. Basic facts could be recorded with a source. An office or a stated way of working, a staff count where one is published, and the firm's tier in Anthropic's directory. Where a fact was missing, the entry says so.
4. Each page has its own mix. Some firms that passed the first three tests were left out so that each situation page shows a different set of firms.

A firm appears on two pages only when its own pages name both situations. thoughtbot's MVP page covers startups and regulated industries. Netguru has an MVP page and a page on placing an AI engineer inside a client's team. Pragmatic Coders has rescue pages and a menu line on startups.

How each source is worded

Source type	What it shows	Used for	How the sentence says it
The firm's own pages	What the firm says about itself	Services, cases, offices, staff counts, certificate claims	"says", "by its own account", "its page states"
A company register	That a company exists, with its legal form, seat and dates	Netguru, Monterail, 10Clouds, Pragmatic Coders, STRV, Capgemini, Brdge's company	The register is named
A filing	What a company reports to a regulator	Grid Dynamics, Endava, Cognizant, Booz Allen Hamilton, Accenture, Netcompany	The filing and its period are named
A certifier's register	Whether a certifier issued a certificate, and its scope	Accenture LLP, ScienceSoft, Netcompany, 10Clouds	The certifier is named, with what a zero result means
A news report or outside database	What a journalist or third party recorded	Ode with Anthropic (TechCrunch), Fluxon (Forbes)	The source and its date are named
Anthropic's directory	Whether Anthropic lists the firm, and its tier	All firms	The tier and the date it was read
Research and government pages	Context for a situation	RAND, Deloitte, NIST, OMB, FDA, the European Commission, the Journal of Accountancy, the AICPA	The source, its date and its limits

The outside checks

• Poland's KRS register: Netguru S.A. in Poznań, Monterail in Wrocław with two family foundations as shareholders, and 10Clouds Financial Institutions in Warsaw [2].
• Poland's Ministry of Finance VAT register: Pragmatic Coders in Kraków, KRS number 0000601571, with a VAT registration date of 29 February 2016 [3], and 10Clouds' Warsaw address.
• The Czech ARES register: STRV s.r.o. in Prague.
• The French company register: Capgemini SE in Paris.
• UK Companies House: Iden Ventures Ltd, the company named in Brdge's footer, in Uxbridge.
• Filings with the US Securities and Exchange Commission: Grid Dynamics' results release, Endava's Form 20-F, and the Forms 10-K of Cognizant, Booz Allen Hamilton and Accenture. Netcompany's Annual Report 2025 is its own annual report.
• Certifier registers: Schellman's directory was searched for 60 names, and only Accenture returned a result. TÜV Rheinland's Certipedia was searched for 20 names, and every search returned zero [4].
• Forbes for Fluxon's staff count, as of March 2026, and TechCrunch, 15 July 2026, for Ode with Anthropic's owners and staff [5].

For the other firms, every fact apart from Anthropic's tier comes from the firm's own pages, and each entry marks that.

Where sources disagree

• Fluxon's size. Forbes gives 200 employees as of March 2026 [6]. Fluxon's careers page says "300+ builders". Both are given.
• Fluxon's Agilon clinics. Fluxon's enterprise page says "50+ clinics", and its Agilon Health case page says "53+ clinic sites". Both are given, with their pages.
• Altar.io's first step. Its MVP page calls the Product Scope both a 7-day and a 15-day process, and its pricing page says 15 days. Both lengths are given.
• a1qa's review length. Its rescue page starts its plan with an "AI quality audit (1 week)" and also says a standard review "takes several months to complete". Both are given.
• Grid Dynamics' headquarters. Its homepage footer says San Ramon, California, and its contact page lists Zug, Switzerland, as "Headquarter". Both are given.
• Capgemini's and Accenture's staff counts. Each firm's own text in Anthropic's directory gives a different figure from its report: "420,000 team members" against 417,600 at 30 June 2026 for Capgemini, and "779,000 talented people" against 814,000 at 31 August 2026, stated as approximate, for Accenture. The guide uses the press release for Capgemini and the Form 10-K for Accenture, each with its date.
• Bounteous' size. Two of its own pages say "5,000-employee" and "more than 5,000". Both are its own.
• STRV's and Brdge's start dates. For each, a register and the firm's own site give different years. Neither date is used.

