40,000 Firms Applied. Most of Them Learned Claude Last Quarter.
When Anthropic launched the Claude Partner Network in March 2026 with a $100 million investment, more than 40,000 firms applied. That number tells you two things: the demand for Claude expertise is enormous, and the supply is wildly uneven.
Over 10,000 consultants have now earned Claude certifications. But a certification page and a partner badge do not tell you whether a firm can actually ship. The gap between a practitioner who builds with Claude every day and a reseller who added "Claude" to their services page in April is the difference between a working system and a very expensive slide deck.
Here are the seven criteria we would use to evaluate any Claude partner - including us. Each one comes with a question to ask and a red flag to watch for.
1. Certified Practitioners, Not Just a Partner Badge
Anthropic's first professional certification, the Claude Certified Architect - Foundations (CCA-F), launched in 2026. It is a 120-minute, 60-question exam that costs $99, and it tests systems design - agentic architecture, Claude Code configuration, deployment patterns - not just prompting. A firm-level badge means the firm met a threshold. Individual certifications mean the people on your project passed the bar.
The distinction matters because badges are earned once, at the firm level. The consultants actually assigned to your account may have joined last month.
- What to ask: "How many of the people who will work on our account hold the CCA-F or an equivalent Anthropic certification?"
- Red flag: The firm cites its partner status but cannot name a single certified individual on your delivery team.
2. They Build With Claude Daily
The fastest way to separate practitioners from resellers: ask to see the tools they built for themselves. A firm that genuinely works with Claude every day has internal agents, MCP servers, automated workflows, and Claude Code running in its own operations. A firm that only advises does not.
If a consultancy cannot demo something Claude-powered from inside its own company, it is selling you a map of a country it has never visited.
- What to ask: "Show me two internal tools your own team uses that are built on Claude. When did you last update them?"
- Red flag: Every example they show is a client mockup, a demo environment, or a screenshot from Anthropic's documentation.
3. Deployment Depth: Claude Code, MCP, and Managed Agents
Rolling out Claude chat licenses is procurement, not consulting. Real deployment depth means the firm works across the full Claude stack: Claude Code for engineering workflows, Model Context Protocol (MCP) servers that connect Claude to your CRM, database, and internal systems, and managed agents that run scheduled work without a human in the loop.
The Partner Network's own tiering reflects this. The Services Track, announced June 2026, requires production deployments - not seat counts - at every level: Select partners need 2 production customer deployments, Preferred needs 15 deployed customers, and Global Premier needs 100 customers across 3 or more regions.
- What to ask: "Walk me through an MCP server you built for a client. What system did it connect to, and what broke in production?"
- Red flag: Their "deployment" case studies are all chat rollouts and prompt libraries - no agents, no integrations, nothing running unattended.
4. They Train Your Team AND Ship Working Systems
Training without deployment produces enthusiasm that fades in three weeks. Deployment without training produces a system nobody inside your company can operate, extend, or trust - which means you are renting your own capability back from the consultant forever.
The best engagements do both in the same motion: your team learns on the exact system being built for them, and by the end they own it. This is capability transfer. The alternative is dependency, and dependency is a business model - just not one that serves you.
- What to ask: "At the end of the engagement, who maintains what you built - your team or ours?"
- Red flag: The proposal includes an open-ended "ongoing management" line item with no plan to hand over the keys.
5. Right-Sized for Your Company
The Claude Partner Network spans everything from the Big Four to boutique specialists. Accenture has trained 30,000 professionals on Claude. Deloitte gave Claude access to its 470,000-person workforce. KPMG, PwC, Cognizant, BCG, and Slalom are all members, alongside specialists like Tribe AI and Caylent.
Those numbers are genuinely impressive - and genuinely irrelevant if you are a 50-person company. A firm built to run 10,000-seat transformations carries the overhead, timelines, and minimum engagement sizes of that world. A boutique that ships in weeks is a mismatch for a Fortune 500 global rollout. Neither is better; one of them is wrong for you.
- What to ask: "What is the smallest and largest engagement you ran last year, and which one did we most resemble?"
- Red flag: The firm cannot describe a past client that looks like your company in size, budget, or timeline.
6. Public Proof: Named Case Studies and Open-Source Footprint
Anthropic bakes public proof into its own partner tiers: Select requires 1 public customer story, Preferred requires 3, and Global Premier requires 15. There is a reason. Anonymous case studies ("a leading logistics company") are unfalsifiable. Named clients with real metrics can be checked.
An open-source footprint is the other honest signal. Published skills, MCP servers, agent frameworks, and repos with real usage show a firm's actual working style before you sign anything - and they cannot be faked retroactively.
- What to ask: "Give me two named clients I can call, and a link to something you have published that I can read tonight."
- Red flag: Every reference is anonymized, every metric is unverifiable, and the firm has published nothing.
7. Honest Scoping: They Tell You When Claude Is Not the Answer
A partner whose answer to every problem is a bigger Claude engagement is a vendor, not an advisor. Some problems are data-quality problems. Some are process problems. Some are better solved with a $30 off-the-shelf tool than a custom agent. A good partner says so - early, before you have spent anything.
The most valuable sentence a consultant can say is "you don't need us for that." Firms that never say it are optimizing for their pipeline, not your outcome.
- What to ask: "Describe a prospect you turned down or descoped because Claude was not the right fit. What did you recommend instead?"
- Red flag: They have never met a problem that was not an AI problem.
How OneWave Fits
We hold ourselves to this same list, so here is the honest version. Every member of our team is Claude-certified - we are certified Claude experts, and criterion one is table stakes for us, not a badge on a website. We build with Claude daily: our own CRM tooling, internal agents, and client-facing systems all run on it, and we will demo them in the first call.
We have shipped 50+ deployments - Claude Code rollouts, MCP servers, managed agents, and team training programs that end with a working system, not a certificate. We are a boutique by design: built for SMB to mid-market teams that want results in weeks, not a 470,000-seat transformation. And we descope when Claude is not the fit, because a client who trusts our "no" comes back for the next "yes." Our success stories are public and named.
The Bottom Line
Run every candidate through all seven criteria and the field narrows fast. Certified individuals on your account, internal tools they built for themselves, deployment depth beyond chat, capability transfer instead of dependency, right-sized economics, public proof, and the willingness to say no.
If you are evaluating partners now, our Claude consulting page covers exactly how we run engagements - or talk to us and put us through the seven questions above. We wrote them expecting to be asked.