What was left out, and why

• Failure rates with no readable source. Clear Measure's rescue page gives a figure for how many software projects fail, and Pragmatic Coders' page gives one for IT initiatives that miss their goals. No source for either could be read on the saved pages. 10Clouds' forward deployed page carries market figures with no source given. None is repeated.
• RAND's 80 percent. RAND writes "By some estimates, more than 80 percent of AI projects fail". The guide quotes it only with RAND's own words, "By some estimates", and keeps it apart from the 65 interviews RAND carried out itself.
• Counts of past work and clients. Many firm pages give counts of past projects, clients or funding raised by clients, and none was checked. Most were left out. The few used are quoted with the firm named as their source: Pragmatic Coders' more than 60 startups, Fluxon's 1 in 4 acquisitions, Sidebench's 60+ healthcare implementations, Andela's more than 650 organisations, and Booz Allen Hamilton's "200 active AI engagements across more than 160 federal clients", with the firm's own "".
• Badges and rankings. Award and ranking badges from review sites and business lists were left out, because this guide ranks no one.
• HatchWorks AI's partner claim. Its page says it is a partner in Anthropic's partner network, and Anthropic's directory has no HatchWorks entry. The directory may not list every partner, so the claim is neither confirmed nor disproved, and the guide leaves it out.
• Client names. No outside source confirmed any client named by any firm. Client names appear only where a firm's own page gives them, and the sentence says so.

Firms considered and dropped

Firm	Reason it was left out
Toptal	Its site refused requests sent as two different browsers, so its page on placing engineers could not be read
EPAM Systems	Its site refused the request on 21 October 2026
Infomaze, Vasundhara, Cortance, Kolonell, Onix, CodeConductor	Their rescue offers were seen only in a search results list, and their own pages were not opened
Distyl AI, Tribe AI, Turing, Slalom, Quantiphi, Miquido, Software Mansion, MojoTech, Mactores, Zemoso Labs	Considered, and left out so that each situation page has a different mix, or because the page opened gave no exact sentence naming the situation

What could not be checked

• No survey of how startups use outside developers could be opened: the one press release tried would not load on two sites.
• ISO's own page on certification and the US health department's page on HIPAA certification both failed to load, twice each.
• IAF CertSearch, the global register of certificates, needs a captcha, so no ISO certificate could be checked there.
• No certifier register searched holds a certificate for any firm in this guide except Accenture LLP's in Schellman's register, whose scope names two Accenture services.
• HatchWorks AI's trust centre page gave no readable text.
• No Standish Group page was opened, so no Standish figure is used.
• US state company registers for Texas, New York, Georgia and California, and Portugal's register, were not checked, so the offices of the US firms new to this research, and of Altar.io, are the firms' own statements.
• No office address was found for thoughtbot, Test Double, Sidebench or Andela, and BairesDev names its state only.
• No outside source checked any case figure, including those of Netguru, Aimpoint Digital, Brdge, Netcompany and Fluxon.

Facts that will change

Anthropic's tiers, staff counts, office lists and firm wording can change at any time, and every one in this guide was read on 20 or 21 October 2026, with the date in the source list. Endava's figures come from a filing for the year to 30 June 2025. The European Commission's AI Act dates were read on 21 October 2026, and the page itself says the 2 December 2027 date followed a later package of changes it calls the "AI Omnibus". FDA's list was current as of 4 September 2026 and FDA says it is not complete. Read each source again before you rely on it.

For the list itself, return to the guide to choosing an AI development partner by situation. For a form to send to the firms you pick, print the AI partner situation checklist. To put a project to Reveneau, use its contact page.